Home

Thesis - Repository TU Delft

1. gendidesign appspot com project 80 c B ce y He 0 Feedback CELEUS or bd Contact support 2 Integrate with code version control PLANNED I suggest you Ea Add support for Python studies Postidea Your feedback is really important and much appreciated Part Ill Code report Appendix A Appendix B Bootstrap package contents The boostrap package zipfile contains the following project folder tree structure and files bin An empty directory documentation gendl dedign assets assets for the off line design viewing application data classes json The class summaries consistency json The in consistency data diagrams a ba SVG The diagrams as svg design xml Project design XML index html Viewer for the class and consistency JSON data Instructions txt Instructions about what documentation belongs in this folder input An empty directory lib An empty directory output An empty directory source file ordering isc Order for compiling files The existing files are added to this already together with some comments on how to use the file package lisp A file that defines a gendl package parameters lisp A file for putting project wide parameters both project independent like default project directory paths and project specific project directory paths are there by default util lisp A file where small generally used utility functions can be put most
2. Oo0000 Note that some meta data might not be mentioned by some authors because they are considered trivial such as the name of the package Because of this and because of the limited number of sources the quantification of how often a meta data type is mentioned should be regarded as indicative 6 1 4 Existing approaches similar studies In Shull et al 2004 knowledge sharing issues in Experimental Software Engineering are addressed In Experimental Software Engineering research there is a large need to reproduce experiments to validate and gain confidence in earlier findings The study addresses the need for an effective collaboration structure between the original and replicating researchers to convey both the explicit lab package and the tacit knowledge This is very similar to the situation of Engineering Automation developers The following aspects of lab packages were highlighted o It is difficult to assemble a complete and consistent package o Configuration management version control is needed o Doing a trial first is useful to learn how to use the package This also reveals knowledge missing in the package o Feedback in the form of discovered issues must be handled properly The concept of sharing executable knowledge among non technical end user developers has been explored by Leshed et al 2008 The developed system CoScripter was intended to share procedural best practise how to knowledge related t
3. html JsonRequestHandler properly outputs JSON data and is used for json routes 10 There are more http verbs but these are the ones used by GenDL Designer Part IIl Code report 46 Access control Access control on most web servers is done with sessions A session is a combination of session data stored on the server e g the username and access rights and a session ID the user sends along with each request these is done with so called cookies With the session ID the server knows which session data belongs to the user who made the request To separate concerns session management was implemented in SessionRequestHandler a RequestHandler class which other classes can subclass perhaps along with other RequestHandler subclasses Subclassing SessionRequestHandler adds the attribute session a dictionary in which session data can be stored Two special request handlers that subclass SessionRequestHandler ate LoginHandler and LogoutHandler They add remove the entry user in the session attribute In practice session is not used by actual request handlers because several convenience decorators have been created for access control check_access_to_project and check_admin do what they say they do The snippet below shows how to use them class ProjectHandler WebRequestHandler SessionRequestHandler check_access_to_project def get self project_id project get_project project_id Before the get me
4. mxGraph is a commercial software package It is however possible to obtain a free trial license 7 2 3 jQuery jQuery is a javascript library that makes many common operations in Javascript easier Currently it is the most popular of its kind It is used extensively in GenDL Designer for DOM manipulation event handling and communication with the web server Compared to other frameworks jQuery is rather small it doesn t provide a graphical toolkit and relies on plugins for less commonly needed functionality 7 2 4 JQuery Ul jQuery UI is a Javascript library for creating the user interface of web applications and is built on top of jQuery GenDL Designer uses the jQuery capability to make any HTML element draggable and sortable 7 2 5 Dojo Dojo is a general javascript library It is an alternative to jQuery but much more extensive It includes dijit a widget library to built web applications The learning curve is harder though as it often builds an unfamiliar layer around the browser concepts you are already familiar with The documentation is in my experience not as good as jQuery s especially not for modules outside the core of dojo Dojo is used in conjunction with jQuery for two reasons The first is a historical one The GenDL Designer user interface was originally developed as a showcase for what an exiting system the IDEA editor could evolve to IDEA used dojo at the time The elements that were added to the user in
5. o You can install this Python version next to an already installed version o Current link http python org ftp python 2 7 5 python 2 7 5 msi Part IIl Code report 16 e Google App Engine SDK for Python o Any recent version should work Up till now GenDL Design was developed against version 1 8 3 o Current link https developers google com appengine downloads Google_App_Engine_SDK_for_Python The provided links might become outdated In that case the bold part of the url should lead you to a web page with links to the downloadable files 3 2 Getting the application code The first step is to obtain the source code The repository is a mercurial Hg repository and can be found here https bitbucket ore pidewitte gend desiener Alternatively you can obtain the source code as files bundled in a zip file by contacting the author Currently the size of the repository is less than 20 MB Most of the code is Python HTML CSS or Javascript which does not need to be compiled There are two exceptions though the codeclient executable code mirror tool and the lisp parser The code client must be compiled into a single executable The parser tables must be generated once before the parser can be used If you do not need to change the code client or lisp parser you can download the codeclient_generation and plygendl_compiled zip files from the download section of the Bitbucket repository The zip files contain instr
6. 2 2 Deployment diagram On the next page the deployment diagram of the system is shown This diagram shows the main components of each subsystem They are explained further in the corresponding subsystem chapter 2 3 Data flows The three big arrows in the deployment diagram represent the main data flows between the subsystems Design modifications and code modifications are pushed to the server by the web client and the code client respectively The web client also retrieves the inconsistencies the server discovered Three small arrows indicate internal data flows One arrow indicates that the consistency area of the web client can trigger design modifications Two arrows indicate the bi directional data flow between the server app and 1 the underlying server data store and 2 the memcache a fast temporary data storage service An arrow with modifications to the code client is absent the web client does not trigger code modifications It was decided that modifying files on the user his system is the exclusive right of the user himself GenDL Designer simply does not mess with the user his code This is clear to the user and makes GenDL Designer easier to implement as well Part Ill Code report 23 a Aeysfhs a qeynoexe wyuny UOYAG uawuojaua ADUa S SU0D m s qam dde savues a10 5 BEG 3V9 Ay yuawuoaua u isaq ampnyseyu BY PIS JAAS 24 Part Ill Code report 2 4 Subsystem main responsibilities
7. Agile software development methods fit well with scientific software development as they share an emphasis on responsiveness to change and collaboration and on an incremental nature Sletholt et al 2012 Segal 2008 An example is its application in the KBE domain van den Berg et al 2011 Agile practises are adopted selectively however Sletholt et al 2012 communication and flexibility are embraced but in the areas of requirements and testing agile practises tend not to be adopted In fact Segal 2008 notes that agile methods are sometimes seen by scientists as a confirmation of their from a Software Engineering point of view flawed development approach A journal paper database Scopus was queried to identify literature that explicitly relates the Software Process Improvement field to Engineering Automation Knowledge Based Engineering Design Automation Scientific Software or Research Software No such literature was found This does not indicate that process improvement has not been applied in these areas rather it indicates that the attention of the Software Process Improvement community has not yet focussed on these areas Adoption sharing collaboration and reuse in Software Development 2 2 4 Segal 2007 and Howison amp Herbsleb 2011 report a limited level of sharing and collaboration among professional end user developers and in the scientific software community Part Il Initial study 13 respectively This
8. Methodologies for Knowledge Based Engineering are the most extensively developed methodologies found within Engineering Automation The two most widely recognized methodologies are CommonKADS for knowledge management in general and the more recent MOKA van der Velden et al 2012 The MOKA methodology is discussed in the next section 2 3 5 The MOKA methodology The MOKA methodology seems not directly relevant for Engineering Automation software development it is a methodology for managing engineering knowledge rather than a software development methodology It is relevant nevertheless because engineering knowledge plays an essential role in Engineering Automation MOKA connects the Knowledge Management background to the Software Engineering background of Engineering Automation MOKA stands for Methodology and software tools Oriented to Knowledge based engineering Applications and is introduced in Stokes 2001 The main contributions of MOKA are the application lifecycle a representation for engineering design knowledge and software support for MOKA Application lifecycle The first contribution the application lifecycle consists of six steps 1 Identify step in which objectives are set 2 Justify step in which a project plan and a business case are developed 3 Capture step in which raw knowledge is captured and structured into an Informal Model 4 Formalize step in which the Informal Model is translated into a Formal Model 5 Pack
9. Part IIl Code report 56 7 3 Components 7 3 1 Introduction The web client is build from modular blocks according to the AMD standard The modules depend on each other and use each other Each of the modules has a particular responsibility Some modules are responsible for a visible part of the user interface the project tree the diagrams the consistency tab the settings tab the class dialogs and the code dialogs Other modules provide a simple interface for specific tasks the entity manager the tab manager the diagram manager the code generator and consistency resolution module Finally some modules simply provide a handy service such as the generic_tree_display and the auto_update_tab modules Modules interface in two different ways One possibility is that a module requires another module to be loaded before it loads itself You can see this as a direct dependency Another possibility is that a module can create objects but only if another object from a different module is provided For example the diagram component can create and show a diagram but the diagram constructor must be given an entity manager to save diagram modifications The entity manager will have been created with the entity_manager module This could be considered an indirect dependency The diagram component is aware of an entity manager but is not concerned with the module where it came from It only matters which API is exposed by the entity ma
10. codeclient MonitorApp pause monitor errorstate_monitor folder_path string file_mask string interval int ip has_single_acton alive_interval int _exi H has_submenu_items last_report_alive int H get_submenu_items triqger_menu_item report_alive set_rightclick_menu check_stop_requested show_balloon filestore re interface FileStorelnterface methods l I I I l I l l I l I I l I I hat am I has_report_alive_interval_passed et_leftclick_menu l l I I I I I l l l l I I I I I 8 3 2 Integrating components codeclient top level MonitorApp integrates the worker the file monitor and the GUI the system tray icon This way both can be developed and tested individually The functionality they expose is easily integrated in the top level MonitorApp This would not be possible if the GUI wrapped the worker which would have been the alternative design decision The GUI uses several icons The file location of these icons varies depending on whether the code is run as regular Python code or as PyInstaller executable The assets module abstracts the difference away with a single function get_asset_path Part IIl Code report 65 8 3 3 Monitoring files codeclient monitor The monitor module provides the FileSystemMonitor It starts a second thread which checks for changes in the monitored folder The changes are r
11. disappear It is strictly intended as development system 3 4 Deployment on the Google App Engine Since the application is already running on GAE http gendldesign appspot com it should not be necessary to create a new installation but for completeness sake e Ensure that your local installation works e Sign up for Google App Engine on https appengine google com e Create an application you will have to choose an application name not yet in use e Open app yaml in the root folder of the source code and adjust the application line to the name you ve just chosen e Click the Deploy button in the Google App Engine Launcher and provide your Google Account credentials e Wait for the deploy process to finish You will see some output that indicates the progress and the success of the deployment e You can now visit http your_name appspot com 4 Administration of an existing installation 4 1 Admin password You will have to set an admin password and a system salt before you can administrate users and projects e Choose an admin password and system salt which is a word of 5 10 random characters e Run python exe Under Windows 7 simply type in the start menu search box e Execute the following lines to hash your new admin password import hashlib hashlib sha256 your password system salt hexdigest e In server src server_app settings py adjust the ADMIN_PASSWORD_HASHED and SYSTEM_
12. 2 1 3 Knowledge Management approachesnnsunnasnner nra ARER E SENEE E N ES 2 2 SOFTWARE ENGINEERING irrena AAE A AAA REE ETA E ER E TE R RAE s 2 2A Inroduc on A S 2 2 2 Software requirements design and testing 2 2 3 Software development methodologies and the software process 2 2 4 Sharing collaboration and reuse in software development 2 3 ENGINEERING AUTOMATION 2 3 1 Introduction wees ae ae Sed es site nee ts eis 0 3 2 End User Developmentr e a a a ee a ET AA Nan nia bene Bee Ban nh ne Rnd DAA ann 2 3 3 Professional End User Development 2 3 4 Design Automation and Knowledge Based Engineering ses ae ue ee hie 2 3 5 The MOKA method Glo gy sii e 22 cessscs trated seats cesasaeagaruescedecaadedassaatedoasesbededuccsndsessradastactdessoatssaussnnderusteececssdducaaats 2 3 6 Adoption of Software Engineering Practices drii E n EEE a E 2 4 DISCUSSION AND CONCLUSIONS sscssssssssssessssesscsceseesescesessessescesesssescsesseesseeesessssseseesessesseseesessessssessesseseseessesasesesesaeesseresessseaserees 3 INTERVIEWS iicsccssescieceosedstecondsessesseleSeesssdedeesetdedennnsdessenssesseesecessesccecceeseseceessesedcuasssedenscceedsaseseecstesesedusesses 17 DEN TRODUGTION EAAS AE S A A AE AAS REE edad coe edad AEE ASNN SANA H NENA AVII a ea DI AE E E E AET E NANA arian peseetce need 3 3 ANDI PAKA E E E E A E E E 3 3 1 How Engineering Automation software is currently developed 3 3 2 Why Engineering Automation software is developed the
13. Child object gt Child type A Class mixin Ce Collection of popular elements Part IIl Code report 7 There are also marker icons Element is not on any diagram i e recently imported Goto documentation Except for the packages all elements can be dragged into the diagram to display that element in that particular diagram Even diagrams can be dragged into a diagram 2 3 Multiple diagrams Putting the design of an entire system of realistic size in one diagram would result in a huge hard to work with diagram GenDL Designer supports creating multiple inter linked diagrams Main diagram Detail of lifting surface Consistency Design Code Main diagram eS ee i Detail of lifting surface a Q Q fa gt E lifting surface v E wing A lifting surface v flap my ifti m gt lifting surface cia ssn saa fas ank volume nil v fuel tank smem A E Multiple diagrams are created with the button next to the save button or through dropping the blue connection arrow of a class into the void and choosing link to diagram The diagram is opened in a new tab Existing diagrams can be opened by double clicking the diagram in the project tree Multiple diagrams are used to keep diagrams simple If the diagram becomes too complicated a diagram reference can be connected to a class This class can then be detailed in the new diagram This is especially interesting for classes that are a mix
14. and operations Lower is better smaller components functions classes etc are easier to understand Misra 2011 A complexity measure proposed by Wang amp Shao 2003 is based on the inputs and outputs of functions and the amount and type of the basic control structures BCS in the function complexity es w ie function N ae function Wacs N acs function BCS Here N stands for the amount of input parameters output parameters or basic control structures W stands for weight The weight of each basic control structure is different e Sequence 1 e Branching 2 3 for case statements e Iteration 3 e Function call 2 3 for recursion Part Il Initial study 50 e Concurrency 4 An alternative complexity measure the Unified Complexity Measure Misra 2011 works more generically on lines complexity For perons line t N perina line i Wacs Iine In any case the complexity measures provide only an indication and should not be trusted blindly 6 1 3 Engineering app store aspects discussed in literature In literature several reports were found which describe the introduction of a repository which has some similarity with the proposed concept of an engineering app store These reports are a valuable source of guidelines and recommendations Focus on how to knowledge Traditional repositories tend to become no more than a library of documents They should do more however mak
15. they really want it to be correct but really testing it they won t o It is described in literature that testing in Engineering Automation is often approached differently than testing in regular Software Engineering due to its special nature To what extend do you recognize these The developer is the user so that the entire usage period is an iterative testing and improvement period Part Il Initial Study Appendix B All participants agree During the project requirements change as it becomes clear what really needs to be done I4 notes that with concurrent engineering this is necessarily the case the external world keeps changing and the software needs to keep up It is not always known what the output should be when testing rather the software is run and it is checked whether the output is sensible Quite true most participants indicate that they do have a vague idea of what should come out but it is true that the expected output is not a crisp value In some cases hard to test theory independent from implementation since the implementation serves to test the theory Only few indicate they encountered this Most would question the implementation rather than the theory Defining a test is difficult because there is no formal list of what has to be tested All but 14 and A1 agree I4 would always use a formal or at least an explicit list The opinion of A1 is that at least for his area testing does not require a form
16. A useful way to think of the system is as individuals each with their own responsibilities 2 4 1 Server subsystem responsibilities Host the web client Generate pre configured code client executables Provide a CRUD API for diagrams blob data Provide a CRUD API for design data elements JSON data Provide a CRUD API and hash summary for code files blob data Generate the bootstrap package zip file including GenDL code JSON data to blob data Parse GenDL code files blob data to JSON data List the inconsistencies between the design and the code JSON data Web client subsystem responsibilities Provide a diagram drawing interface Provide a project overview tree Display the list of inconsistencies Provide inconsistency resolution options o Generate and show code snippets o Import design elements from the code Code client responsibilities Monitor files for changes based on their content hash Push file changes to the server Show an icon and menu in the system tray user interface Compile into configurable executable Note that both the server subsystem and the web client are responsible for code generation To avoid code duplication the same code generation templates are used by both the client and the server This will be elaborated on in the server and web client subsystem chapter 2 5 Communication API The communication between the server and the clients is done though RESTful HTTP i e the HTTP verbs GET
17. One directional change propagation simplifies the problem by limiting the propagation from the higher level to the lower level only Three mechanisms proposed for this are partially editable code and three way diff Angyal et al 2008 as well as transformation matrices Tri amp Tho 2012 With partially editable code the code generated from a higher level model is divided into editable and non editable sections Typically this is done by placing commented tags or markers on the border of the editable sections in the source code or splitting editable and non editable sections into separate files if possible Only the editable sections or files will be preserved when the code is re generated Angyal et al 2008 The drawback of this approach is the limited robustness when changing more than simple implementation details merging design level changes performed in the code into the design easily becomes problematic Williams 2006 With three way diff the re generated code is compared to previously generated code The difference is then applied to the actual code file which might have been manually edited The main drawback here is the limited robustness of the merge operation it is difficult to guarantee that the final code file is even syntactically valid let alone correct as percetved by the developer Angyal et al 2008 To avoid the fragile merging one could use transformation matrices to generate lower level models from higher le
18. Thesis article 27 Those who used it as a documenting tool for existing code found it to be a fast and accurate documentation tool faster and more accurate than alternative ways of documenting Nearly all users expressed their appreciation for the modern interface One user noted that he could even introduce the tool and diagrams to those who did not know GenDL Some users ran into bugs which made them loose time Most users also found missing features that in their opinion would have been valuable and time saving like integration with the GenDL compiler and full automatic diagram creation 6 5 Risk of reflexivity The outcome of the research might be distorted due to reflexivity Reflexivity refers to the possibility that a participant responds and behaves according to a perception of what he thinks the researcher expects For example spending more time explaining the intended usage of GenDL Designer increases the risk of finding an artificially high conformance to that intended usage In other words there is a risk that the findings of this research are not valid for realistic engineering environments because there the particular experiment coordinator and his guidance are not present Therefore the interaction between the users and the coordinator is described here The get users started they were introduced to the user interface with a small demonstration This is also when they were informed that manuals and screencasts were availab
19. This documentation is supposed to be useful for the author himself as well when he wants to pick up his code later again Apart from this concise high level documentation and comments in the source code no other documentation is expected I2 Design documentation examples and source code with good comments is everythin need g you A2 An activity diagram would be handy so you know what happens The connection with the actual code must be clear however You need to see how elements in the diagram map to what you have to do in the code 14 Making the engineering process clear is exactly what we had in mind with a recent research tool project We choose to implement the tool as spreadsheets with scripting The sheets in the workbook correspond to the steps It is fairly linear in terms of its sheet layout so that you can go through each of the steps Quality code is generally described as clearly documented and well structured The correctness is also considered to be part of the code quality Some attribute high value to performance and conciseness while others in particular novices insist on simple to understand and easy to read code I1 Code from more experienced team members can be so compact it becomes difficult to understand 3 4 Discussion In this section the results of the interviews are discussed and compared to the conclusions of the literature review on Engineering Automation in chapter
20. as design documents The second part is meta data about the software An example interpretation of the first part the software and related development artefacts is the Engineering Knowledge Resource EKR Bermell Garcia et al 2012 An EKR consists of o Knowledge informal and formal knowledge the source code and traceability links o Process the workflow that can be performed with the knowledge o Cases a history of process applications preferably automatically collected For the second part meta data to describe engineering automation applications was searched for but not found As an alternative 3 reports on similar packages has been consulted and compared Basili amp Rombach 1991 gives meta data for reuse candidates in general software engineering Stokes 2001 provides a list of items to mention in a user guide of Knowledge Based Engineering applications and in Shull et al 2004 requirements for a lab package in Part Il Initial study 51 Experimental Software Engineering are stated here a lab package describes the procedure and background of an experiment for the purpose of reproduction in a similar environment Table 6 1 compares the proposed meta data T RA 3 2 Nek ae ah ZT Efe Sy He 23 a g S Eg Roe Se 3 8 vo 3 aD Mo Sg n Bs aA Ss Ba 6 O SY ao ia g NS 5 z 8 sane 8 3 3 g Name X Type
21. effectiveness evaluated This project will be carried out at the chair of Flight Performance and Propulsion FPP of the Delft University of Technology TU Delft in cooperation with EADS Innovation Works IW These are the organizations in which the research will take place an academic and industrial one respectively The research method adopted for this project is Action Research Avison et al 1999 Baskerville 1999 It was selected because of its suitability for studying human organizations and its ability to solve problems in human organizations Action Research is well established in social and medical science and is increasingly adopted in Software Engineering research Action Research is based on the assumption that deep understanding can be gained from introducing changes to a particular research environment and studying the effect even when the Part Il Initial study 2 exact same experiment cannot be repeated because the research environment is a human organization Due to practical limitations research involving organizations is usually limited to a few experiments with similar conditions rather than many experiments with nearly identical conditions The major limitation of Action Research is that generalizing results must be handled carefully reproducibility and refutation are not supported as well as with some other research methods Therefore there is a large emphasis on repeating studies in similar environments Action
22. management features should be expanded to deal with large projects and other target platforms than GenDL should be included This will allow experiments in which GenDL Designer is deployed to a large and diverse set of Engineering Automation developers in industry and used in realistic projects 9 2 Extending GenDL Designer The issues from the initial study GenDL Designer does not address are the lack of systematic testing and the need to document the high level steps of an engineering process This section provides a proposal to incorporate these aspects in GenDL Designer 9 2 1 Systematic testing Just like code generation can encourage documenting test code generation could encourage testing If each design element would have tests associated with it test code or a part thereof could be generated along with the rest of the code Tests are developed anyhow now mostly as example runs Generating the tests from a clear and convenient overview can be about as fast but makes them explicit and persistent 9 2 2 High level process steps The initial study pointed out that engineers find it helpful to see the steps of the engineering process clearly when working with engineering automation software In contrast GenDL applications are not supposed to have these steps encoded in them information should depend on other information and the GenDL framework automatically determines the right order of execution The notation currently provided by
23. the urls form a tree structure like a file system To keep this structure as clear as possible the flat list of routes required by webapp2 is generated from a tree description The tree acts as a table of contents of the serverapp functionality It makes the application more self documenting The transformation step also allows automatic escaping of regular expression characters which also improves clarity An extract of the tree structure is shown below The full an up to date tree can be found in src serverapp routes py Part IIl Code report 42 rest_interface un StartPage projects html projects ProjectsHandler json projects ProjectsJsonHandler Project key id Project dt projects ProjectHandler html projects ProjectHandler summary json projects ProjectSummaryJsonHandler design integrity html projects design DesignIntegrityHandler infinity xml projects design DesignXmlHandler data Tree to list transformer The transformer itself can be found in src serverapp util route_generation py It generates routes like d serverapp handlers StartPage T cy serverapp handlers StartPage Jprojects Ahmi serverapp handlers projects ProjectsHandler projects json serverapp handlers projects ProjectsJsonHandler projects d serverapp handlers projects ProjectHandler projects dt serverapp han
24. 0 8 1 0 Consistency 0 1 Figure 6 Consistency and flow for each user in the trial runs The project of user 7 was not finished at the time of writing Flow evolution average 0 65 Flow from design to code 1 0 Part 1 Part 2 Part 3 Part 4 Figure 7 Evolution of the average flow direction throughout the project of user 2 Total design and code sessions 23 5 or 6 in each of the 4 parts Part Thesis article 26 Flow evolution average 0 27 Flow evolution average 0 54 10 1 0 0 5 0 0 Part 1 Part 2 Part 3 Part 4 i Part 1 Part 2 Part 3 Part 4 0 5 0 0 Flow from design to code Flow from design to code Flow evolution average 0 70 Flow evolution average 0 68 Flow from design to code Flow from design to code Figure 8 Evolution of the average flow direction for 4 of the 7 trial runs clockwise user 1 3 4 and 7 The project of user 7 was not finished at the time of writing Total design and code sessions 81 42 13 and 31 The flow discussed so far is an average for an entire project yet different parts of the project can have different flow directions To investigate this the sequence of alternating design and code sessions of each project is split in 4 parts of equal length and the flow is calculated for each part Figure 7 shows the evolution of the flow for the same trial run as Figure 5 The project starts with a large flow from the d
25. Error conditions are routinely handled O Assertions are used to validate logic O The code is exception safe O Errors are propogated not hidden C There are no resource leaks O The code uses multiple threads C It is thread safe O There isn t potential for deadlock Structure O There is no redundant code O There is no cut and paste programming C Continue to next section C Stop review here Statementlevel review Fill out the table below and move on fo a new sheet as required Rate issues on a scale from O cosmetic nice to have to 5 must fix File Line Issue Rating Continue on a separate sheet or mark up a paper copy of the code Follow up Record the outcome of the review here Conclusion Complete work by O Code OK O Rework and verify O Rework and re review Part Il Initial Study Appendix C Assigned verifier Source Goodlife 2007 PART III CODE REPORT Usage administration and development of GenDL Designer P J A R Dewitte B Sc January 26 2014 Code Report Usage administration and development of GenDL Designer Pieter Jan Dewitte contact pj dewitte gmail com version 0 5 2014 01 26 Gendl Designer demo gendidesign appspot com pro v ii demo Main diagram Consistency Design Code Settings Help center Main diagram Detail of lifting surface a gt E wing gt E lifting surface wv amp Popular buil
26. II System Administration Testing It was chosen to use functional tests rather than unit tests for the handlers since it was desirable to test the whole server subsystem including routing as realistically as possible They therefore test the overall behavior including routing datastore and memcache behavior rather than the behavior of a single function 11 Imagine a database configuration setting being wrong and the admin page to set it right again being unreachable because the admin cannot login Part IIl Code report 47 For each handler there are one ot mote functional tests A functional test e sets up datastore and memcache stubs e initializes the whole GAE app e sets up any data in the datastore or memcache required for the test e fires a request to the app simulating a the web client and the code client e checks the response and or side effects in database and memcache stubs The test test_serverapp test_handlers folder structure mirrors the src serverapp handlers folder so that each handler is tested 6 3 6 server_app code_tools The server_app generates two code tools on the fly the code client executable monitor exe and the bootstrap package Both are actually customizations of pre build files Code client generation The code client executable monitor exe is pre built but still has to be configured for the right project on the fly It is built with a non sensible configuration file which has to be repl
27. Last design change unknown Download Remove GDL Robot Last design change unknown Upload 2 9 User accounts Every project is associated to a user Only this user and the system administrator can see open and edit the project GenDL Designer cannot be used without logging in Part Ill Code report 11 Login Help center Design amp Implementation Support for GenDL Projects 2 10 Help center The help center contains the quickstart manual screencasts and frequently asked questions It can be accessed from any page in the upper right corner If you are trying to get familiar with GenDL Designer and you have not read the quickstart manual or viewed the screencasts you should go to the help center now http gendldesign appspot com help html Help Center Getting started Quickstart manual Read the Quickstart manual don t worry easy to read lots of screenshots Screencasts Watch the screencasts which introduce GendlIDesign the basics of using GendiDesign and some individual topics You can watch the entire playlist in i chronological order or select individual screencasts 1 Introduction 2 GendiDesign basics 1 Designing your application 2 Generating code 3 Propagating code changes to the design 3 Topics o Multiple diagrams o Reverse engineering importing existing GenDL code Frequently Asked Questions Code mirror tool e Do l risk loosing my code Answer 2 11 Admin panel Gen
28. PUT POST and DELETE are used in combination with URLs that try to point meaningfully to a resource like a path to a file on a file system For example 6 CRUD Create Update Delete Standard situation where it is needed to get create update and delete data objects 7 Blob Binary large object Chuck of binary data 8 JSON Javascript Object Notation a plain text data format comparable to xml but simpler Many programming languages support importing and exporting JSON Part IIl Code report 25 GET http localhost 8080 projects 4967730973245440 code 6093630880088064 consistency json is a request to get the consistency view on the code file group 6093630880088064 in project 4967730973245440 as JSON data Whether the file is available on the server as a static pre computed file or needs to be generated on the fly is of no concern to the client In fact the latter will be the case in this instance Part IIl Code report 26 3 Code conventions 3 1 Documentation The main documentation of GenDL Designer is the code itself together with comments in the code not this report Documentation outside the source code tends to become obsolete and forgotten hence this approach The following policy was kept in mind during implementation 1 First try to write the code as clear and understandable as possible The code should be self documenting if there is the opportunity for it 2 If then the code still is not cle
29. Projects Last projects Demo Pieter Jan Last design change unknown Download Create project poddns g 2eqp 4 A Important Notice the Feedback amp Support button on the right When you click it you can submit ideas feedback bugs but also vote for features and issues and see what is happening with them Or you can use the button to contact me directly Your feedback is really important and much appreciated Feedback browse ideas GCT LEGE of bd Contact support Integrate with code version control I suggest you Add support for Python 2 Export diagrams to printable format Make three levels of detail in diagrams Your email address _ Postidea Part Ill Code report Appendix A Project diagram Now continue by opening a project The project will show the main diagram when opened Firefox T Gendl Designer Demo Pieter Jan e Be t wv e B amp P G e Q v Ez demo pieter jan Main Diagram Consistency Design Code Settings Main diagram v Popular built ins E geom base box i gt Hi geom base gt i surf i gwl idns g y2eqp 4 AN Hod The diagram is where you can quickly create and edit the design of your application The screenshots below show the notation used in the diagrams Firefox e Gendl Designer Demo Pieter Jan gendidesign appspot com projects 5681726336532480 htmi fo B Ge P
30. Rae baad hava hae he a ee ee Re ee ee ee 3 1 Code mirror tool R6034 Runtime erfOt cesses 3 2 Cardinality field of a composition link in diagrams is not laid out nicely 3 3 Delete button does not work with multiple diagrams cesses 3 4 Tree collapses and expands for no apparent reason cccccesseseeessesessessesessessessesesssssessessessessessesssseesseseeseaseseesees 3 5 Line NUMDErS akiro Er EE EEE AEEKO EEKE EEA EEA EEA NEE EAE S EE S E EE E E S Oy SYSTEM ADMINISTRATION ccscccssesscsssssescsnssesasecseseeseresasnesssaseesaseeseseesenesacaesesaseesasecseseesenesacaeescaseenes 15 1 ENTRODUGIION EEE A AeA cose ER AAE AAS ARAE ES S EE E EE 2 THE CHOICE FOR SAAS AND GOOGLE APP ENGINE 3 SYSTEM REQUIREMENTS AND INSTALLATION 3 1 System requirements for a local installation sive its an sas ae ae 3 2 Getting theap plication cOdesci ihei cer ieekeitaacsanieateas AE E EE E as ARSE EEE E SAS E A Aa AEE EE AEE SEE SEEE 337 Wecal d ployiment taa rite Peseta acces AERA RR dat oe A el Oc 3 4 Deployment on the Google App Engine 4 ADMINISTRATION OF AN EXISTING INSTALLATION AT Admin passwotd cece 4 2 lt Admurmistrating userts and projectsia niin unaniianci A E EEA R A RT 4 3 Monitoring the system with on line logs cccceseeseesessesesessessessessssessessesesesseseeseessssessesssssssessessssseseeseseesesseess 4 4 Available logs sas a oath es a ASS Analyzing the lost cccsccvice sects anana RER RERE E RTIRA AARE SYS
31. Research involves an iterative research practise feedback loop consisting of a problem diagnosis phase an action intervention phase and a reflective learning phase This report covers the first problem diagnosis phase and prepares the action intervention phase This report is structured as follows In chapter 2 literature on Engineering Automation is reviewed to establish a background theoretical framework and collect contributions to the discussion that have already been made In a next step interviews are held in the organizations to validate the applicability of the literature review results and to provide deeper understanding of discovered problems by looking for additional explanation This is described in chapter 3 In what follows chapter 4 a several solution concepts ate reviewed and one is selected In chapter 5 the selected concept is reviewed in depth For future reference unselected concepts are reviewed briefly in chapter 6 The review of the solution concepts concludes the literature study phase of the project The next phase of the project will implement the selected solution concept and evaluate it in an experiment Part Il Initial study 3 2 Literature review on Engineering Automation and its background Existing literature related to Engineering Automation its development and its reuse is reviewed to establish a reference frame for the subsequent research Knowledge Management is reviewed because knowledge is develop
32. SALT e Ifthe SDK development server is running you can now log in as the user admin with the chosen password e To use the new admin password in live GAE applications re deploy the application The salt is used to protect the passwords of users For developers it is explained in part III how it works Part IIl Code report 18 4 2 Administrating users and projects e Log in to GenDL Design as admin e On the bottom of the Projects overview page there is a link to the admin panel e Inthe admin panel you can O Create users o Remove users their projects will persist but will only be accessible by the admin o Change the password of users o Create projects o Remove projects like users can when a user removes a project it is not really deleted it only becomes invisible o Remove projects permanently Code and Design integrity buttons are only for development see Section ILI of this report Currently each project has exactly one user Giving multiple users access to the same project is at the moment not made convenient This scenario has been foreseen though The underlying data model already creates a Team with one User for each Project To enable projects with multiple users only minor adjustments are required such as controls to add more Users to a Team and safety mechanisms to prevent concurrent changes 4 3 Monitoring the system with on line logs With the SDK development server you can view log messages and error
33. There is more time to iterate and find optimal designs From the perspective of academia reuse is closely related to reproducibility and contributing to the body of knowledge Reuse facilitates validation and further research Software has a high potential for reuse in the form of both black box and white box reuse Black box reuse corresponds to reuse of complete code artefacts to solve a particular group of problems without additional human effort White box reuse corresponds to reuse of the underlying ideas i e the knowledge embedded in the code software can be executed and also necessarily contains the knowledge required to execute an engineering task The combination of description and demonstration make Engineering Automation software an interesting potential learning resource 1 4 Research approach The objective of the project is to develop a more detailed understanding of Engineering Automation and in particular of measures to increase the reuse of Engineering Automation software by conducting an experiment with Engineering Automation developers To give the experiment solid ground Engineering Automation will be explored first with a literature review and expert interviews Potential measures solution concepts will be collected and reviewed One solution concept will be selected as the experiment subject In the experiment which will be based on this report the solution concept will be applied and eventually its
34. Tokens integer lt 1 9 plus lt P Grammar rules program lt expression expression lt expression plus expression expression lt integer This grammar would recognize sums of integers Reducing an example input program to the start symbol could be done as follows what is in bold is reduced in the next step Input program 1 2 3 Lexer integer plus integer plus integer Parser integer plus integer plus integer becomes expression plus expression plus expression expression plus expression plus expression becomes expression plus expression expression plus expression becomes expression Part IIl Code report 35 expression becomes program With more grammar rules more complex languages can be parsed Note when the grammar rules are restricted to context free grammar rules the existence of a reasonably efficient parser can be guaranteed and in fact tools like PLY can generate this parser from the grammar rules This explains the use of context free grammar rules PLY grammar descriptions The grammar descriptions for both the PLY lexer and the actual parser have to be supplied as Python modules with mainly functions The docstring of those functions contains either the regular expression for the lexer or the grammar rule for the actual parser The function can perform some processing on either the text that matches the regular expression or the tokens and symbols that match the grammar tule An exam
35. What determines whether it is worth to develop some software What motivation and incentive is there e In literature some common views of engineers on software are described To what extend do you recognize these Oo Software is a research tool which is developed in the first place to get some output needed by the research at hand First you must get it working and then you could consider its usefulness in later projects o Activities with long term objectives such as documentation activities and training activities ate under resourced Part Il Initial study Appendix A o You are rewarded for publications and results not for software o Software developers are at the bottom of the research ladder The career aspirations of many involve moving away from developing software o There is a high lapse of developers o Maintainability is not considered until it is already a problem Current Practise in Engineering Automation Reuse 15 min e To what extend are applications currently shared and reused e What material is typically available to learn about how the software performs its task e To what extend is networking used to share and reuse Engineering Automation Requirements for Engineering Automation Reuse Relevance Assessment 5 min e Situation Imagine you hear during a coffee break about an application from a past project which might be relevant for your new project to learn from or even to develop further e What will you
36. a high performance language such as Fortran etc Examples of software that is not included in the definition are top of the market simulation packages CAD software and PLM solutions because that software is typically implemented by or with extensive support from software engineers rather than regular engineers A clarifying example of what is and what is not Engineering Automation is CAD software with a scripting interface e g CATIA V5 allows the execution of Visual Basic scripts In this example the CAD software written by a professional team of trained software developers is not Engineering Automation A script written by an engineer to automatically draw a common part feature is Engineering Automation Notice how the engineer is both a user and a developer It can be argued whether engineers should write software Writing software could be left to software engineers trained professionals who can develop quality software in a systematic and disciplined manner However there are practical arguments for not hiring or contracting a software engineer e Design and research work is highly iterative Engineers and researchers revise their choices and assumptions based on the results they find Software developers would have to keep up with changing requirements Both actors their work would frequently come to a standstill while waiting for the other e For small problems it might be faster and cheaper for an engineer to develop a solution h
37. a physical hardware platform or a software platform which takes care of mapping the provided model further down to machine code If the higher level model describes the solution completely and unambiguously the model interpretable by the target platform can be generated automatically An example is the LabVIEW simulation model shown previously in Figure 5 2 If the higher level model is not complete or formal enough the mapping must be performed manually An example is the implementation of the software design shown in Figure 5 5 in a general purpose programming language Wachsmuth 2011 Mapping in the other direction from a lower level model to a higher level model reverse engineering is more difficult A higher level model doesn t necessarily exist for a given lower level model since lower level models inherently allow for more variation than higher level models Limiting the solution space to sensible options is exactly what higher level models do Part Il Initial study 40 After forcing a lower level model into a higher level one anyway the information which cannot be expressed in the higher level model will be lost This mapping issue is the reason why propagating changes under forward engineering is much easier than under reverse engineering and explains the general lack of tool support in the reverse direction Angyal et al 2008 5 3 3 Code generation forward engineering Code generators map a higher level description
38. activities other than software construction time consuming e g designing documenting testing integrating supporting e They need to justify the time spent on those activities directly e In general they have limited training and or experience in Software Engineering e They iteratively revise the entire software solution throughout the project 4 2 Solution concept exploration The solution concepts were sourced informally from Knowledge Management and Software Engineering and are shown in Figure 4 1 categorized along a solution axis technical and behavioural solutions and an improvement area axis software quality or software sharing rs Engineering App Repository Communities of Practise L3 oF E 9 Engineering App Repository ENESENN E N E EE EE SA EEEN Quality Metrics Graphical Software e Design Tool Code Review we ss Coding Policy sos Technical solutions Behavioural solutions Figure 4 1 Overview of solution concepts From the Software Engineering side solutions were found in three books on Software Engineering Goodliffe 2007 McConnell 2004 Bicar amp Dogru 2011 and cover both the technical and social behavioural area graphical software design tools quality metrics coding policies and code reviews Quality metrics are combined with a repository because it is expected they will strengthen each other Part Il Initial study 26 From the Knowledge Management side a solution from the technocrati
39. already an encouraging result Still users who do not see the point of documenting have no incentive to use the tool For them further time savings must be pursued e g with better layout mechanisms The users did not need a background in Software Engineering to work with GenDL Designer This was as intended one barrier less when convincing test users and will be equally beneficial in real life engineering environments The users who started from scratch worked indeed iteratively revising along the way and GenDL Designer was suited for that approach Timelines show that they kept switching between the design environment and code environment The flow evolution plots reveal that information flowed in both directions throughout the entire project Users kept using GenDL Designer during these iterations Part Thesis article 30 8 Conclusions 8 1 Initial study Understandability and validity are currently the most important issues that impede the reuse of Engineering Automaton Later on more reuse will also require scaling up the internal team interaction on which reuse now relies and supporting others has to become more rewarding The success of solutions to these issues largely depends on how well these solutions take into account several non technological obstacles related to the current engineering automation culture There is only an incentive for the answers that will be obtained with the software Laborious tasks that do no
40. and implementing your GenDL applications You can also use it as a documentation tool and it might help you to understand how a UML diagram maps to GenDL code GenDL Designer is implemented as a web based application in combination with a small dropbox like utility to upload your code to the server The utility is provided as a download in the web based application There is no need to install any software whatsoever The best way to get started is to read the quickstart manual or watch the screencasts not to read this code report Both can be found in the GenDL Designer help center There you can also find some frequently asked questions The quickstart manual is also added to this code report in appendix A The remainder of this section of the code report is a more formal overview of all implemented features known bugs and their workarounds 1 The help center can currently be found here http gendldesign appspot com help html Part IIl Code report 4 2 Features In this chapter each of the features is described along with a screenshot to give an overview of the implemented functionality Most features relate to the main interface of the web application In the figure below the different parts of the interface are named Help center Diagram tabs Consistency tab Settings tab Firefox v i demo Main diagram Consistency Design Code Settings Help center 2 Main diagram Det
41. answered These answers are input for the discussion Interesting quotes made by the participants are added to clarify the results and make them mote tangible to the reader 3 3 1 How Engineering Automation software is currently developed The findings about software development style itself are split into two parts software construction and software testing Software construction The software is developed individually and moderately to very ad hoc without a formal process It is a highly iterative and interactive development process which is not separated from the process of using the software Going through the calculation process step by step and seeing the output leads to new insights Code is developed without coding standard or policy Several participants indicate that they would be happy to use one instead of the implicit conventions currently used if any Requirements are implicit from project goals Only few consider it feasible to write detailed requirements in advance Short term objectives and bugs are scribbled down on a notepad Design is not planned to a high degree At most some thought is given to the top level data flow what goes in what must come out and vaguely the steps to make that happen Part Il Initial study 18 I3 unless the problem is complex There is no prior plan for the code I just start doing it and see what happens see what falls out A2 I found it difficult to write down beforehan
42. check and try to find out about this application before you decide to actually invest a serious amount of time in understanding this application deeply Actual Reuse 10 min e What determines how well you can work with someone else s source code Both characteristics of the source code itself and other factors e There is a trade off between quality and time resources required With that in mind what is the required level of the following items if you have to work with other s their code o Requirement docs Design docs Tests Documentation amp traceability and on what levels Code quality what is code quality to your O00 0 Part Il Initial study Appendix A Appendix B Aggregated interview answers Aim of the Interview Engineering Automation software is software written by engineers which automates an engineering task This includes spreadsheets Matlab scripts pre and post processing simulation data etc Much if this software is not very future proof it is hard to adapt to new situations in future work and thus reuse is low In my thesis I need to find strategies to improve the reuse of Engineering Automation software Apart from finding strategies that can improve reuse it is also an exercise in finding the strategies which engineers are willing and able to adopt With this interview I would like to find out how Engineering Automation software is currently developed by engineers why it is developed that wa
43. complete 2 the design description is consistent and 3 the design description fulfils the requirements The notions complete consistent and fulfils depend on the domain theory and context During the process not only the design description but also the requirements are gradually completed along with the design description This model is generic enough so that it can be applied to the implementation of a software design 1 The software design must be complete 2 The software implementation must be consistent 3 The software implementation must fulfil the software design These conditions can be checked partially by the graphical design tool when synchronizing the design to the code Like requirements the software design is gradually completed along with the software implementation itself During the process the above conditions will frequently be violated and proper actions should guide the project towards a state where all conditions are met 5 3 2 Mapping between models of higher and lower abstraction While humans prefer higher level models with concepts and abstractions that relate to the problem domain computers are essentially limited to a low level model in which the only abstractions over the hardware are computing and memorizing values To overcome this tools known as compilers map higher level models to machine code More generally higher level models need to be mapped to a target platform forward engineering This can be
44. container for the tab contents is returned Part IIl Code report 58 7 3 5 Project tree modules The tree_display module provides the project tree functionality It mainly interfaces with the entity manager diagram manager and implicitly the diagram module The content of the tree is extracted from the entity manager which provides a tree of nodes and from the diagram manager which provides a list of diagrams Together with the built in packages and popular built ins that data is shown in a tree The tree subscribes to the entity manager and diagram manager If entities or diagrams are added changed or removed the tree will be updated The tree uses the diagram module so that its contents can be dragged to diagrams The built in packages and popular built ins are configurable in the settings at the top of the tree_display module source file This system was preferred over connecting GenDL Designer to the documentation of a live GenDL GDL installation because e it is much simpler to implement e itis faster e it is much more robust genworks com documentation pages is regularly unavailable e it allows GenDL Designer to operate without GenDL or GDL which is required because it runs on Google App Engine and desired because of the license Actually rendering the tree is done with generic_tree_display a project agnostic module which simply renders a tree based on prepared data 7 3 6 Diagram module Introduction The dia
45. differs from the earlier mentioned research techniques with a much longer tradition the notation for calculus is subject to strong conventions by now after a great deal of experimentation in the eighteenth and nineteenth century and the best practices for orderly and safe lab experiments are taught as straight forward guidelines to any engineering student to set foot in a lab At the moment that level of guidance is simply not available to engineers writing software Instead engineers spent a great deal of their time looking for a good and workable approach or they spent time working with an inefficient one The amount of understanding we can obtain with a research technique is limited by our ability to use that research technique If applied ineffectively we will find only a fraction of the results that we could have obtained If applied downright wrong the validity of the research results is under threat This motivates the need for research about how we do research We need to bring the guidance for doing research with software development to the same level as the guidance available for other well established techniques This thesis is part of that endeavor During my thesis investigated how engineers write software and what problems they encounter first with a literature review and later with interviews With this knowledge investigated several possible improvements and implemented one of them specifically for engineers a graphical s
46. display controls list color orange line thickness 2 Amn ER ROR CE TE KITS CTUCRED csssssssinnnnnnn Part Ill Code report 10 2 7 Project bootstrap package After a first initial design session where the most important diagrams are set up one typically wants to generate all code and start coding To get you started as fast as possible GenDL Designer offers a bootstrap package for your project It is a zip file that contains a standard project structure all code that can be generated and the code mirror tool required for consistency checking The contents of an example package are shown below Ex Organize v Include in library v Share with v Wr Favorites Name W Desktop J bin r3 Downloads d documentation D Dropbox d input El Recent Places H lib b Google Drive di output J source a Libraries Ji temp L Documents J test ad Music monitor exe Pictures monitor info txt al Unief README bt z Videos A full overview of the files included in the zip file can be found in appendix B The bootstrap package can be downloaded in the settings tab like the code mirror tool 2 8 Downloading and restoring diagrams You can back up and restore your whole design by downloading and uploading your project as XML This is for example useful when experimenting or when reverting the code to an earlier version This functionality is accessible from the projects overview page Last projects k Demo
47. e g follow a link e The diagram manager in turn uses the tab manager to create a new tab The tree display requests all design data from the entity manager The tree display requests all diagram data from the diagram manager e The consistency resolution module uses the entity manager to save design modifications 7 3 2 Entity manager module The entity_manager module creates the entity manager for querying and modifying the design model on the server Entities are the individual elements in a diagram the blocks and the relations They are stored separately on the server besides the diagram itself the server doesn t know how to interpret the diagram it is just an image to the server With the entity manager you can get one or more entities update them remove them adjust in which diagrams an entity is referenced etc Other parts of the web client can subscribe to the entity manager to receive notifications when any entity is modified The usage of this mechanism is demonstrated in an example in section 7 4 7 3 3 Diagram manager module This module defines the diagram manager It manages the tabs with the diagrams and communicates with the server to retrieve and store diagrams It has a subscribe mechanism so that other components can be notified of added and removed diagrams 7 3 4 Tab manager module This module defines the tab manager A tab manager allows creating activating and closing tabs When creating a tab the
48. et al 2010 Angyal et al 2008 Part Il Initial study 43 Automatic synchronizing propagates changes over traceability links These can be established manually or if that is too time consuming based on naming conventions and tracers Neumiiller amp Griinbacher 2006 Establishing traceability in the reverse direction might be difficult however due to the mapping issue When traceability between elements in higher level and lower level models is established a rule can describe how changes to an element affect the element it is linked to Detecting changes can be done in batch mode during a synchronisation activity or in real time which is more robust when making significant changes Williams 2006 One can imagine however that real time synchronization is more demanding to implement There is little room for mistakes in the synchronization system as the user doesn t have the chance to make a backup before synchronizing 5 3 6 Testing Testing should be considered during design already Firstly applications that are designed for test tend to be more modular maintainable and reusable McConnell 2004 Secondly testing as you write helps to find mistakes quickly Goodliffe 2007 Stokes 2001 This suggests that when designing in a design tool testing should be given attention too so that when source code is generated also tests can be generated Part Il Initial study 44 5 4 State of the art UML modelling tools with round trip f
49. even on whether or not it is used The user with negative flow user 3 might have stopped using GenDL Designer in a real project since he was rather creating documentation than designing Under the pressure of a deadline that documentation activity would probably have been dropped Part Thesis article 29 When used as a documentation tool the flow is naturally highly negative One user indicated that he found deficiencies in his code while documenting The flow evolution for this trial run lower right in Figure 8 confirms this After some documenting activity visible as highly negative flow the user fixed the issues at the design level and pushed these changes to the code visible as positive flow in part 3 of the project 7 3 User feedback In summary the feedback of the users is that using GenDL Designer reduces chaos adds structure and provides overview in the development process and in the application itself This directly and positively affects the understandability Users also appreciated the ease with which valid documentation could be created The deficiencies one user found while documenting illustrates that understandability and validity are related increasing the understandability uncovers mistakes that threaten the validity For people who start from scratch GenDL Designer was not perceived as a time saver but not as a burden either In other words the tool gave them the documentation more or less for free That is
50. is in the same order of magnitude The new products integrate tightly with a bundled editor instead of being an additional tool This enables real time automatic round tripping i e the design models are updated instantly when making changes to the source code Presumably this is more robust than the batch mode of Rose at the expense of the obligation to use the bundled editor Enterprise Architect Enterprise Architect from Sparx Systems is a low cost alternative to IBM Rational software A one off licence costs a couple of hundred dollars Enterprise architect allows the generation of code the reverse engineering of existing code without automatic layouting the model though and automatic batch mode synchronization Synchronization is fairly robust but if changes are not properly recognized there is no way for the user to intervene until the synchronization has been completed An interesting feature of Enterprise Architect is to generate code in a user defined language with templates For the current research this is however not usable because the source model is fixed to be generic UML and only forward engineering is supported for custom languages not reverse engineering or synchronization ArgoUML ArgoUML is a free and open source UML modelling tool Additional features include design support and feedback code generation and reverse engineering ArgoUML allows modelling use case diagrams class diagrams sequence diagrams coll
51. low Best practises are not picked up due to a lack of training the development process is left unstructured and little time is spent on testing and documenting Apparently the return of investment is expected to be too low I2 Testing everything takes too much time I4 I don t use automatic testing because I don t have the need for it Since I now develop individually I am more able to pick up what s happening A1 Your software they supervisors really want it to be correct but actually testing it they won t 3 3 3 How Engineering Automation software is currently shared and reused Code is reused only inside teams Reuse occurs by copying legacy code This seems to be because there is not enough discipline to keep a shared library working code easily gets broken or is moved A1 A shared repository introduces the risk that someone else breaks your code Their change might work for their input but it can break yours It happened and it took me half a day to find the problem When things like that happen you quickly loose interest in pulling changes from others and rather keep working on your own version None of the investigated organizations i e teams use a central repository for finding software Knowing about existing code is done through internal team interaction The ideas behind the software are shared informally if one asks for it Only in exceptional cases there is more documentatio
52. makes the datamodel already complicated enough Application level constraints A practical GAE needs to handle requests in at most a couple of seconds soft limit and must store data in chunks of at most 1MB hard limit To prevent crossing these limits it is sometimes necessary to put application level constraints in place GenDL Designer at the moment has one such constraint the amount of CodeFiles allowed in one FileGroup is limited to 100 that is 100 1lisp files More files would mean longer Part IIl Code report 45 consistency checking up to the point where the server cannot keep up Also with more files the aggregated parse results are more likely not fit anymore in one memcache entry 1 MB Tests Most of the datamodel especially the utilities are unit tested The test folder structure mirrors the src folder 6 3 5 server_app handlers server_app request_handler Introduction The handler is what actually handles a request Creating a handler is easy subclassing webapp2 RequestHandler and defining get put post and or delete methods is sufficient One request handler class handles all HTTP verbs for a given URL GET PUT POST and DELETE This fits nicely with REST e A given resource say data is referred to by a URL Similar resources have a similar URL design data is under projects my project design data code under projects my project code etc e The URL matches a route e The
53. of the interrelated digital models 34 37 What is well understood is the practice of generating domain specific models from a central product model such as generating finite element models from CAD data This corresponds to information flowing down More difficult is the propagation of changes up again Propagating changes by hand is time consuming and error prone leading to undocumented alterations and outdated documents 36 38 Propagating changes automatically requires tool support which is currently lacking A mechanism similar to incremental code and design documentation generation can support the propagation of changes up and down The first steps in this direction have already been made In 39 an approach is described for integrating two particular engineering disciplines through model synchronization XML data files in this case A further advancement would be a hub framework to integrate multiple engineering disciplines as for example envisioned in 37 The hub can provide the framework to check detailed subsystem descriptions against a higher level system description and propagate changes Note that this hub vision is a decentralized vision in contrast to the centralized vision of a single source model often found in literature 1 38 40 Each engineering discipline plugin for the hub would contain consistency checking rules and optionally provide advice on how to resolve the inconsistency The domain specifi
54. of the importance of multiple test cases and testing both realistic cases and extremes All participants except 14 use an ad hoc testing method of running the software with example input as the code is developed one off testing and inspecting the output Some 11 A2 even change the source code to be able to test a part of the code commenting replacing variables by fixed numbers In general there is not felt the need to keep individual parts of the code testable The software is tested by running it as a whole None of the participants not even 14 uses automatic testing for parts of their software None seem to be familiar with the practise or have a clear idea of the advantages A2 does run all his test cases in batch mode though but the output is inspected manually to see whether it is reasonable When one sees an error or strange value the typical approach is to print out intermediate values and look for the ones that based on experience ate suspicious O A2 testing everything takes too much time o A2 The problem is that when you make changes in one file and finally get it working the program will crash in another file So you keep fixing files for a week until you can finally commit But still you might have broken something you aren t aware of o A1 about a fellow student He needed results for him it wasn t interesting to make sure that his adjustments worked in all situations o A1 Your software
55. on http www python requests or 8 2 2 pywin32 Microsoft Windows interaction in Python To place an icon in the Microsoft Windows system tray your Python scripts need to interface with the Windows system The pywin32 library contains the necessary extensions for Python to do this These can be downloaded from http sourceforge net projects pywin32 This library is not included in the repository and needs to be installed separately if you want to run the code client from source code not the pre build executable or if you want to build the executable 8 2 3 Pylnstaller making executables from Python code PyInstaller can combine your Python scripts the Python runtime environment itself and any additional files into a single executable This executable extracts all files to a temporary location and then starts the program as usual You can get it from http www pvinstaller or Simply download the zip file and extract the files somewhere on your computer Part IIl Code report 64 8 3 Components 8 3 1 Introduction The top level file of the codeclient package defines MonitorApp which integrates a FileSystemMonitor with a SysTrayApp The FileSystemMonitor does the hard work monitoring files and reporting changes to an underlying file store in practice a ServerInterface object The SysTrayApp shows the GUI of the code client in the system tray The class diagram below shows which modules contain which classes
56. process or seeing output triggers new insights and revisions of the entire software solution The causes of low reusability have consequences that stretch further than missed opportunities to capitalize on previous investments Low understandability i e transparency and lack of validation make the contributions to knowledge questionable impede the reproducibility and undermine future research These consequences strengthen the need for an investigation into possible solutions Part Il Initial study 25 4 Solution concepts The literature review and interviews showed a very limited level of sharing collaboration and reuse of Engineering Automation software while reuse is highly desirable This chapter identifies several solution concepts and evaluates their potential to improve the level of reuse Eventually one solution concept is selected for further elaboration 4 1 Selection criteria The solution concept will be selected based on the highest potential to improve the level of reuse The potential will be evaluated with respect to two key issues as described in the conclusions of chapter 3 Engineering Automation software should be easier to understand and be mote systematically tested and validated In addition the solution concepts are required to take into account the profile the way of working and the incentive scheme of Engineering Automation developers also described in the conclusions of chapter 3 e Engineers find software
57. rather than grading the whole organization based on all processes together Also it does not prescribe an improvement path Paulk et al 1995 2 2 4 Sharing collaboration and reuse in software development Software development requires extensive knowledge from multiple domains to start with the application domain and the software domain Usually this knowledge is not all in the head of one single person but distributed between the developer and the external world and as such the developer needs to collaborate with the external world As a design activity software development requires collaboration with the right information on the right time As a distributed cognitive activity it requires knowing your way around through interaction and reflection Ye 2006 Two support mechanisms are available to support collaboration Along the technical axis cognitive support helps to interact with external cognitive tools such as repositories manuals and intelligent feedback critique systems Along the social axis mediating support helps to engage with knowledgeable peers This includes finding them but also motivating their participation Ye 2006 Ye 2006 provides a model for knowledge reuse in Software Engineering where first information is filtered based on a quick relevance assessment before deeper understanding is gained and the knowledge is applied Supporting each phase effectively requires a layered presentation of the knowledge Unf
58. readily found with search tools Use existing code documenting conventions Constructs such as functions and classes neatly always need some documentation to record e g their purpose When comments are written in the right place and formatted properly off the shelf documentation generators can extract this documentation In addition it provides structure to a large part of the comments in the code The Javadoc conventions are probably the most fully developed code level documentation standards currently available Part Il Initial study 57 7 Conclusion The purpose of this report was to develop a detailed understanding of the development and reuse of Engineering Automation software and to prepare the next phase of the project in which an experiment will be conducted based on the findings in this report The detailed understanding was obtained with a literature review and further deepened with expert interviews Literature and interviews showed that the level of reuse in Engineering Automation is limited The two most pressing issues that impede reuse are understandability and validity When these issues are resolved raising the level of reuse further will require scaling up the internal team interaction on which reuse now relies Also several non technological obstacles related to the current Engineering Automation culture were identified These must be accounted for when introducing change laborious tasks are skipped where possible there i
59. recent changes are indeed detected and processed Saving the timestamp in the database is expensive every time the timestamp gets updated the whole entity i e Project or FileGroup needs to be retrieved from the database adjusted and stored again Both functions can continue gracefully when the timestamps are not available the user interface can show unknown and the logger can take a snapshot anyway Therefore it is sufficient to store these values in the memcache and not in the datastore Project FileGroup by user Frequently it is needed to get all FileGroups the user has access to in a Project Calculating this is rather expensive because of the access rights Therefore this information is cached as a mapping from User keys to FileGroups FileGroup Content hashes and parse results for each code file In two situations the content of each CodeFile in a FileGroup is required when the code client starts it needs the hashes of each file and when the consistency tab is open the consistency analysis which is refreshed every 5 seconds needs the parsed contents of each file Retrieving and processing the contents of all files i e the entire codebase of a project is too expensive in terms of datastore traffic CPU usage and time Therefore hashes and parse results are memcached Utilities Utilities were kept outside the main datamodel file to improve the conciseness of that file The datastore and memcache abstraction
60. required boolean datatype string body optional parseType string value sting line optional string l Children name string olacss name string COMPLEX_EXPRESSION used when there are if clauses etc package string cardinality string constraints f name string hody parseType string value string l line optional string l methods name string parameters name string stereotype required optional datatype optional string description string J parameters_line alternative for parameters string Part Ill Code report 30 datatype string description string body parsetype string value string line optional string file optional string line optional string 4 2 2 Function name string description string parameters name string stereotype required optional datatype optional string description string J parameters_line alternative for parameters string In some cases the class data model is split child object diagram elements for example will contain the data in the format of Class gt children Part Ill Code report 31 5 Setting up a development environment 5 1 Introduction To edit the source code you can use any text edi
61. reviewers don t need to be higher in rank experts Reviewers need to know how good code looks like though If the amount of material is overwhelming the author might need to provide some overview but in general the code should speak for itself so that the reviewers avoid the same traps the author might have fallen into Everyone must have familiarized himself with the documents or even have reviewed them in detail before the start of the meeting The meeting itself is about finding defects not about finding solutions The defects are noted down in the meeting notes If something is unclear there is no need to go into detail about whether or not it is a defect it needs to be clarified so it is a defect For a standard code review the following sequence is proposed In a couple of minutes the author explains the purpose and structure of the code Then the reviewers can make structural design comments e g about the breakdown of functionality into classes or the file structure Then general code comments can be made e g about the overall adherence to best practises and standards Only after these general comments the code is talked through The last step should be done by someone else than the author After the meeting the meeting notes with the defect list are given to the author If necessary a follow up meeting can be scheduled but mostly it is sufficient for the author to go through all the defects on the list In the case
62. server This way the server knows about the current state of the code and can match the code to the design The code client has a minimal user interface a system tray icon to stress its unobtrusive profile It is available as a simple executable which doesn t need to be installed or configured the user just has to run it The code client subsystem is written in Python The changes are reported to the server subsystem over standard HTTP so that issues with firewalls are less likely The codeclient package has three main modules the monitor module which actually scans for changes the communication module which communicates with the server subsystem and the systray module which provides the graphical user interface Code modifications Python Runtime executable codeclient systray The required level of Python to understand and develop further the code client is fairly low The three most difficult aspects are using two threads using the Windows API for the system tray and building the executable with PyInstaller These aspects will be discussed later in this chapter Part Ill Code report 63 8 2 Technologies used 8 2 1 requests HTTP requests in Python Python v2 x has two built in libraries for doing HTTP web requests urllib and urllib2 They are unfortunately not very convenient to use Instead the requests library is used which wraps around the two built in modules More information about requests can be found
63. the Software Engineering community Secondly in the context of code generation UML is a semi formal language not amenable to formal analysis consistency checking and eventually code generation A possible work around is to create a UML like formal notation M ry amp Singh 2011 Part Il Initial study 39 5 3 Code synchronization A key aspect of the software design tool to be implemented is that it will provide a tangible benefit for software design work in the form of generated code At the same time the software design tool must take into account iterattve development code and design will be revised regularly To realise these features code synchronisation is investigated This section starts with the Knowledge Level Theory of Design and a discussion of mapping between levels of abstraction which provide a theoretical foundation for code synchronization Next the current practise in forward engineering code reverse engineering code and iteration of these are discussed Finally the impact of code synchronisation on testing is considered 5 3 1 Knowledge Level Theory of Design The Knowledge Level Theory of Design described in Klein 2000 provides a model that describes the conditions which must hold in order to have a solution for a design problem A design description design process output is a solution to the problem formulated by a set of requirements design process input if and only if 1 the requirements are
64. the UML diagrams is to take one or more screenshots Exporting them as a vector image e g SVG would greatly enhance the way the diagrams can be used in documents presentations etc The diagrams are currently rendered in the browser as SVG The only part not rendered with SVG but with HTML are the lt lt stereotype gt gt markers because of the lt lt characters If this is done with SVG as will the browser could store the diagram as SVG to the server in addition to the XML and JSON formats and users could download it as an SVG file 9 3 Off line documentation Continuing this idea not only the images but even the whole design could be stored along with the source code on the user his computer The user can already download the design as XML manually but this could be done automatically too With this system the code client uploads the code and downloads the design documentation whenever they change This is not difficult to implement but requires some time a good interactive viewer for the diagrams and underlying data must be created 9 4 Version control support Off line documentation allows the user to put his design documentation under version control This is a great way of supporting versions no new system has to be developed for managing versions and the user is already familiar with the tool of his choice Ultimately the code client should recognise when a different version of the code is checked out locally i e
65. the parser recognizes pre defined code snippets correctly These can be found in test test_plygendl test_fragments py Run these tests after modifying the parser description to reduce the risk of accidentally introducing new bugs 6 3 2 server_app Introduction The server_app package handles all requests of the web client and the code client except requests for static files Requests are handled with a Webapp2 app This Webapp2 app is the interface of the server subsystem to the external world The Webapp2 app uses GAE services such as the datastore the memcache and logging functionality to handle the requests Modules sub packages The server_app package consists of several sub packages modules These are discussed individually in the following sections e Routing routes requests to the appropriate handling logic e Datamodel defines the data structure and provides tools for storing retrieving and manipulating data e Handling defines the handling logic for each type of request e Code tools generates the code client executable and bootstrap package zipfile 6 3 3 server_app routes server_app util route_generation REST routes The system has been built with the REST principle in mind as described in section 3 4 This results in urls like http lt server gt projects html http lt server gt projects design diagrams SVg http lt server gt projects code The grey parts are variable As you can see
66. the potential of Knowledge Based Engineering in manufacturing design TU Delft 2013 H Stoewer Missing Link in the digital Enterprise Strategy Keynote to the MBSE Workshop in INCOSE Los Angeles International Workshop 2014 Object Management Group Systems Modeling Language SysML Specification v1 3 2012 Online Available http www omg org spec SysML 1 3 R J Hamann and M J L van Tooren Systems Engineering amp Technical Management Techniques 2006 M Bajaj D Zwemer R Peak A Phung A G Scott and M Wilson SLIM collaborative model based systems engineering workspace for next generation complex systems in Aerospace Conference IEEE 2011 pp 1 15 J Gro and S Rudolph Generating simulation models from UML A FireSat example in Proceedings of the 2012 Symposium on Theory of Modeling and Simulation DEVS Integrative M amp S Symposium 2012 M Lauder M Schlereth S Rose and A Schurr Model driven systems engineering state of the art and research challenges Bull Polish Acad Sci Tech Sci vol 58 no 3 Jan 2010 K Amadori M Tarkian J Olvander and P Krus Flexible and robust CAD models for design automation Adv Eng Informatics vol 26 no 2 pp 180 195 2012 A Maedche and S Staab Ontology learning 2004 World Wide Web Consortium W3C Resource Description Framework RDF Specification 2004 Online Available http www w3
67. to a lower level one Examples are compilers and existing UML modelling tools that can generate code for general purpose languages or at least code stubs These Computer Aided Software Engineering CASE tools can generate code from UML although it must be noted that mainly the generation of class stubs from a structural view class diagrams is established Generating code or code stubs from behavioural views such as activity diagrams is not widely supported Bennett et al 2010 Gessenharter amp Rauscher 2011 Several techniques are used to generate code That code must be understandable by developers in the case of partial code generation Goodliffe 2007 The most popular technique for model to text transformation is template based text generation A template possibly with limited scripting embedded is combined with parameters to fill slots It is the same technique that has established itself as a standard technique for generating web pages Bork et al 2008 Angyal et al 2008 An alternative technique is to build an Abstract Syntax Tree AST and use a pretty printer to represent this tree in textual form i e source code Angyal et al 2008 An example of source code and an equivalent AST are shown in Figure 5 6 With an AST a larger variety of code can be generated but the increased generality comes at the cost of a higher development effort def fib n FunctionDef if n FEDS return Name n if n 1 If Com
68. to change the BASIC program In terms of cognitive dimensions one would say that the LabVIEW model has a high visibility for dependencies and data flows but at the same time a high viscosity A strength of the textual notation is that related statements can be grouped while the LabVIEW model has to respect the dataflow lines which restrict the possibilities to group related elements Also note that a common pitfall of graphical notations is that commenting facilities are overlooked 5 2 3 Multiple views and levels of abstraction To deal with the complexity of increasingly large design models models can be split in multiple models each displaying a different view or a different abstraction level of the model and each consistent with the others Multiple views Two broad categories for views exist under varying names static vs dynamic models structural vs behavioural models product vs design process models etc Blaha amp Rumbaugh 2005 Stokes 2001 Object Oriented software languages such as the languages for Knowledge Based Engineering ICAD GenDL tend to have source code that is closely related to the static models Nevertheless most software is best modelled with models from both categories linked to each other A synergy between the product model and design process model exists a process view facilitates comprehension Bermell Garcia 2007 Describing KBE modules as processes makes them easier to understand to en
69. to develop the web client further is fairly high Javascript is not the easiest language and the amount of Javascript code might be slightly overwhelming The demands on HTML and CSS knowledge of the developers on the other hand are rather low Part Ill Code report 53 7 2 Technologies used Several existing technologies are used in the web client mxGraph is used as the basis for the drawing component jQuery jQuery UI and Dojo are Javascript libraries that make building web applications easier Bootstrap is the CSS styling library that gives GenDL Designer its looks Finally jQote is a templating system to be used in conjunction with jQuery 7 2 1 Javascript The web client consists mostly out of Javascript code Javascript is an object oriented language but it uses a very different flavour of object orientation than e g Java Often people attempt to use a Java like approach in Javascript but because the keyword this in Javascript is trickier than most programmers realize that quickly gets out of hand this in a function refers to the object on which the function is called not the object on which the function is defined as in Java An example my_function defined on my_object_1 var my_object_1 object_name My object 1 my_function function console log this object_name J Create a shortcut to the function window global_shortcut_to_my_function my_object_1 my_function This will print out undefined bec
70. to validate the findings from literature and to gain deeper understanding through practical examples and additional explanation The results from the literature review and the interviews are presented side by side in this chapter 3 1 Focus questions The initial study was set out to answer the following questions How is engineering automation software currently developed Why is engineering automation software developed the way it is How is engineering automation software currently shared and reused What would more reuse of engineering automation software require The answers to these questions provide the required understanding of both the needs and culture of Engineering Automation developers in particular in relation to reuse 3 2 Interview participants In total six interviews were held with engineering automation developers in industry and academia All participants devoted a large part of their time to writing software Except the oldest participant all had followed basic programming courses in university All of them were self educated in particular programming language s while on the job Four participants were recruited in an engineering company Airbus Group Innovations They represent a wide spread in work experience from a couple of months to more than ten years with a median of 1 38 years and used Engineering Automation for various tasks automating conceptual design and simulation workflows post processing s
71. un configured executable The un configured executable is built with an empty configuration file containing 256 spaces This file is added uncompressed to the executable so that later it can easily be replaced by a string of equal length i e the project URL followed by spaces up to the total of 256 characters The rest of the executable will not be touched and will work as it should This step in fact generates two files as explained before in section 6 3 6 the un configured executable and a Python file which contains the byte position of the configuration file inside the executable This step is performed by codeclient src builder build py It invokes PyInstaller on the main py file which starts the code client and logs errors to the server The results are placed in server src serverapp code_tools codeclient_generation so that the web server can perform the second step In build py two paths must be provided as settings at the top the installation path of Python and PyInstaller Then it can be run as a regular Python script PyInstaller builds the executable based on the specification it finds in monitor spec This file is created from a template monitor spec template in which the file target path and the project configuration file to use in the build process are filled out 8 4 3 Configuring the executable In the second step the web server reads the un configured executable and replaces the right bytes with the correct configu
72. way it is 3 3 3 How Engineering Automation software is currently shared and reused s sssssssssssrsssssssssessssrreessssrseessnsrerossnreeessss 20 3 3 4 What more reuse of Engineering Automation would actually require o DIRUS SILON EEEE NE A EEEE E E A N aksi on sede 3 5 CONCLUSIONS o sssscsssssssssssssessssessnsessnsessnsessascssnscnsnscsssscsssecsssecsssecsnsessnsessnsessnsessasessasessasessasessascssascasaessssscensessnsecsnsessasessnsessaressasensas 4 SOLUTION CONCEPT Sonn a e A A A E O O R 26 Al SELECTION CRITERIA EAE TE TATAE EEEE AEA 4 2 SOLUTION CONCEPT EXPLORATION s s vost eee nae ay seve tie 4 2 1 Graphical software design toolis ennn E EERE saves ESEE EEE SEA E EEA E SEEE EEEE 4 2 2 Engineering app repository with quality metrics 4 23 CODE TEVIEW E E AREPA 4 2 4 Coding policy 4 3 SOLUTION CONCEPT SELECTION 5 LITERATURE REVIEW ON GRAPHICAL SOFTWARE DESIGN TOOLS essseesssseeeseseerseseesssreesssee 33 ARENAL DI BLSE H KOIN DEMAS A AES AER N RR E ERARE A AE ARE 5 2 GRAPHICAL PROGRAMMING NOTATIONS 5 2 1 Cognitive Dimensions of Notations 5 22 Limitations Of oraphiGal motati on Ssi r G E E E E 5 2 3 Multiple views and levels of abstraction 5 2 4 Unified Modelling Language soe ais ae Pon aa aes 5 3 CODESYNCHRONIZATION x cccssvevenssoncescttasteacesenvosescugtetensngtesueaunseseucupsunetuanduses dandetes coven T E E E E E E 53l Knowledge Level Theo or DOS a a aaa a aaeei Aaaa a Ae Aaea 2 Su
73. would be a notation which makes sense to engineers Usually a dedicated modelling environment is made available that allows quick and convenient graphical modelling i e drawing combined with a code or even executable generation feature The shift towards modelling is currently one of the two main strategies within Software Engineering to deal with the growing complexity of software systems the other being agile software development Bicar amp Dogru 2011 Kelly amp Tolvanen 2008 Part Il Initial study 27 Domain Specific Modelling could be used for full code generation ie generating an executable from a fully detailed model as suggested by Kelly amp Tolvanen 2008 The simplicity of the approach is advantageous but the approach has several drawbacks for Engineering Automation developers Firstly it requires a complete model so that it becomes unwieldy and no longer provides an overview Secondly this approach is an all or nothing strategy whenever the pre defined building blocks of the domain specific language are insufficient e g because the insights have changed the right solution cannot be expressed For situations where full code generation is not suitable iterative code generation strategies exist which intend to preserve changes made to generated code Lauder et al 2010 Angyal et al 2008 Williams 2006 Neumiuller amp Griinbacher 2006 4 2 1 3 Supporting design reflection and documentation Creating a software d
74. zip it Then start monitor exe which is the code mirror tool It is a simple executable no need to install or configure it it is pre configured when you download it You can see it in your system tray when it is running SEFBVE ua 15 46 0 The code mirror tool will scan the folder and the subfolder where it is placed and send all updated files to the server as long as it is running This way you don t need to upload anything manually Note typically it takes about 5 10 seconds before changes on your disk have propagated to the server and your browser To quit the code mirror tool right click the icon in the system tray and choose Quit Part Ill Code report Appendix A Project consistency code to design When you switch to the consistency tab again you will see that all inconsistencies have been resolved This is because the downloaded bootstrap package matches exactly with the design Freio _ Gendl Designer Demo Pieter Jan gendidesign appspot com projects 5681 20 htmit v El demo pieter jan Main Diagram Consistency Design Code Settings Main diagram gt Ewing Design Code Description high lift device v 1G Popular built ins geom base box i gt Hi geom base gt H suf f No inconsistencies detected Hurray gt E gwi yoddns 8 yDeqpee4 a The tool can also work in the opposite direction If you add elements your code you will see the inconsistency with the option t
75. 2 Part Il Initial study 21 The overall level of sharing collaboration and reuse found in the investigated teams matches the limited level described in literature What is shared is shared within teams in an informal way No close collaboration takes place Knowledge Management It was confirmed in the interviews that the level of documentation is low but the lack of stable sharing networks has to be nuanced Small sharing networks do exist informally within teams These networks are used to share pieces of legacy code and explain them upon request not for close collaboration Being asked for explanation is preferred over writing documentation because it saves time The lapse of developers within these networks is contrary to reports in literature not experienced as an important problem This small scale networking has several downsides Firstly this limits sharing to team members working in close collaboration Secondly knowledge remains mostly in the head of one developer Only when colleagues know about legacy code and ask for explanation the knowledge is shared These networks show the large importance of the social aspect in reuse both discovering existing software and reusing software are now closely linked to internal team interaction Software Engineering The overall development approach is as expected iterative and interactive and rather ad hoc Both requirements and systematic testing are notoriously absent The lack of r
76. 2007 advises to support professional end user developers in sharing software development knowledge and in testing It is claimed that proposed changes to current practise need to acknowledge the iterative nature of research and the position of software development as very much a secondary activity Part Il Initial study 14 2 4 Discussion and conclusions Engineering Automation is in the intersection of Engineering Design and Research Knowledge Management and Software Engineering Figure 2 3 Engineering Design amp Research Knowledge Management Software Engineering Engineering Automation Figure 2 3 Position of Engineering Automation in relation to other domains Knowledge Management is needed in Engineering Automation to manage the knowledge created in both the engineering and the software domain It comprises a technocratic school concerned with capturing and formalizing knowledge and a behavioural school concerned with knowledge in a social context a view that is gaining more and more importance Software Engineering is concerned with the best way to create and maintain software in general Software Engineering provides well established practises for predictable software development as well as development methodologies such as agile methods which seem to fit with iterative engineering work In practise the reusability of software frequently turns out to be a problem Engineering Automatio
77. 6 17 18 19 20 21 22 23 24 25 26 27 28 J Johansen and Jan Pries Heje Software Process Improvement SPI Manifesto Softw Qual Prof Mag vol 12 no 3 2010 A Abran and J Moore Guide to the software engineering body of knowledge no 2005921729 IEEE Computer Society 2004 H Lieberman F Paterno M Klann and V Wulf End user development An emerging paradigm in in End user development Springer 2006 pp 1 8 P J Dewitte Development and Reuse of Engineering Automation Software Literature and Interviews TU Delft 2013 Available in Part II D F Kelly A software chasm Software engineering and scientific computing Software IEEE vol 24 no 6 pp 119 120 2007 M T Sletholt J E Hannay D Pfahl and H P Langtangen What Do We Know about Scientific Software Development s Agile Practices Comput Sci Eng vol 14 no 2 pp 24 37 2012 F Elgh Supporting management and maintenance of manufacturing knowledge in design automation systems Adv Eng Informatics vol 22 no 4 pp 445 456 Oct 2008 G La Rocca Knowledge Based Engineering Techniques to Support Aircraft Design and Optimization Delft University of Technology 2011 G Schreiber and B Wielinga CommonKADS A comprehensive methodology for KBS development IEEE Expert vol 9 no 6 pp 28 37 1994 M Stokes Managing en
78. CELL_ADDED event GET http localhost 8080 projects 4967730973245440 code 6093630880088064 consistency json This request indicates that the user opened the consistency tab for project 4967730973245440 and his code file group 6093630880088064 within that project GET http localhost 8080 projects 4967730973245440 logging show_code_snippet subject Wing This request is special the web client only made this request to log that the user asked for a code snippet of the class Wing in this case The request has no side effects The logs of all these requests can be analyzed to give a good idea of how users interact with the system The application logs are log messages the GAE app filed either because it encountered an interesting event or because it encountered an error which is also an interesting but less fortunate event The application logs are very important because one of the interesting events captured are switches of the user from design to code or vice versa When such a switch is detected the consistency list is logged so that the inconsistency evolution can be traced through design code cycles The consistency logs are entries in the database which contain the full details of the consistency list at the moment of a design code code design switch 4 5 Analyzing the logs 4 5 1 Step 1 downloading log files bit by bit It would be inefficient to download the entire log file from Google App Engine over and over again Inste
79. Code data models which can easily be analyzed plygend1 uses the PLY library described in section 6 2 1 Parsing process The parsing process is shown in a diagram below PLY Source code Tokenize Tokens Source code Eran Classes and Functor Python objects Convert to JSON Code model JSON Modules The main interface to the outside world is the plygend1 parse module Furthermore plygend1 has these modules plygendl parser_lex defines the tokenizer lexer rules plygendl parser_yacc context free grammar rules as well as processing logic That logic produces objects of the types defined by plygendl ast e plygendl ast describes the data structures produced by the parser Part IIl Code report 41 Compilation PLY compiles the parser description lex and yacc into parser tables and caches them on disk for later runs This is done in the compiled subdirectory PLY will update the parse tables during the next run after the parser description was changed This has to be done locally though GAE does not allow writing files during production In the test test_plygend1 directory there is a script run_build py to clean the compiled subdirectory and run build the parser Tests There ate automatic unit tests which verify whether
80. DL a KBE system Based on literature about graphical programming notations and code synchronization the following recommendations are made generate both code and design documentation incrementally and iteratively use a customized UML notation without unnecessary detail provide soft consistency warnings and generate and parse code with industry standard techniques Part Il Initial study The expected outcome of the project is clarity about the feasibility and potential of incremental code and design documentation generation in an engineering environment Also it should yield a more detailed understanding of how Engineering Automation software is developed and how that activity is best supported Part Il Initial study Table of contents 1 INTRODUCTION seis cescase se scieasesoctuateasutecdesestusdesentestesecteesssavancdesocenedeesudessesvcteassoveteddensceasdubestedssescesssvesogessves 1 TT SOFTWARE DEVELOPMENT BY ENGINEERS rart ittir eiea taoa aT E ATE EA LEA RATE AE 1 1 2 DEFINITION OF ENGINEERING AUTOMATION 1 3 REUSE OF ENGINEERING AUTOMATION 1 4 RESEARCH APPROACH anaona n Aa aa a a e a a aaa araa E Aaa ea ee ae aE AAAA araia E aaa R E E 2 2 LITERATURE REVIEW ON ENGINEERING AUTOMATION AND ITS BACKGROUND 4 2C CKNOWEEDGE MANAGEMEN rena REER E EE AREE REER E O REA ARER 2 1 1 Definitions of knowledge and Knowledge Management 2 1 2 Knowledge Management theories s ssssssssssssrsesssssressssreeessss
81. DL Designer includes an administration panel to administrate users and projects The admin panel can be accessed form the bottom of the main projects page Part Ill Code report 12 Logged in as admin Logout Users Pieter Jan Overwrite password aie Overwrite password Create user pas Haon Projects Demonstration Pieter Jan Last design change 11 seconds ago me Code amp Design integrity E permanently Create project Part IIl Code report 13 3 Known bugs 3 1 Code mirror tool R6034 Runtime error Some users get the following error message when running the code mirror tool Microsoft Visual C Runtime Library xs x Runtime Error Program D ftian Dropb R6034 An application has made an attempt to load the C runtime library incorrectly Please contact the application s support team for more information OK Oddly enough the code mirror tool does work The icon will appear in the system tray and the files will be synchronized You can ignore the error message windows shows 3 2 Cardinality field of a composition link in diagrams is not laid out nicely The cardinality field currently sometimes overlaps with other elements This does not affect the functionality of the application To export the diagram for print consider doing minor fixing with a simple drawing program like Microsoft Paint 3 3 Delete button does not work with multiple diagrams T
82. GenDL Designer reflects this it is geared towards modeling compositions of GenDL classes entities with information slots Part Thesis article 33 that depend on other slots The current notation is unsuited for modeling high level process steps GenDL Designer must provide both a notation for the high level process and sufficient incentive to model the process Since the high level process is not explicit in the code code synchronization cannot provide this incentive 9 2 3 Proposed modifications It is proposed that a second drawing canvas for a procedural view is added and that the models on both canvasses are connected to each other The procedural view could be based on Unified Modeling Language UML activity diagrams 32 Process steps and classes can be created on the respective canvasses A process step and a class can be related to each other though sharing slots attributes Slots are both part of a class as an attribute and part of a process step as an input or an output Relating a class and a process step can be as simple as dragging and dropping slots Each step will eventually have inputs and outputs Users can clarify the process steps and provide test cases for them by entering several sets of example values for the inputs and outputs The consistency between the models can be checked each slot should be the output of at least one step slots can only be input if they were output in an earlier step and each step mu
83. Google App Engine It is very different from relational databases such as SQL database because of its focus on scalability The GAE datastore is a hierarchical datastore a bit like a file system Every entity has either a key name or an id and optionally a parent The key name or id is unique among all children of a parent Any entity can serve as the parent of other entities The parent and the key name or id Part IIl Code report 37 cannot be changed after creating the entity They determine the fixed path and corresponding key which is used to access the entity in the datastore The GAE datastore is also an unstructured datastore Any entity can have different properties However the default interface provides a framework to impose a structure based on object oriented modeling GAE memcache Both the datastore and memcache persist data The datastore is very reliable but datastore operations are costly in terms of time and quota limits The memcache on the other hand might get cleared at any time but is fast and more importantly provided free of charge The usual paradigm is 1 retrieve data from the memcache 2 if not available in the memcache a get the data from the datastore b store the data in the memcache for the next time The memcache can also store data not in the datastore e g aggregated data which is updated on every change of the datastore GAE quota Google App Engine provides free daily quota so you c
84. Hover over them to see their respective mouse shortcuts 2 4 X_ E Delete elements if DEL key broken please use right click Bug The remaining buttons are discussed at the end of this manual in Advanced Features Part Ill Code report Appendix A Project consistency design to code The next step is to compare the design to the code This is done in the consistency tab Firefox gt Gendl Designer Demo Pieter Jan gendidesign appspot com proje 480 ht mit v Hi demo pieter jan Main Diagram Consistency Design Code Settings Main diagram gt Bwing ie high lift device Warning There are no files to analyze v Popular built ins geom base box i Info To set up file synchronization to the server go to the settings tab Once files are available on the server they will be analyzed and the results will be shown here gt iz geom base gt ii surf gt Es gwi The UML class wing inthe design doesnthave a corresponding define object wing declaration in the code Design Code Description code snippet The UML class high lift device inthe design doesnthave a corresponding define object high lift device declaration in the code poddns g y2eqp 4 a code snippet When using this for the first time you will see a warning that there are no files to analyse yet We will add files later The consistency tab shows that there are two classes in the design whic
85. Important GenDL designer doesn t change your code That s a guarantee On the downside that also implies that automatic resolution options in that direction are not available The provided options are semi automatic the code is generated but you need to copy paste or save the code yourself During the consistency checking process errors might be found in the design or code such as duplicate names or invalid GenDL files These errors are shown on top of the table if any If no code files have been uploaded yet a warning is displayed as well as information on how to set up code synchronization In some cases it is undesirable to show particular elements present in the code also in the design if they are merely an implementation detail in the code and do not contribute to the overview of the system they do not belong in the design In such a situation you can put implementation_detail in the documentation of that element define object slot function etc This will suppress the inconsistency warning The consistency tab displays in the upper right corner the last modification time of the design and the code This allows you to verify whether the diagrams were saved properly and code mirror tool has synchronized properly Part Ill Code report 9 2 5 Code Generation To stimulate the creation of UML diagrams GenDL Designer provides the capability to generate code from UML diagrams Code can be generated either by right cli
86. J Segal Some Problems of Professional End User Developers IEEE Symp Vis Lang Human Centric Comput VL HCC 2007 pp 111 118 Sep 2007 J Segal Models of scientific software development First Int Work Softw Eng Comput Sci Eng 13 May 2008 Leipzig Ger 2008 J Howison and J Herbsleb Scientific software production incentives and collaboration in Proceedings of the ACM 2011 conference on Computer supported cooperative work 2011 pp 513 522 M Lehman Laws of software evolution revisited Softw Process Technol pp 108 124 1996 J E Hannay H P Langtangen C MacLeod D Pfahl J Singer and G Wilson How do scientists develop and use scientific software in Proceedings of the 2009 ICSE Workshop on Software Engineering for Computational Science and Engineering 2009 pp 1 8 R Sanders The development and use of scientific software Canadian theses 2008 P Goodliffe Code craft the practice of writing excellent code No Starch Pr 2007 S McConnell Code complete Microsoft press 2004 Y Ye Supporting software development as knowledge intensive and collaborative activity in Workshop on interdisciplinary software engineering research WISER 06 2006 p 15 D Cooper GenDL Generative Programming and KBE system embedded in Common Lisp 2013 Online Available https github com genworks gendl Part Thesis article 37 14 15 1
87. MASTER OF SCIENCE THESIS Development and Reuse of Engineering Automation Incremental Code and Design Documentation Generation P J A R Dewitte B Sc 25 February 2014 Faculty of Aerospace Engineering Delft University of Technology 5 a AIRBUS T Delft Unversity of GROUP Development and Reuse of Engineering Automation Incremental Code and Design Documentation Generation MASTER OF SCIENCE THESIS For obtaining the degree of Master of Science in Aerospace Engineering at Delft University of Technology P J A R Dewitte B Sc 25 February 2014 Faculty of Aerospace Engineering Delft University of Technology G TUDelft s AIRBUS GROUP This report is the result of research conducted at the Delft University of Technology with support of Airbus Group Innovations formerly EADS Innovation Works Copyright P J A R Dewitte B Sc DELFT UNIVERSITY OF TECHNOLOGY DEPARTMENT OF AERODYNAMICS WIND ENERGY FLIGHT PERFORMANCE AND PROPULSION The undersigned hereby certify that they have read and recommend to the Faculty of Aerospace Engineering for acceptance a thesis entitled Development and Reuse of Engineering Automation by P J A R Dewitte B Sc in partial fulfilment of the requirements for the degree of Master of Science Head of department First supervisor Second supervisor Reader Reader Dated February 25 2014 prof dr ir L L M Veldhuis Delft University of Technology Depar
88. TEM DEVELOPMENT ccccccsssssessssessscsssscscsscsessasessesassccesasaeessanessasesensesensassseseasasessasessecesecceeasasenses 21 1 EINGT ROD UG TION ech E A AE EOE O OEE EN EEEE SL Na Nal ts bees ec 2 SYSTEM COMPONENTS OVERVIEW Qed ABMMCHOM AAS 1 0 PEER EENEN E EA EAE AAN AEE AS A N E 2 2 Deployment diata srana A a A A AAEE EAE VINAVER AA T 23 Data flows secietcc chunk eh 2 4 Subsystem main responsibilities 2 5 Communication API tes ats sts Ia ahs 7 3 CODE CONVENTIONS Ahasaia cian a AA S AA A Sil Documentation n i E E E O A EAA Rr Palen So ER 3 2 Folder structure a is sd a an ay S ede oes 3 9 gt lt Dependencies ian aeie a ea eich siath mana E E NAA EE A AE A AANS 4 DATA MODE S rae e A ak a A ee Ee Ee ee ee 4 1 Database data model wi 4 2 Unified Design Code data models ccc00 5 SETTING UP A DEVELOPMENT ENVIRONMENT aes 5 1 TntroductoN sarsar rE a EEK A EE KEE EEE KE E KEE AEE canna dandebebdoatanc EE EE 32 DA gt CEE a ETNIE E E EEE N ESS ARR EE 32 5 3 e Cr ating an Eclipse Projectie sieer rsii tir neee r sia a EEEE aE ORATE E EAEEREN ER SEES S AETAT VESE 32 Part Ill Code report 6 SERVER SUBSYSTEM virinan A EN AE E OL tak aR 34 Oal sn pa 16 1 u 6 n Reenter ene PRR Sn ee eee 34 6 23 Technologies Used sive cessose a RRRA TREE REAN ARS E denies ANDRN 35 G3 Components eoa ww Al 6 4 Serverapp configuration and loading sees Dl 7 WEBCLIENT SUBSYSTEM esccsscssesseesee
89. The code elements represent code files added to the project Each user in the team has his own set of files he can write to a FileGroup Under the FileGroup CodeFiles are stored 4 2 Unified Design Code data models Some concepts are present in both the design and the code e g classes and functions Although the same concept is represented differently in the design and the code a single data model is used to describe the same concept This is possible because the design and the code are basically at the same level of abstraction except that the code contains extra detail Single standardized data formats were created for class and function entities to be used anywhere where model entity data Part Ill Code report 29 is created transmitted or processed in the diagram in the datastore in the code generator in the parser and in the consistency checker The data needs to be stored as text in the datastore and must be processed both client side e g in the diagram editor and server side e g in the consistency checker This situation can be handled conveniently with JSON data The structure of the JSON data was inspired by the XML format to describe KBE classes used in the IDEA project This structure is described next 4 2 1 Class name string mixins name string package string external boolean description string ownedAttributes name string stereotype input computed
90. Y Ype Main Diagram Consistency Design Code Settings Main diagram gt Ewing high lift device Popular built ins geom base box 1 gt Hi geom base gt ii surf a wil ainput slots input slots profile fueltank volume chord computed slots length acomputed slots a D 0 A nm objects flap Part Ill Code report Appendix A high lift device input slots ainput siots fueltank volume computed siots functions objects flap Currently you can draw e Classes with slots e Child objects e Functions e Relations o classis composed of child object Oo child object has type class o class has mixin class O function ot class uses function To create new elements double click somewhere on the canvas To create relations hover with the mouse over the starting element and drag the blue arrow that appears to either an existing element or drop the arrow on an empty area in the diagram to create a new relation and element at the same time Deleting elements is done with the delete key on the keyboard or with the right click menu Note the delete button does not always respond that s a bug However the right click menu always works Saving is not necessary the diagram is saved automatically There is a save button for peace of mind you can click it hover it to see how long ago the diagram was saved There are buttons on the bottom to zoom in and out
91. a software design tool based on incremental code and design documentation generation was selected as the most suitable approach to start tackling the issues identified To contribute to understandability and validity and ultimately reuse the tool aims to encourage the creation of accurate design documentation and to encourage the creation of that documentation before implementing the corresponding code Creating a design beforehand encourages a well thought and understandable application structure yet this is rarely done in an Engineering Automation context The approach was implemented for a specific community of Engineering Automation developers namely users of the GenDL software framework The resulting tool GenDL Designer features a simplified version of the Unified Modeling Language continuous consistency checking with the code and support for incremental resolution of inconsistencies e g by generating code skeleton fragments or by proposing design diagram modifications GenDL Designer was developed with Engineering Automation developers in mind and therefore differs significantly from general software engineering tools with similar objectives To address the potential and feasibility of incremental code and design documentation generation for engineering automation development a large scale academic experiment with GenDL Designer is planned in spring 2014 In anticipation of that trial runs were held which only allow for preliminary con
92. aboration diagrams statechart diagrams activity diagrams and deployment diagrams Part Il Initial study 45 The design support and feedback provided by AgroUML consists of to dos design critics and checklists The motivation for providing these feedback systems is grounded in cognitive psychology To dos mark missing elements in the UML models The design critics are about errors in the model expert designer s advice target platform specific issues etc Many critics offer to automatically improve the design Checklists guide users through complex processes The support and feedback mechanisms are such that they do not interrupt the designer but instead are available to the designer when he requires them ArgoUML can generate code from class diagrams There is no support for generating code from other diagrams such as sequence diagrams activity diagrams or statechart diagrams In the other direction ArgoUML can import existing source code reverse engineering The imported entities are then available to use in diagrams In practise there is no support for round trip engineering ArgoUML implements bi directional model generation i e switching between the UML models and the code but it is too time consuming to be practical the diagrams need to be re constructed manually However ArgoUML does handle the other issue of bi directional model generation the loss of information between transitions lower level implementation details are stor
93. aced in the executable The code client build process in fact generates two files the un configured executable and a Python file which contains the byte position of the configuration file inside the executable The web server reads the un configured executable on the fly replaces the right bytes with the URL to the project before sending the executable as a download to the web browser of the user Bootstrap package generation The bootstrap package is a zipfile with fixed and variable files The fixed files are in a pre built zipfile The webserver opens this zipfile adds the variable files such as generated code files and a configured monitor exe and sends the zipfile to the user Tests The code client generation and bootstrap package generation are not tested individually because they are already tested through their associated request handlers Those tests can be found in test test_serverapp test_handlers test_projects test_codetools test_codetools py 6 3 7 dataprocessing In line with the policy to move the heavy lifting out of the handlers two big data processing tasks were moved to separate modules consistency checking between design and code models and design model integrity checking Integrity checking is a debug tool only Consistency checking Consistency checking is one of the core features for GenDL Designer The server_dataprocessing consistency module provides a single function analyse design_data code_data which re
94. ad a script has been made to download the log files of the last N days If this script is ran regularly e g every week with N 10 a complete set of overlapping log files will have been obtained The download script can be found in experiment src download_logs py 4 5 2 Step 2 Aggregate log files The overlapping log files need to be aggregated into 1 log file before the analysis scripts can be applied This is done by a script in experiment src aggregate_logs py 4 5 3 Step 3 Run analysis scripts Several analysis scripts are available The common functionality between them in particular reading log files and filtering relevant messages is offered by the package in experiment src log processing calculate_project_metrics py plot_flows py and plot_timeline py process data for one specific project that poject has to be set as an input setting in those scripts plot_map py shows the consistency and flow metrics for each project together in 1 plot view_logs_admin py finally shows a list of all actions of all users to get an idea of the recent activity Part IIl Code report 20 Section III System Development Part IIl Code report 21 1 Introduction The third and last Section of this report explains the internals of GenDL Designer to the reader who wants to change the application functionality and behavior This Section can be read selectively it is meant as a reference to be used on the side when developing GenDL Designer
95. age step in which the MOKA models is translated into a KBE system description 6 Activate step in which the KBE application is distributed introduced and used These steps usually have to be executed iteratively This is shown in Figure 2 2 Part Il Initial study 10 Development of new KBE system possibly resulting in new system or process knowledge or new requirements representing changes in business needs New Knowledge provides a to Identify what is involved in order to Justify m Maintenance of existing KBE system requiring no longer Capture justified and verification of the required knowledge C stop gt before the knowledge engineer can Formalise and incorporate it into the corporate knowledge repository to ensure consistency and completeness prior to extracting the appropriate knowledge in order to Package it in the KBE system test the resulting system which may ready to i trigger a Distribute sian the KBE system to intended users and then Introduce it to and train the intended users so that they can Use the system to gain benefits Figure 2 2 Model of the KBE System Lifecycle Source Oldham amp Kneebone 1998 Part Il Initial study 11 Knowledge representation The second contribution is a representation for engineering design knowledge based on an informal an
96. ail of lifting surface a a a a X gt Ewing gt E lifting surface amp Popular built ins A g roy geom base base object surt box solid G surf cylinder solid surf extruded solid gt E8 geom base yoddns 8 y2eqpeae4 gt SE surf gt HE gwi Project tree Drawing canvas Feedback button 2 1 UML drawing The most visible feature of GenDL Designer is the ability to draw UML like class diagrams high lift device cinput Slots input siots profile fueltank volume chord computed siots functions functions objects flap Part IIl Code report 5 The UML notation is slightly tweaked compared to UML the options and possibilities have been reduced Also attributes and methods have been replaced by input slots computed slots and functions as they are called in GenDL Last but not least objects are drawn outside of the class box as a box instead of as a named edge This is to better represent the modeled product tree structure Currently you can draw the following elements by double clicking on the canvas e Classes with slots e Child objects e Functions e Diagram references You can relate these elements to each other When hovering over an element a blue connection arrow appears To create a connection drag and drop the blue arrow to another element Not all relations are allowed A green or red marking indicates
97. ake it a performance review 6 2 3 Material to review The review can be about code but also about design documents or other project documents The review can be a personal one to one or meeting like review In the latter case it is recommended to make it a formal review meetings need structure to be effective A formal review uses checklists which are continuously updated to focus the attention and has clearly defined roles for the participants moderator author reviewer and scribe It can be chosen to review every line of code some argue that the time invested will eventually pay itself back If not all code is reviewed one should carefully select the code to review Good candidates are o Central components Areas where most of the CPU time is spent as measured with a profiler Areas with a high complexity as measured with complexity analysis tools Areas with a high bug count Code written by new or untrusted programmers OQ 0 0 0 Part Il Initial study 54 6 2 4 Code review procedure The generally recommended procedure for conducting a review consists of the preparation the actual meeting and the follow up The preparation allows the participants to have an effective meeting later on For a less formal review the author distributes the documents to review and schedules a meeting at a convenient time in a quiet room For a formal review this is done by a moderator who also distributes up to date checklists The
98. al 2006 These professionals tend to have a highly technical field of expertise in which case they are used to formal languages and abstraction and hence tend to have few problems with coding per se Segal 2008 This is how they differ significantly from regular end user developers The following definition for professional end user developers is based on a description given in Segal 2008 2 develop further can mean both altering existing software components or build new software components within an existing framework or environment which makes development easier than general purpose programming Part Il Initial study 8 Professional end user developers are people working in highly technical knowledge rich professions who develop their own software in order to advance their own professional goals While they are very sophisticated in their field of expertise they have received little Software Engineering education or training A similar definition for professional end user developers is given in Lieberman et al 2006 except for the constraint on technical professions Note that this definition doesn t exclude professionals who attended courses on particular programming languages which is common for engineers nowadays Useful as these courses are they hardly go beyond the coding itself and don t treat Software Engineering in depth It must be admitted though that professional end user developers frequently turn out to b
99. al list if you know what you re doing Do you use one or more formal methodologies when developing Engineering Automation software development methodologies knowledge engineering methodologies None of the participants uses a formal methodology except for 14 when working in large projects o 14 I ve heard of MOKA when I was in the KBE team but it was perceived as a fairly theoretical overhead o A1 Using a formal methodology seems like a to involve a disproportionate effort o Do you and to what extend do you use these Incremental development All participants use this Char iterations with clear goals Highly varying some do some don t some a little Ai prioritized TODO list with features and bugs All participants indicate they have such a list on a notepad or similar Small sub task descriptions with example inputs and outputs Only I4 uses these Coding standards policies conventions Are they made available in the first place Only I4 is aware of any standards Several participants indicate that they would be happy to use one instead of the implicit conventions currently used if any Do you find yourself effective in producing software Effective in the sense that you can get it to the level you find sufficient within the time you find reasonable Answer varies Two common arguments are the lack of experience if they find themselves ineffective or the high reward that surpasses the time inve
100. ame string password_hashed string ModelEntity Diagram name string content string referenced_in CodeFile get_content_hash string The datamodel is defined in src serverapp datamodel __init__ py The rest of the datamodel package consists of utility functions to work with the data in the data store and memcache As usual the main documentation is the code itself together with the documentation in there What is discussed here concerns the rationale behind architectural decisions e Single interface that hides datastore and memcache details e What data is stored in the memcache instead of or in addition to the datastore e Utilities that are kept outside the main datamodel code Single interface that hides low level details The data model provides a single interface to the datastore and memcache for the rest of the application they simply request data and store data Behind this facade it tightly integrates datastore and memcache access to provide good performance and acceptable quota usage The interface is used by the request handlers which contain the bulk of the server_app logic The main functions that the handlers use from the data model are the get_ and get_ _key functions together with the put and delete methods of the retrieved objects Data stored in the memcache The memcache caches three sorts of data e data from the datastore e aggregated data comp
101. an giving me more work my work would have been more valuable for the following students Among the interviewed participants team members who leave was not felt as an issue that had affected them Each of the participants was asked about what they would desire when reusing code Requirements are not expected All participants mentioned the need for high level documentation a clear structure i e the software components and understandable n u u steps also referred to as engineering process story storyboard or flow Several participants stressed the need to see how the high level documentation maps to the code In the source code comments are always expected Some attribute high value to performance and conciseness while others in particular novices insist on simple to understand and easy to read code Code from more experienced team members can be so compact it becomes difficult to understand The practical experience of the two academic participants who had to extend existing software showed two reuse traps it was difficult to understand the code because of its low quality and poor and sometimes inaccurate documentation and there were hidden assumptions and flaws under the hood 3 5 Discussion The results from both literature and interviews are discussed per focus question 3 5 1 Howis engineering automation software currently developed Overall the software is developed ite
102. an use the app engine free of charge for small applications The Google App Engine dashboard of a deployed application shows the status of all quota how much has been used and how much is remaining for that day The relevant quota for GenDL Designer are discussed here The most important quota are datastore quota At the moment per day GAE allows 50 000 read 50 000 write operations This seems a lot but you have to take into account that each attribute of an object accounts for one additional operation So reading a Diagram object from the datastore with a name and some content accounts for 3 read operations and more if it would also have a timestamp or lock information A consistency check which requires the full code and design model has for a large project with 100 files a quota usage of 100 files x 2 reads CodeFile 100 classes 500 children 500 child relations 500 child type relations 150 mixin relations x 5 reads ModelEntity 8700 reads Hence you would only be able to run 6 checks before running out of quota Clearly this will not work in combination with a 5 second auto refresh interval The solution is to store aggregated data temporarily in the memcache which is provided free of change Traffic volume bandwidth and storage volume are limited too but are in the order of gigabytes which is much more than what is currently in use The Google App Engine scales your application by starting up more instances whic
103. anagement and Software Engineering practises such as communities of practise or documenting software designs will benefit engineering automation developers but the practises will need to be adapted to fit with the profile way of working and Part Il Initial study 15 the incentive scheme of engineering automation developers who are professional end user developers not software engineers Part Il Initial study 16 3 Interviews 3 1 Introduction The problem of limited reuse in Engineering Automation found in literature and described in the previous chapter is further investigated with interviews in the field to improve the reference frame for solutions later on The main requirement for solutions is to match the profile way of working and incentive scheme of engineering automation developers The interviews will be used to validate the applicability of the literature review results to the organizations and to provide deeper understanding of the problem by looking for practical examples and additional explanation To obtain validation and understanding it is tried to answer the following four questions e How is Engineering Automation software currently developed e Why is Engineering Automation software developed the way it is e How is Engineering Automation software currently shared and reused e What would more reuse of Engineering Automation software actually require Interviews are a common practise in engineering design res
104. apt to new situations in future work and thus reuse is low In my thesis I need to find strategies to improve the reuse of Engineering Automation software Apart from finding strategies that can improve reuse it is also an exercise in finding the strategies which engineers are willing and able to adopt With this interview I would like to find out how Engineering Automation software is currently developed by engineers why it is developed that way how Engineering Automation software is currently reused and what more reuse would actually require This should result in a set of constraints and opportunities to base my strategies on Context questions 10 min e What is your position in the company e What is your professional history e What do you use Engineering Automation for e Do you develop individually or in a team e What training did you receive in software development formal and informal How do you look back to that Part Il Initial study Appendix A Current Software Development Practise General 30 min e Can you draw a timeline activity diagram of a typical project e How do you approach the requirements e How do you approach the design e How do you approach testing and validation o It is described in literature that testing in Engineering Automation is often approached differently than testing in regular Software Engineering due to its special nature To what extend do you recognize these The develo
105. ar enough and cannot be simplified add comments 3 Regardless of whether the code and comments are clear enough add a docstring to functions and methods unless their purpose and usage is completely clear from their name and arguments 4 When the general idea behind the code in a particular file cannot be sufficiently understood by studying that file alone describe the general idea in this code report Hence this report gives overview and focuses on larger aspects Although the adherence to the policy is not perfect Pm convinced I came pretty far Only rule number 3 was violated more than it should docstrings were sometimes omitted when their purpose and usage could be inferred from the context 3 2 Folder structure There are three top level folders one for each subsystem Each subsystem folder has a src folder a test folder and a vendor folder The src folder contains all code written for GenDL Designer that will run in production This is the most important folder The test folder contains the test code written for GenDL Designer The vendor folder contains external libraries used Only in very exceptional cases this code has to be modified or tested Therefore it is separated from the self written code Some other folders are sometimes present as well inspiration contains examples found on the internet playground contains code snippets for tying out ideas quickly These folders are not under version control 3 3 Dependencie
106. articipants 6 2 3 Material to review O24 Code review procedure site ses csetseciecead secede S R S A RA RAA AA AAA AT 6 3 CODING POLICY wieiasevsechesavcsereasesaveaetcsyeccyowenevapucwansayescucuequven chayvowsseayeceucnensreadousevouitecyaven vnetecsansavecrbeveueuenevayacwacvayeccucvepereicnagecaunedseces 56 6 3 1 The purpose of a coding policy sits i ste 623 2 Amtroduciiig ac COGIN PS POL Cys sviat csseseluesescasestideasascisedestesad phsulaegitscsseotsudndesasadedeshotdedeshoulateaieadedsoicadhasahshdel lt phsedeceeseecaunrasts 63 3 Language independent guidelines iniaa EAT EE ESE EES EE EEEE EE EEEREN 56 7 CONCEUSTON ia ss cocvsvcscastenstvosssecesevensssuaseinsssesuvontee Sotasssasesasdseseagssssevassengesesiabescisvepesgsisosaiecesesssossbvsnseiots 58 8 REFERENCES avis e ECEE E O Sovensissevecsbesevusalesesscsdedussaslasvestshe Susatctesevcas e TN 59 APPENDIX A INTERVIEW AGENDA AND QUESTIONS APPENDIX B AGGREGATED INTERVIEW ANSWERS APPENDIX C CODE REVIEW CHECKLIST Part Il Initial study 1 Introduction 1 1 Software development by engineers Increasingly engineers create software to support their daily engineering activities The goal of this project is to gain understanding in how the reuse of these automated engineering models and solutions Engineering Automation software can be improved there is no need to re invent the wheel over and over again With the ever increasing computational pow
107. ass dialog 9 11 Filter text field for the project tree Large projects contain many classes and functions Finding them in the project tree becomes easier with a filter text field on top or on the bottom of the project tree Part IIl Code report 69 9 12 Maximum size of projects Currently both the design model and the code model i e the JSON data of the design and the JSON data from the parsed code files are each stored in 1 memcache entry They are in the memcache because in the datastore the data is spread out over multiple entities and all these entities would have to be retrieved to construct the models That consumes too many resources 1 memcache entry is limited to 1mb For large projects with say 50 large classes this can become a problem the design model and code model can be bigger than 1mb Requests that involve retrieving or constructing the design or code model will then simply result in an error To work around this the models would have to be split in multiple memcache entries and the list of all memcache entries to retrieve would have to be stored in a master memcache entry or perhaps a database entity As a sensible safety guard projects are currently limited to 100 files 9 13 Store timestamps in the datastore instead of memcache The last changed timestamps for both the design and the code are stored in the memcache not the database to avoid some database traffic on every change This causes the unknown
108. ause this in the function refers to the object on which the function was called in this case window And window object_name is undefined window global_shortcut_to_my_function This creates major problems when passing around functions as arguments The function will behave differently when called in different places and that is weird Therefore the keyword this is avoided where possible Instead the following pattern is used illustrated by a counter function create_counter initial_count var _public var count initial_count undefined initial_count _public increment function count 1 _public read function return count return _public This pattern also has the advantage that private data is possible everything which is defined inside the function and which not accessible through the returned object _public is private there is no way for code outside the create_counter function to get access to count except by calling the read function Part IIl Code report 54 7 2 2 mxGraph mxGraph is a browser based graph drawing library with extensive options It provides features aimed at applications that display interactive diagrams With the right configuration you can use it to create practically any 2D drawing application in a browser The best starting place for mxGraph is the tutorial and the user manual Both can be found on http jgraph github io mxgraph
109. bout graphical programming notations and code synchronization is reviewed next Graphical programming notations ate needed for the graphical design whereas code synchronization is needed for iterative code generation This chapter is concluded with a review of state of the art software design modelling tools some critical notes about software design tools in general and a summary of recommendations for the prototype of the software design tool to be implemented Part Il Initial study 33 5 2 Graphical programming notations In software design graphical notations are a frequently used alternative to textual notations They tend to be more appealing and provide an overview that is easier to grasp Diagrams can be used to quickly capture and share thoughts acting as a reflection and communication tool or they can serve as formal or informal documentation Goodliffe 2007 Well known examples of graphical modelling languages within engineering are LabVIEW and Simulink marketed by National Instruments and MathWorks respectively Within Software Engineering the Unified Modelling Language UML is currently the most popular and well specified graphical language Goodliffe 2007 This section discusses graphical programming notations on several levels Starting from the most fundamental they are discussed from a cognitive perspective with the Cognitive Dimensions of Notations framework followed by their main limitations More practically the us
110. c and the behavioural school is included An engineering app repository is a combination of a knowledge repository a concept from the technocratic school and an application repository such as the app stores for mobile devices The communities of practise concept is a solution from the behavioural school which distinguishes 3 levels The two lower levels engaging in communities and communities refining practise correspond to code reviews and coding policies respectively and are positioned in the software quality area The third and highest level interconnected communities in organizations is positioned in the software sharing area since these communities are more likely to exchange working software rather than software quality improvement remarks exchanged information tends to be larger and more finished as the distance between knowledge workers increases A first selection of the solution concepts is based on the two key issues identified in chapter 3 understandability and validation These are pre conditions for sharing the software needs to have acceptably high quality Therefore the pure software sharing concepts are discarded in favour of solution concepts that can improve the quality 4 concepts remain e Graphical Design Tool e Engineering App Repository with Quality Metrics e Coding Policy Code Review In this section these solution concepts are reviewed with respect to the selection criteria 4 2 1 Graphical software
111. c nature of consistency checking rules and inconsistency resolution raises the need for engineers to create and extend the plugins themselves A promising approach is to let engineers provide patterns of what is considered inconsistent The semantic web 41 provides a uniform way of representing data RDF 42 and provides a pattern based matching mechanism SPARQL 43 that can be used find inconsistencies based on patterns The data that is matched by the inconsistency pattern can be fed into a related template or form also created by an engineer to generate advice or precise instructions on how to resolve the problem Part Thesis article 35 Part Thesis article 36 References 1 2 3 4 5 6 7 8 9 10 11 12 13 W J C Verhagen P Bermell Garcia R E C van Dijk and R Curran A critical review of Knowledge Based Engineering An identification of research challenges Adv Eng Informatics vol 26 no 1 pp 5 15 2012 G La Rocca Knowledge based engineering Between Al and CAD Review of a language based technology to support engineering design Adv Eng Informatics vol 26 no 2 pp 159 179 2012 G Leshed E Haber T Matthews and T Lau CoScripter automating amp sharing how to knowledge in the enterprise in Proceedings of the twenty sixth annual SIGCHI conference on Human factors in computing systems 2008 pp 1719 1728
112. ce AI In AI knowledge based systems are systems which have explicit knowledge encoded in them Applying such a system to engineering in particular CAD yields a knowledge based engineering system The discipline that develops these systems became known as Knowledge Based Engineering even though the distinguishing qualifier knowledge based in fact applies to the developed systems not the discipline itself Verhagen et al 2012 La Rocca 2011 Part Il Initial study 9 budget constraints which is often the case in industry but comes at a price van der Velden et al 2012 argues that Design Automation solutions fall short on one or more of the following system qualities for KBE systems e Reusable can be used in several business processes e Generic applicable to a range of different problems e Generative preserves the design process rather than the generated product model thereby allowing for design changes e Integrated interfaces with other software preferably through standardised formats e Detailed implements a high level of knowledge e High level a high level of abstraction is available to express knowledge Knowledge Based Engineering has not found widespread adoption yet La Rocca 2011 To improve the maturity and to support the adoption of Knowledge Based Engineering several methodologies have been developed which support the full lifecycle of KBE systems including knowledge acquisition and knowledge modelling
113. cking an element in a diagram or through the consistency tab The consistency diagram provides the resolution option code snippet when an element was found in the design and not in the code In both cases a dialog is shown with the code snippet The code snippet can be copy pasted into the user his code Firefox 7 7 Gendi Designer Demo Pieter Jan e gendidesign appspot com projects Code snippet for wing in package demo pieter jan define object wing documentation sauthor lt name gt lt username gt fkorganization gt com description input slots fuel tank volume nil todo default value computed slots robjects flap type high lift device For classes or functions the code snippet can also be downloaded and saved as a file with the button that appears in the top right corner The file will already have the correct name eLlement_name lisp which encourages clearly naming code files 2 6 Lookup code GenDL Design can also trace design elements to code by looking up the file and line that corresponds to a diagram element This feature is found in the right click menu of elements in diagrams The file will be opened in the browser define object robot base object input slots torso type box settable head angle body angle arm angle right arm angle left pincer distance right Ppincer distance left shininess transparency computed slots
114. clusions GenDL Designer seems to encourage the creation of accurate design documentation and seems to encourage designing before implementing The principle of incremental code and design documentation generation appears to have the potential to improve the understandability of applications the validity of their documentation and even the validity of the code itself due to the improved transparency that uncovers defects Finally introducing incremental code and design documentation generation in an engineering automation context appears to be feasible but some potential users will not be convinced with a short introduction alone These promising but preliminary findings will hopefully be confirmed with the large scale academic experiment and later on with experiments in industry vi PART THESIS ARTICLE Incremental Code and Design Documentation Generation for Engineering Automation P J A R Dewitte B Sc February 25 2014 Contents PREFACE nnie ea a a a a Huda aa a a a ieee a a aa ar ht EEK ACKNOWLEDGEMENTS wiccccecssecstecteeccceusescunacsecaecestiuelst i iarsi aea e a aA See Eoia aaRS II SUMMARY eserini ARES ANE RAE AEEA ARNT E R eon detest en tee adeeb eerste V Tt INTRODUCTION Sisco eeart anesan oeira e e E aaa Ea AE E OaE O AA N RA E OnE EUAS 1 2 DEFINITION OF ENGINEERING AUTOMATION n sssssssssissssrrssssrrssrrrrssrerrsrrrrssreressrrerssrreesse 3 3 ENGINEERING AUTOMATION PRACTICE eeeessececeeseeeceesaeeeceeae
115. ctiveness of GenDL Designer both when experimenting with the application setup and during refactoring 9 8 Dynamic class type notation The type of a child of a GenDL object can be dynamic it can be decided upon at run time The design notation currently allows multiple has type relations for a child but the consistency tab will complain about this If this bothers users support for dynamic types could be implemented It requires a notation for it and improved parsing and consistency checking logic 9 9 Automatic layouting and mass import When diagrams grow they can become cluttered and some effort is needed to re organise and clarify the diagram This could be done partially automatic with layout algorithms Automatic layouting is especially useful when reverse engineering existing code Currently design data can be imported into the project but the diagrams themselves still have to be created by dragging elements into the diagrams The biggest challenge for mass import is what to place in which diagram each diagram should not be too extensive but not too minimal either 9 10 Assign colours to classes Classes can have different roles some are product classes others are capability modules etc The expressivity of diagrams could be increased by letting the user choose a background colour for each class This can easily be implemented by adjusting the code that renders classes on the canvas and adding a colour picker to the cl
116. d a formal model The informal model provides 5 standard forms for describing a variety of concepts Entities such as structures and functions Activities of the design process Constraints i e limitations Rules applied during Activities and finally Illustrations i e background information These forms are referred to as ICARE forms The formal model is divided in two pillars a product model and a design process model The product model describes a family of products It expresses the product in terms of structure function behaviour technology such as materials or manufacturing techniques and representation such as a geometrical representation or a discrete FEM model The design process model on the other hand describes how the product model is instantiated The model elements in the formal model are usually linked to elements in the informal model Entities and Constraints typically become part of the product model while Activities and Rules typically become part of the design process model The formal model can be expressed in the MOKA Modelling Language MML This notation is based on the UML modelling language which has become the de facto standard graphical notation used within Software Engineering UML is more extensively discussed later in this report Software support To support structuring knowledge according to the guidelines of MOKA a software tool was created In addition the MOKA project produced conceptual plans f
117. d code base 14 used to develop in a team Cooperation is limited to sharing old code Part Il Initial Study Appendix B What training did you receive in software development formal and informal How do you look back to that All participants except 14 had introduction course s which mainly concerned language basics syntax loops conditions etc Looking back these courses were too simple and not very helpful for the real work All participants are self educated in particular programming language s while on the job by using the internet asking colleagues and or reading books o 13 asking more senior colleagues is a bit intimidating But we definitely try to help each other o 13 I tried to go through Python tutorials but those were so boring and I struggled see the application the bigger picture is missing I prefer to search on the internet for a particular piece of code or a particular problem You ll always find a good answer Similarly A2 prefers googeling compared to the built in help o I3 There is no formal training program We wouldn t even talk about that It s almost assumed in the team that people can program They don t expect you to do a brilliant job but they expect you to do a good enough job There will always be an underlying assumption that you can code o 13 Taking snippets from existing code in the company was a pretty good way of learning you could see what the input was you
118. d reuse reported by the participants matches the limited level described in literature Small sharing networks do exist informally within teams These networks are used to share pieces of legacy code and explain them upon request not for close collaboration Being asked for explanation is preferred over writing documentation as it saves time These networks show the large importance of the social aspect in reuse both discovering existing software and reusing software are now closely linked to internal team interaction 3 5 4 What would more reuse of engineering automation software require There is a discrepancy between what is currently done by engineering automation developers and what they desire when they have to reuse code This discrepancy is formed along two lines understandability and validity The discrepancy is shown in Table 1 Low code quality and poor documentation make software hard to understand The validity on the other hand is undermined by the lack of systematic testing which makes it hard to guarantee the correctness of the code and the consistency of the documentation with the code What is needed on the other hand is simple to understand code with adequate comments the steps and structure in the code clear and documented tests to ensure the correctness of the code and finally documentation that matches the code Part Thesis article 11 Table 1 Difference between what is currently done by engineering automation de
119. d what I m about to do It was never taught how to do that for programming while we did learn how to do this for say mechanics Documentation is limited to source code comments written informally for the author himself Some feel that external documentation won t be read Instead people ask questions directly which saves time If there is external documentation such as a thesis or paper traceability is limited to indicating the origin of equations in the source code If version control is used it is only used as a way to transmit code 13 I ve never been formally introduced to version control If I write code within a day I find it pointless to use version control I can remember what changes I made Software testing The software is tested while it is developed by running the complete program and manually verifying the output There is a prior expectation for the output but not a crisp value Automatic testing is not used Testing parts of the code individually is only done during debugging Multiple test cases are used but there is no explicit test plan of what has to be tested At the same time some participants mention problems for which the standard solution in Software Engineering is systematic and automatic testing A2 The problem is that when you make changes in one file and finally get it working the program will crash in another file So you keep fixing files for a week until you can finally commit But st
120. delete methods e An instance of the handler class is created e The appropriate method is called with as arguments the matched groups in the regular expression e g ProjectsHandler get 1234 e The handler method can now process the request Typically it reads from self request reads and writes from to the database and writes to self response but anything is possible A Webapp2 application is nothing else than a group of routes and their linked handlers GAE app configuration app yaml The starting point of a Google App Engine app which is more than just the Webapp2 app is app yaml It is a configuration file that declares what kind of GAE app the folder contains and which URLs are handled how URLs belonging to static files must be routed to the physical files URLs belonging to dynamic web pages must be routed to an app from a Python module which can handle those requests A clarifying example of the folder structure for a simple GAE app app yaml Marks folder as GAE app directs static urls to static_files and directs dynamic urls to my_app app static_files Contains js css etc files script js style css my_app py Creates an app e g Webapp2 app and assigns it to the variable app More information about app yaml configuration options can be found on https developers google com appengine docs python config appconfi GAE Datastore The GAE datastore is the database service available in
121. derstand Validation time ability consuming Graphical 4 0 software 3 1 Ok Ok Ok Ok 4 design tool x X 6 E 8 Pnpnicenag 3 3 Not ok Not ok Ok Ok app store 2 10 8 Poe 5 5 Not ok Not ok Ok Ok review 2 i 5 4 oe 3 2 Ok Notok Ok Ok policy 1 Table 4 1 Trade off table for the selection of a solution concept The highest score is awarded to the graphical design tool concept This concept is selected mainly because it fits best with the current way of working of engineering automation developers This concept will be reviewed extensively The other concepts engineering app store code review and code policies will only be reviewed briefly for future reference in order not to waste the information obtained already Part Il Initial study 32 5 Literature review on graphical software design tools 5 1 Introduction In chapter 4 the graphical design tool concept was selected out of four solution concepts A graphical software design tool will be implemented for GenDL a KBE system so that KBE developers can create the design of a GenDL application quickly and generate code from that design The created designs must provide an overview of GenDL applications The tool must account for users with limited Software Engineering training and for an iterattve development style where both the design and the code are regularly revised For recommendations and best practises to build a graphical design tool literature a
122. describing the modules that use these technologies 6 2 1 Parsing PLY PLY is the library used to implement plygend1 one the two main components of the GAE app as shown in the figure on the previous page PLY is a Python implementation of the Lex Yacc parsing tools With PLY you can build a parser from a grammar description of the language you would like to parse This reduces the effort needed to build an efficient parser compared to writing a parser yourself from scratch A basic introduction to PLY is given below mainly so you can understand existing PLY grammar descriptions Before writing or altering one you probably want to follow some tutorials on the PLY website first to same yourself some time http www dabeaz com ply General two stage parsing Lex Yacc follows a standard paradigm first a lexer breaks the input stream into tokens Then the actual parser applies grammar rules The tokes are described by regular expressions and are used to recognize e g numbers variable names operators etc The grammar rules used by Yacc are so called context free grammar rules These have the form symbol lt some expression with tokens and symbols There is one start symbol e g program When parsing the parser tries to recognize the pattern on the right and replace it with the pattern on the left After repeatedly doing so eventually the parser should end up with the start symbol To clarify a simple example
123. design tool 4 2 1 1 Introduction Based on Domain Specific Modelling a graphical software design tool supports working out the design of an application before the application itself is implemented The design is a plan a higher level description which omits details not needed to understand the global view The use of a software design is comparable to the use of a conceptual and preliminary design before starting with the detailed design in Aerospace Engineering The benefits of creating a design for software in particular are widely described e g in Goodliffe 2007 Blaha amp Rumbaugh 2005 it becomes easier to write the code easier to understand and improve the code and perhaps most importantly it helps to prevent the code from becoming an unmanageable chaos The motivation for this solution concept is formed by the benefits of domain specific modelling the possible improvement of understandability through reflection on and documentation and the possible reduction of implementation effort with code generation which reduces the coding effort Each of these is discussed next 4 2 1 2 Domain Specific Modelling Domain Specific Modelling strives for simplicity by turning software development into modelling in the problem domain rather than in the software solution domain This lowers the barrier allowing problem domain experts to take part in or even take over the development In the case of Engineering Automation the modelling language
124. design tool GenDL Designer can also be used as a learning and tool as a reverse engineering tool Novice GenDL programmers can draw designs which is relatively easy to pick up and learn how it maps to code by requesting a code snippet for an element in the design Also the Part Thesis article 18 consistency check helps them to spot where the software is different from what they intended GenDL Designer can be used to reverse engineer existing code by comparing the code to empty diagrams GenDL Designer will propose to resolve missing design documentation by importing it from the code All the user has to do is drag elements from the project tree into diagrams Part Thesis article 19 Part Thesis article 20 5 Validation experiments 5 1 Introduction GenDL Designer needs to be deployed to GenDL developers to verify and validate it as a solution On the level of verification the question is whether the solution is working as it should does incremental code and design documentation generation as implemented in GenDL Designer encourage engineers 1 to document applications correctly and completely and 2 to create a design before writing the corresponding code Metrics will be proposed to address this quantitatively On the level of validation it must be determined whether incremental code and design documentation generation is an appropriate solution to the problem it is supposed to tackle does it promote shar
125. dlers projects ProjectHandler projects d _summary json serverapp handlers projects ProjectSum projects d design infinity xml serverapp handlers projects desi projects d design integrity html serverapp handlers projects de projects d html serverapp handlers projects ProjectHandler Tests The transformer is tested with unit tests The routes tree itself is more a configuration file than program logic and is therefore not tested individually Instead the routing is tested along with the handlets as described later on 6 3 4 server_app datamodel Introduction The datamodel package has two functions First it defines the structure of the data i e the data model itself This is the data model presented in section 4 2 The UML diagram is repeated below for your convenience Second it provides an abstraction layer over the datastore and memcache services The rest of the application does not have to deal with the difference between the memcache and the datastore the datamodel package provides a unified interface which offers the required functionality Under the hood the datamodel package decides when and how to use the datastore and or memcache Part Ill Code report 43 name string package string visible bool True when_was_last_design_change Date when_was_last_code_change Date when_was_last_change Date usem
126. e generation and minimal effort This is highly motivating and reduces the overhead of sharing Markus 2001 Nevertheless both literature and interviews showed that it is time consuming to prepare software for sharing clarifying generalizing etc and maintain software that is shared bugfixes new requirements feature requests To warrant participation the overhead must be balanced by proper incentives Howison amp Herbsleb 2011 Leshed et al 2008 4 2 2 4 Conclusion An engineering app repository with quality metrics can contribute to the level of reuse by improving the visibility of Engineering Automation software for reuse and by making the quality explicit For knowledge consumers software and or the underlying knowledge become easier to find understand or validate and the software quality becomes easier to judge For knowledge providers visibility can enable formal or informal reward systems and trigger quality improvement through peer pressure Unfortunately this concept is likely run into some of the non technological obstacles There is no reason why an engineering app store would make improving and sharing software less time consuming The incentive that could compensate the time spent on improving and sharing softwate visibility leading to reputation and or other benefits is currently not present Therefore an organizational change in reward model is required On a positive note the solution concept can account for user
127. e Engineering community need to be adjusted to the different environment background and domain knowledge of the scientific computing community or more general professional end user developers general software engineering solutions only work for general software engineers Part Il Initial study 28 Stokes 2001 goes even further and proposes not only a domain specific but even a target platform specific editor for engineering knowledge MOKA The generated code can then be targeted to the specific platform and code style expected by the engineering automation developer 4 2 2 4 Conclusion A graphical software design tool and associated way of working can contribute to the level of reuse by improving the quality of software through increased understandability of the code It encourages a well thought application structure and provides the documentation of that structure It also improves the level of validation to some extend by checking for conflicts between the design documentation and the code This concept could handle all 4 non technological obstacles The concept makes designing software less time consuming than with general purpose drawing tools as well as easier to justify because with code generation the time spent on design results in tangible results The domain specific modelling approach which is in fact drawing keeps the entry level for creating a design low so that it isn t difficult to convince developers without a Soft
128. e capable of building complex large and working systems while usually from a Software Engineering perspective their approach is flawed Segal 2007 2 3 4 Design Automation and Knowledge Based Engineering The literature on the application of Professional End User Development to engineering Le Engineering Automation is fairly limited Most contributions are in the area of Design Automation and Knowledge Based Engineering This is also the area of Engineering Automation in which the investigated organizations are active Design Automation and Knowledge Based Engineering KBE are technologies that provide the capability to rapidly design and produce a large number of product variants Their objective is to reduce the time and cost of product development by automating repetitive non creative tasks and where applicable support the integration of multiple disciplines by retrieving and combining analysis results from several disciplines much faster than is possible with a manual approach Design Automation is used to refer to the development of design related Engineering Automation solutions without the overhead of a comprehensive modelling activity but also without the benefits of such an activity Knowledge Based Engineering is at the other end of the spectrum stressing the need for a detailed study of the domain Verhagen et al 2012 van der Velden et al 2012 Actual projects are situated somewhere in this spectrum Knowledge Based Engin
129. e control structures such as for loops are also closely related to the equivalent Python control structures All HTML web pages outputted by the server app are generated with Jinja2 templates Jinja2 is available in the Google App Engine standard library Jqote2 Jqote is a template language implemented in Javascript The expressions you use are in fact plain javascript Jqote is used to generate parts of the web interface dynamically e g the class dialog or the settings page The code can be found on Github https github com aefxx jOQote2 Full documentation can be found on http aefxx com api jqote2 reference Django templating system The Django templating system is part of Django a web framework implemented in Python The templating system of version 0 9 was ported to the Dojo Javascript framework Summary In the end three template languages are used e Jinja2 for Python is used on the server for generating HTML e Jqote2 for Javascript is used in the client for generating HTML e Django for Python is used on the server for generating code for the bootstrap package e Django for Javascript is used in the client for generating code to show to the user in a dialog Unfortunately the quality of the template engine implementation is low Part IIl Code report 40 6 3 Components 6 3 1 plygendl Introduction The plygendl package parses GenDL code and returns it as data in the format described in section 4 2 Unified Design
130. e findings from literature were confirmed by the interviews the participants were not aware of software engineering practices or they hardly used them correctly The requirements are implicit from project goals The design is not planned to a high degree One participant said There is no prior plan for the code just start doing it and see what happens see what falls out At most some thought is given to the top level data flow what goes in what must come out and vaguely the steps to make that happen The software is tested while it is developed by running the complete program and manually verifying the output There is a prior expectation for the output but not a crisp value Automatic testing is not used Testing parts of the code individually is only done during debugging Multiple test cases are used but there is no explicit test plan of what has to be tested During the interviews participants mentioned problems which could have been avoided with more testing one participant mentioned that after making a particular change he was fixing bug after bug for one week until finally the program ran again Another participant told that only after weeks of reverse engineering he could find out that the code of his predecessor was not entirely correct Documentation is limited to source code comments written informally for the author himself Traceability is limited to mentioning the origin of equations The development method f
131. e interactive with jQuery event binding add remove elements and with jQueryUI sortable order of elements Standard jQuery functionality is used to read out the data again when the dialog is closed 7 3 8 Consistency tab module This module displays consistency notifications in the consistency tab and refreshes them periodically The refresh mechanism works as follows It is started when viewing the consistency tab and stopped when another tab is viewed When running it requests the list of inconsistencies every X seconds It sends to the server also the timestamp of the last list If no changes were made since then the server will respond with 304 NOT MODIFIED In that case the list is still accurate and not updated After Y minutes of inactivity the refresh mechanism is stopped the list is blocked resolution links can no longer be clicked and a message is shown on top When clicking a link or anywhere in the list in fact the list is blocked for Z seconds to avoid flickering of the list This will happen because with clicking issues are resolved The user then expects the issue to be gone but due to the time it takes to process the change the list might get refreshed earlier and show the notification again while it actually is getting resolved already Rendering is done with dojo s implementation of Django for historic reasons Jqote2 could have been used as well Resolution links are included in what Django ge
132. e of multiple views and levels of abstraction is discussed Finally the most common application in software UML is presented 5 2 1 Cognitive Dimensions of Notations The Cognitive Dimensions of Notations CD framework of Blackwell amp Green 2003 provides a vocabulary of system properties to discuss how well user activities are supported by the structure of a Human Computer Interface HCI This vocabulary serves as a discussion tool similar to the established vocabularies in other more mature design communities The CD vocabulary has a wide range of application from classic notations such as programming languages to interactive devices Analysing a notation with the CD framework starts by identifying representative user actions An action has an ideal profile in terms of the CD vocabulary some dimensions properties will be highly desirable while others are not important for the given task With the CD vocabulary it can then be described in what ways a notation does or does not support an activity The most important dimensions are given below Viscosity resistance to change Visibility ability to view components easily Premature commitment constraints on the order of doing things Hidden dependencies important links between entities are not visible Role expressiveness the purpose of an entity is readily inferred Error proneness the notation invites mistakes and the system gives little protection Abstraction types and avai
133. e policy should be based on current best practise in the team The team should be involved in drafting the policy by giving feedback even though it won t be possible to satisfy everyone The document should be very accessible and direct For the more contentious decisions some justification should be included Also helpful is to construct and introduce the standard piece by piece 6 3 3 Language independent guidelines Code presentation and language use are dependent on the syntax and semantics of the programming language used Because of this they are not discussed here see instead McConnell 2004 instead Variable naming and commenting guidelines are mostly language independent and some general advice is given here Favour clarity over brevity Fully and accurately describe what you re naming Choose names for easy reading not for easy writing In only very few cases abbreviations or short names Part Il Initial study 56 are acceptable An example is a loop counter when they are easy to recognise as such e g by consistently using name 1 their brevity actually improves the clarity of the loop code Understand what you re naming If you can t find a good name you probably don t really know what you re trying to do with it and what you re trying to name should perhaps even not exist Make sure you can name things well right away Every name should make perfect sense in its context Use a convention for predictable
134. e prescribed by the MOKA methodology Bermell Garcia 2007 2 3 6 Adoption of Software Engineering practices Being professional end user developers rather than software engineers it is interesting to investigate to what extent Software Engineering practises are adopted by Engineering Automation developers Based on experiences of her own and of colleagues Kelly 2007 states that the average academic Engineering Automation researcher is far removed from the Software Engineering world Simple practises commonly used by software engineers such as using a debugger are not adopted while the software development practises that are used are error Part Il Initial study 12 prone In one example an attempt to introduce Software Engineering to students failed because students felt like it wasn t for them The discussion here goes into detail about the adoption of the practises as discussed in section 2 2 Software Engineering Adoption software requirements design and testing 2 2 2 Requirements in research projects tend to be highly dynamic due to the exploratory nature of reseatch and design This makes requirement elicitation difficult It should be noted that the lack of requirements is in some cases partially compensated for by the permanent availability of the client i e the developer himself Sletholt et al 2012 On the other hand La Rocca 2011 states that the engineering design process to be automated should be well underst
135. e this form to help you perform a code review About the code Module name Version reviewed Code author Reviewed by Date Language Number of files Automated inspection O The code compiles without errors O The code compiles without warnings O There are unit tests O They are sufficient include all boundary cases etc O The code passes them Design O The code is complete against its specification O There is a good choice of algorithms O Optimizations are necessary and appropriate O Any missing functionality is marked clearly in the code General code comments Style O The code layout is clear O It follows project style guidelines O There is a good unambiguous public API O There is a good choice of names Defensive programming O Array access is guarded and safe C C O There is a correct choice of types O All input is validated O There is no use of compiler specific features General comments O The code is kept under source control O The code has been tested with inspection tools Tool name Results O Continue to next section Stop review here General observations about the code s design The code is well structured O There is design documentation O The code matches the documentation O Continue to next section C Stop review here General comments about the quality of the written code Error handling O
136. earch to discover requirements for future support systems Bermell Garcia 2007 It is not the intention to arrive at results that are valid for the entire Engineering Automation community This would require a far larger amount interviews than can be conducted within the scope of this project In fact only two independent organizations will be covered with just a few types of application domains At most these will support the findings already found literature rather than being generalizable themselves This chapter starts by describing the interview setup The interview results are provided as summarized answers to the questions above Finally a discussion with conclusions is presented about the interview results in relation to the earlier findings in literature 3 2 Interview set up In total 6 interviews were held with professional end user developers who develop software to satisfy a professional goal rather than to deliver software Four interviews were conducted with engineers in industry EADS Innovation Works and 2 with engineering students TU Delft Aerospace Engineering Flight Power and Propulsion The participants from industry were selected such that there was a large diversity in work experience from a couple of months to more than ten years and in usage of Engineering Automation The participants with an academic background were selected because of their experience with a project involving reuse and collaboration of Engineering A
137. ected concepts Besides the selected solution concept three more concepts were investigated as part of the selection process In order not to waste the information already obtained they will be reviewed in this chapter for future reference but not as in depth as the selected concept 6 1 Engineering app store with quality metrics The engineering app repository intends to be a central virtual portal where automated engineering solutions apps can be found and evaluated An app includes both the executable process and documentation to understand the process Like any knowledge repository the purpose of the engineering app repository is to lift knowledge out of the minds of individuals and make it accessible throughout the organization or perhaps even beyond Among the means to evaluate apps are software quality metrics ratings and indicators developed within Software Engineering that point out possible flaws App developers might pro actively react to the feedback provided by the metrics because they know the metrics are visible to colleagues 6 1 1 Theoretical foundations of knowledge reuse Knowledge reuse In Markus 2001 the first steps towards a theory on knowledge reuse are taken Central in the discussion are the different types of knowledge reusers and their different needs during the knowledge reuse process the role of intermediaries and the need for participation incentives to balance the participation costs Different t
138. ed by all in some way or another is the need for a clear structure i e break the program down in clear modular steps All but 14 also desire a clear and understandable flow story storyboard etc The commonality between the rest of the answers is very low Only 14 mentions that components should have well defined responsibilities O I2 In summary design documentation examples source code with good comments is everything you need I4 collaboration introduces communication and integration overhead I4 He was interested in how I solved the problem at the design level which modules and how they interact This was transferred in an informal way as a slideshow This discussed the input the business requirements algorithms steps of the process etc 14 Making the engineering process clear is exactly what we had in mind with a recent research tool project We choose to implement the tool as spreadsheets with scripting The sheets in the workbook correspond to the steps It is fairly linear in terms of its sheet layout so that you can go through each of the steps It starts in the first sheet with the input then the next sheet is the first process thing the next sheet the second process thing etc There is a trade off between quality and time resources required With that in mind what is the required level of the following items if you have to work with other s their code O O R
139. ed engineering codes as entreprise knowledge resources Cranfield University Bicar V amp Dogru A H 2011 Modern Software Engineering Concepts and Practices Advanced Approaches IGI Global Bj rnson F O amp Dingsoyr T 2008 Knowledge management in software engineering A systematic review of studied concepts findings and research methods used Information and Software Technology 50 11 pp 1055 1068 Blackwell A amp Green T 2003 Notational systems the cognitive dimensions of notations framework HCI Models Theories and Frameworks December pp 1 21 Blackwell A F 2006 Psychological issues in end user programming In End user development Springer pp 9 30 Blaha M amp Rumbaugh J 2005 Object oriented modeling and design with UML Pearson Education Bork M et al 2008 Towards roundtrip engineering a template based reverse engineering approach Model Driven Architecture Brooks F P 1987 No Silver Bullet Essence and Accidents of Software Engineering IEEE computer 20 4 pp 10 19 Cockburn A amp Highsmith J 2001 Agile software development the people factor Computer IEEE 34 November pp 131 133 Dick J 2005 Design traceability Software IEEE 22 6 pp 14 16 Van Dijk R E C 2013 Oral discussion about code synchronisation Elgh F 2008 Supporting management and maintenance of manufacturing knowledge in design automation systems Advanced Engineering Inf
140. ed in both the problem domain the engineering discipline and the solution domain the software During the development of Engineering Automation software the understanding of the requirements and the solution gradually increases along with the construction of the solution itself Software Engineering is reviewed because it studies software development It provides a framework to describe and classify current Engineering Automation development practises and is expected to suggest improvement to those practises This chapter further contains a review of Engineering Automation itself and a concluding discussion 2 1 Knowledge Management Bjornson amp Dingsoyr 2008 performed a systematic literature review on Knowledge Management in the context of Software Engineering Rather than duplicating their research their main findings are summarized in this section and compared to other academic contributions Software Engineering is indeed a knowledge intensive activity The need for managing knowledge has long been recognized and much can be learned from the knowledge management community which bases its theories on well established disciplines such as cognitive science ergonomics and management 2 1 1 Definitions of knowledge and Knowledge Management Knowledge is information which can be acted upon and in turn information is data put in context Data simply exists and has no significance beyond its existence Data turns into informati
141. ed invisibly in the higher level model Part Il Initial study 46 5 5 Critical notes about software design tools A graphical software design tool is not the ultimate answer for all software problems of engineering automation developers some critical notes about design tools can be found Firstly creating a software design is part of a larger workflow and this workflow must be considered when introducing a design tool Besides programming professional end user developers have to perform almost all tasks software engineers have to perform although they might not do them formally maintenance revision control testing debugging etc If the process has a serious bottleneck in one of these activities the benefits of the design tool are overshadowed by the issues in these activities and the whole approach will still be considered hard and troublesome Blackwell 2006 Secondly when introducing a design tool the usefulness must be clear to the developers who are supposed to adopt the design tool A design tool should therefore only be used where it is really helpful Goodliffe 2007 Part Il Initial study 47 5 6 Conclusions and recommendations Graphical programming notations and code synchronisation were reviewed both in literature and in state of the art software modelling tools to obtain recommendations for the prototype of the graphical design tool to be built The recommendations are summarised here together with pointers for subs
142. edge Management approaches All papers reviewed by Bj rnson amp Dingseyr are classified into two schools a technocratic school and a behavioural school The technocratic school is focused on capturing and formalizing knowledge and has developed technologies techniques and processes to do so The formal knowledge is made available with e g knowledge repositories knowledge maps and knowledge flows The behavioural school is more focussed on knowledge on an organizational and social level This school emphasizes the importance of social interaction in knowledge sharing The behavioural school is concerned with knowledge sharing networks including companies and how to foster them The technocratic and behavioural schools represent the two main strategies in Knowledge Management codification store knowledge itself and personalization store information about knowledge sources e g experts Put simply this is a debate between two solutions formal repositories versus communities of practise Although it is clear that one doesn t exclude the other this distinction is common in the literature see e g Markus 2001 or Segal 2007 The main critique on Knowledge Management found by Bjornson amp Dingsoyr was that published research is biased towards optimism the codifiability of knowledge and the utility of IT are believed to be overemphasised 1 Explicit knowledge is knowledge that can be articulated stored and readily distributed whe
143. eecesaeeeeesaaeeeeesaaeeeeneaaaes 5 A GENDL DESIGNER aiee eraa ee EEA E ulin 15 5 VALIDATION EXPERIMENTS crniiin iaieineea aiia a a le Giedeceesias de eiaa iee aR 21 6 EXPERIMENT TRIAL RUN S nineteen ati aE ERE EEE ANE A Et 25 7 DISCUSSION oei aaa a a EEr T E es 29 O gt CONCLUSIONS e an ioaea eaa Tani EA oth ae ea a a a cs ea Ea a aea aa 31 9 RECOMMENDATIONS wise cecivescccevessscceessbeecevesspecevesetcevvesbteeevesseeuvs scceeveseaseeuvesccesenveseceeves ect 33 REFERENCES oe oi e r E EE ere ax tvs eu E TEE T toasters E A carnetess a e 37 Part Thesis article Part Thesis article 1 Introduction Increasingly engineers in for example Aerospace Engineering create software to support their daily engineering activities This self written software helps them to solve similar problems faster by automating parts of their engineering work The software implementation of engineering models and solutions by engineers themselves to automate their own engineering tasks is referred to as Engineering Automation A prime example of Engineering Automation is Knowledge Based Engineering a discipline that captures engineering knowledge and applies that knowledge as digital rules in reasoning systems 1 2 It is desirable to reuse software for similar applications and share the software with colleagues rather than re creating it over and over again developing engineering automation software requires time and resources Developin
144. een created that build a parser based on a grammar description Usually parsing is performed in two steps first a lexer breaks the input stream into tokens Then the actual parser applies the grammar tules to perform some processing directly or to build an AST for later transformation Wachsmuth 2011 The two most well known parser generators ate Lex Yacc and Flex Bison available in several languages Goodliffe 2007 5 3 5 Iteration and round tripping Software development is an iterative design process If used forward and reverse engineering are likely to be needed multiple times Designers continuously question earlier decisions and therefore switch between levels of abstraction Ye 2006 Developers thus should be able to apply modifications to the artefact in the abstraction layer they find most suitable modifications should be performed where this is the most simple and efficient which is some cases means that manually writing code is preferred over modelling or vice versa Angyal et al 2008 Unfortunately most methodologies disallow changes to the higher level model once generated code has been changed Sitiol amp Lee 1999 The challenge lies in propagating modifications to other models without invalidating those models or overwriting manual changes made to the models A synchronization feature is essential to avoid error prone manual change propagation Several mechanisms to support this behaviour were found in literature and
145. eering has several related definitions Historically these definitions involve the methods tools and even programming paradigms used within the discipline such as high level programming and CAD automation More recent definitions set Knowledge Based Engineering apart from more conventional automation approaches by stressing a focus on knowledge capture knowledge retention and knowledge reuse Verhagen et al 2012 Nowadays it is also commonly accepted in Knowledge Based Engineering that once knowledge is captured it first needs to be structured modelled and developed further before it is transferred into software Studer et al 1998 There is discussion however on whether it is best to complete the modelling phase before the implementation phase or that modelling and implementing should be alternated van den Berg et al 2011 In contrast to the focus of Knowledge Based Engineering on knowledge and modelling Design Automation is a mote pragmatic approach focussing on a specific need and developing a system to meet that need with tangible benefits in terms of reduced lead time and cost van der Velden et al 2012 This pragmatic approach is especially appealing under stringent time and 3 The term Knowledge Based Engineering might be confusing because it seems to imply that other engineering disciplines are not based on knowledge a statement which many would find offensive The name stems from its historical roots in Artificial Intelligen
146. engineer the code if you know software and have some comments as clues The participants from academia A1 A2 found besides the code also the thesis where the code belonged to To what extend is networking used to share and reuse Engineering Automation For the participant from industry 11 12 13 14 internal team interaction is very important to get help and get to know about reusable code In the case of Al and A2 sharing and reuse was limited to starting your thesis from an existing piece of software and occasionally merging with each others code 14 would find a repository to access the code of others helpful So thinks 13 too but mainly as a learning resource o 13 It would be good to have a central repository where high quality code is available but I think mostly as a learning resource Actual reuse will be difficult because it is unlikely that what you need is exactly available and because to know all the assumptions you need to go through lots of code or lots of documentation if the author had the time to write it Requirements for Engineering Automation Reuse Relevance Assessment Situation Imagine you hear during a coffee break about an application from a past project which might be relevant for your new project to learn from or even to develop further What will you check and try to find out about this application before you decide to actually invest a serious amount of time in understanding this application dee
147. eported to the FileStoreInterface like object given when creating the FileSystemMonitor FileSystemMonitor checks for changes by making a list of all files that match the given file mask reading their contents and comparing the hash of the content to the hash on the server The hashes on the server are downloaded when starting the monitor Because the hashes in the client and on the server have to be the same webhash is used previously discussed in section 6 3 8 The FileSystemMonitor also reports the folder it is monitoring root path when starting and that it is still running every 2 hours 8 3 4 Server communication codeclient communication The responsibilities of the communication module are very simple report certain events to the server The ServerInterface class has methods that can be called to send these events to the server as HTTP REST requests file_created file_changed file_deleted etc If a connection error occurs a StoreUnavailable exception is thrown so that the GUI can be notified of the problem 8 3 5 System tray icon codeclient systray This module allows quickly making a System Tray GUI in Python It is intended not to be specific to the code client The features are Show hide an icon Set the menu and menu actions for left and right click Set the tooltip to shown when hovering the icon Show a pop up notification balloon The code itself looks quite cryptic because the module uses the ba
148. equent research Notation The basis of the notation is preferably UML as it is the de facto standard at the moment Slight modifications will be necessary though To suit the target Engineering Automation developers as much as possible the notation should be simple to use and map design and code component to component To facilitate code generation the formality of the notation might have to be increased Finally to avoid death by detail it must be carefully chosen what information is really high level as opposed to lower level details that are better left out The interface should not be too restrictive Soft warnings instead of hard constraints allow the user to make modifications in the order as he thinks of them and without unnecessary detours It would benefit the overall system if the notation covers both static and dynamic aspects e g product and process models and connect the two Also including testing aspects would improve the overall system In terms of cognitive dimensions the final notation should have low viscosity low resistance to change reduced premature commitment few constraints on the order of doing things high visibility view components easily and high role expressiveness the purpose of an entity is readily inferred A trade off between these properties might be necessary Code synchronization The simplest synchronisation mechanism that is expected to work for the prototype is notification generation I
149. equired because without quality control a knowledge repository quickly becomes a chaos They are useful to abstract structure index sanitize and synthesise the knowledge They also can provide the different views on the same knowledge that are needed by the different reuse types Part Il Initial study 49 Cost and incentives One should strive to create recognition and valuation for producing and consuming reusable knowledge and to reduce the effort required to do so Mottvating factors for participating are documenting for yourself immediate benefits and minimal efforts reciprocity and a short distance between the producer and consumer Note that when documenting for yourself fitting for reuse will usually be required because of different quality requirements and it must also be captured how the knowledge must be applied Possible stumbling blocks are a lack of organizational norms to back up reuse activities trust issues i e fear for inappropriate knowledge application and a competitive environment Knowledge repositories McAfee 2006 discusses information technologies used by knowledge workers These technologies are positioned somewhere between two extremes channels and platforms Channels allow everyone to create and distribute but the audience is limited In this part of the spectrum there is e g email and instant messaging Platforms have a larger audience but usually less people feel the need or are even allowed to au
150. equirement docs Only A2 find it important rest doesn t expect it Design docs Part Il Initial Study Appendix B All find that it can be expected to some extend the structure the flow used theories 14 notes that this kind of documentation is useful for both reusers and the original other when he picks up his old code again Oo Tests All participants require example input and output for the whole program Tests for individual functions are not expected In fact only 14 and A2 see their benefit for evaluating functionality and correctness in depth Documentation C traceability and on what levels It is felt that besides design documentation mentioned earlier comments in the source code is sufficient I3 stresses that if external documentation is provided it should be concise and high level Code quality what is code quality to you It varies among participants what is quality code There seems to be consensus that quality code is clearly documented and well structured But while some stress performance I1 Al and A2 others insist on simple to understand and easy to read code 12 and 13 Some also take into account obviously the correctness of code 14 A2 and having a clear story flow 14 A1 e I1 Code from more experienced team members can be so compact it becomes difficult to understand Part Il Initial Study Appendix B Appendix C Code review checklist Code review CHECKLIST Us
151. equirements are discovered along the way as your insight in what the program should do is evolving and growing According Part Il Initial Study Appendix B to I1 that is a main reason why software work cannot be outsourced to pure software professionals 14 captures requirements formally and translates requirements from an engineering point of view into software requirements How do you approach the design All participants have a different approach Only I4 thoroughly considers the design I1 12 and I3 start with the top level what goes in what must come out and vaguely the steps to do that The rest of the design emerges A1 and A2 had to start from existing code and their design work was about understanding and improving the design 14 starts from the tasks and then determines the architecture i e the components of the system and their role o 13 There is no prior plan for the code I just start doing it and see what happens see what falls out Note for complex problems I would quickly sketch something to see the flow and connections between modules o A2 I found it difficult to write down beforehand what I m about to do It was never taught how to do that for programming while we did learn how to do this for say mechanics All the intermediate steps are small in size and the notation makes clear what you re doing How do you approach testing and validation There is a clear general awareness
152. equirements is somewhat excusable there is a project description and the developer is his own client The lack of systematic testing on the other hand is more severe Testing is done by manually and passively looking for signs of incorrect behaviour As a result code is fragile when introducing changes or even downright incorrect Examples found were a week of bug fixing after making a change in a single module and an obscure code base which didn t function as documented respectively Formal methodologies are discarded as unnecessary overhead Engineering automation developers are also unfamiliar basic tools such as version control systems and debuggers There is a discrepancy between what is currently done by Engineering Automation developers and what they desire when they have to reuse code This discrepancy is formed along two lines low understandability and low validity The discrepancy is shown in Table 3 2 Low code quality and poor documentation make software hard to understand The validity on the other hand is undermined by the lack of systematic testing which makes it hard to guarantee the correctness of the code and leaves flaws under the hood unrevealed What is desired however is a clearly documented structure of the code and flow through the code in addition to simple to understand code with adequate comments Some participants indicate a preference for high performance code and concise code While high performance is sometim
153. er more numbers can be crunched input and output can be processed faster data can be stored and retrieved swifter etc all with the ease of pressing a button For many commonly encountered engineering problems software is available as a teady to use tool Examples are aerodynamical analysis tools structural analysis tools statistical analysis tools and plotting tools However for more specific problems such as problems with unique constraints and problems involving phenomena which are not well understood yet it is less likely software is already available Lieberman et al 2006 Then it is up to the engineer or researcher to create the software and any underlying theoretical models 1 2 Definition of Engineering Automation Frequently engineers and researcher undertake the endeavour of software development themselves It is for this reason that engineering curricula nowadays include software development courses Their software development work is referred to as Engineering Automation and is defined here as the implementation of engineering models and solutions by engineers themselves to automate their own engineering tasks This software releases the engineer from repeatedly solving similar problems manually Examples of Engineering Automation are a production cost spreadsheet a heat transfer simulation in Simulink or LabView pre and post processing scripts for an aerodynamic analysis an experimental finite element method implemented in
154. er log data The server log data provides a timeline of user activity and inconsistency events An example of such a timeline is visualized in Figure 5 One can clearly see how the participant switched back and forth from the design environment dark grey to the code environment lighter grey and how he resolved inconsistencies in both directions Also clearly visible is how the inconsistency sometimes goes down during code sessions positive flow and sometimes goes up negative flow 6 3 Metrics The two metrics consistency at the end of the project and the average flow during the project are calculated from the timelines for each project For the 7 trial runs the results are plotted in Figure 6 In the plot it can be seen that users who started from scratch for the toy project or for a regular project obtained a high consistency and all but one also a positive flow direction The users that used GenDL Designer as a documentation tool turn out to have a high variation in consistency and obviously have a very negative flow direction Part Thesis article 25 Design modification Code to design resolution View consistency Design to code resolution Code modification Inconsistency 0 0 0 5 1 0 15 2 0 25 Duration h Figure 5 Example timeline of user activity during a trial run user 2 a a Toy project User 1 2 and 3 x x Documenting User 4 5 and 6 a Regular project User 7 Flow 1 1 0 0 0 2 0 4 0 6
155. es during the first run after the grammar was last modified This should be sufficient to make sense out of PLY grammar descriptions when you see them Before actually altering them remember it will probably save you time and perhaps frustration if you follow some tutorials on the PLY website 6 2 2 Web app Google App Engine The Google App Engine GABE is a cloud computing service with some free quota Basically you can develop a Python web server application locally and when finished deploy the application app to Google infrastructure Apps can use several services available in the Google App Engine GenDL Designer only uses 2 services the datastore and the memcache Part IIl Code report 36 The introduction given here is by no means sufficient to understand GAE fully as a developer It is only meant to give a quick introduction so that the existing code of GenDL Designer can be understood properly Excellent documentation and tutorials can be found on https developers google com appengine docs python GAE web server application webapp2 The default web server framework available in GAE is webapp2 This is also the framework used by GenDL Designer It handles requests as follows e A request e g GET projects 1234 html comes in e The URL of the request is matched against the available routes A route is a regular expression linked to a handler class A handler class is a Python class with get post put and or
156. es required the general advice in Software Engineering is to prefer clear maintainable code over fast code until performance is proven to be an issue Concise code in the sense of compact but less readable is discouraged at all times Table 3 2 Difference between what is currently done by Engineering Automation developers and what they desire when they have to reuse code Currently done Desired Understandability e Low code quality but usually e Simple clear code with comments with comments e High level documentation about Part Il Initial study 22 e No external documentation the structure and the steps Validity e No systematic testing e Verifiable correctness Incentive problem The incentive problem rewarded for results is clear as the need to provide answers not software Any activity that isn t part of getting the answer is not rewarded and has to compete with more urgent work This applies to activities such as designing before implementing documenting testing systematically integrating with existing software and supporting reusing developers The interviews confirmed that not only are these activities unrewarded they are also perceived as time consuming In literature it is claimed that career aspirations which move away from software cause the lack of interest in Software Engineering training However in the interviews it was found that regardless of the career aspiratio
157. esign encourages reflection and creates documentation about the application structure Both are beneficial for the understandability of the software Software design involves reflection and subsequently iteration like nearly all other design tasks Reflecting individually or with peers which is recommended improves the design of the application A graphical description of the software design helps the developer to communicate the software design to others or perhaps to him or herself McConnell 2004 Having documentation about the software design improves the understandability and is especially helpful when changes to the system must be made To see how such an overview contributes to the ease with which a software system can be changed consider the Just In Time Comprehension model introduced in Singer amp Lethbridge 2010 The Just In Time Comprehension model describes how both experts and novices approach a software improvement task Both try to understand just what is necessary to make the changes properly The difference between the two is that while the expert can rely on his conceptual knowledge of the application and limit himself to recalling or learning the details the novice needs to learn both the conceptual level and the detailed level The lack of conceptual knowledge causes him to focus on irrelevant details he accidently learns more than he had to Unfortunately both the expert and novice quickly forget the details after making t
158. esign to the code Afterwards the average flow becomes close to zero indicating that about the same amount of information flows in both directions Figure 8 shows the flow evolution of the other projects except for the ones with purely negative flows i e flow 1 6 4 User feedback Besides measuring user activity through log files users were also asked for their opinions and remarks about GenDL Designer Due to the limited number of users so far the results here are mostly based on individual quotes The most often heard remark is I wish had this tool available when started with GenDL Some users indicate that it stimulates them to work more structured that it provides them with a clear overview of their project and that they found the overview helpful when explaining their software to others Several users feel that applications become more consistent and orderly One user predicts that if the concept would be used in an engineering company it would have a chaos reducing effect on the continued development of existing software Another user reported that making the design documentation for existing code revealed deficiencies in the code The users who started from scratch usually do not mention that using the tool saves them time but they do not mention that it costs time either It is noted however that fiddling with the diagram layout and lines takes time and that the tool is missing features to make this easier Part
159. esseseeseessssessessessessessssssssssessesssssessesssssssssssssssesseessesssassesseeseeseess 9 12 Maximum size of projects 9 13 Store timestamps in the datastore instead of memcache APPENDIX A QUICKSTART MANUAL APPENDIX B BOOTSTRAP PACKAGE CONTENTS APPENDIX C ADDING A STYLE TO IDEADIAGRAM Part Ill Code report Abbreviations 2D AMD API AST Blob CRUD CSS DOM GAE GDL GenDL GUI Hg HTML HTTP ID IDE JSON KBE MB REST RPC SaaS SDK SVG UI UML URL XML Yaml 2 Dimensional Asynchronous Module Definition Application Programming Interface Abstract Syntax Tree Binary Large Object Create Update Delete Cascading Style Sheets Document Object Model Google App Engine General Declarative Language General Declarative Language Graphical User Interface Mercurial version control Hypertext Markup Language Hypertext Transfer Protocol Identifier Integrated Development Environment JavaScript Object Notation Knowledge Based Engineering Megabyte Representational State Transfer Remote Procedure Calls Software as a Service Software Development Kit Scalable Vector Graphics User Interface Unified Modelling Language Uniform Resource Locator Extensible Mark up Language YAML Ain t Markup Language Part IIl Code report 1 Introduction The GenDL Designer is a web based application to support the design and implementation of GenDL applications This should make creating KBE applications easier and improve thei
160. from one central place A large cloud application hosting provider was preferred because of the level of continuity one can expect Many things could go wrong with a server of your own e g under some desk in the faculty Incidents are less likely to happen and much less likely to persist in a large cloud computing data center Finally the Google App Engine was chosen because of the free quota the low entry barrier the extensive documentation and the support for the Python language 3 System requirements and installation GAE applications can run on both the production servers of Google and your local development machine In the latter case the app is run and tested with the development web server included in the GAE Software Development Kit SDK You need a working local development installation before you can deploy install to GAE production servers 3 1 System requirements for a local installation The installation has only been tested in a Windows environment but should run on OSX and Linux as well as only platform independent technologies are used You will need to install e Python 2 7 32 bit o Python 2 7 not 3 x Python 3 is not supported by GEA at the moment o 32 bit because monitor exe will be compiled for the Python version you are running and you ll want to create a 32 bit version since that works also on 64 bit systems If you do not need to change the code client you could also use the 64 bit version of Python
161. ftware Engineering in Computational Science and Engineering 13 May 2008 Leipzig Germany Part Il Initial study 60 Segal J 2007 Some Problems of Professional End User Developers IEEE Symposium on Visual Languages and Human Centric Computing VL HCC 2007 pp 111 118 Shull F et al 2004 Knowledge Sharing Issues in Experimental Software Engineering Empirical Software Engineering 9 1 2 pp 111 137 Available at http www springerlink com openurl asp id doi 10 1023 B EMSE 0000013516 80487 33 Singer J amp Lethbridge T 2010 An examination of software engineering work practices CASCON First Decade Sitiol A amp Lee S 1999 An Approach To The Development of Hybrid Object Oriented Design Tool Malaysian Journal of Computer Science 12 1 pp 47 54 Sletholt M T et al 2012 What Do We Know about Scientific Software Development s Agile Practices Computing in Science Engineering 14 2 pp 24 37 Software Engineering Institute 2002 Capability Maturity Model Integration CMMISM Version 1 1 Speel P H amp Aben M 1998 Preserving conceptual structures in design and implementation of industrial KBSs International journal of human computer studies 49 4 pp 541 575 Stokes M 2001 Managing engineering knowledge MOKA methodology for knowledge based engineering applications Professional Engineering Publishing London Studer R Benjamins V R amp Fensel D 1998 Knowledge engineer
162. further You can read just the paragraphs that are relevant to the code you are interested in The table of contents can be used to quickly find the background information GenDL Designer was developed to improve the reusability of GenDL applications by encouraging the creating of correct documentation and encouraging design before code How this trickles down to business functions is shown in the diagram below GenDL Designer is written in Python server and code client and HTML CSS Javascript web client Both the web client and the code client communicate with the server through standard HTTP requests 5 To give an idea of the size of the code base currently the code base without external libraries consists of 281 files In total there are about 11 000 non empty lines of Python code 9 000 lines of HTML CSS Javascript code of which more than 6 000 is javascript and 500 lines in configuration files Part Ill Code report 22 2 System components overview 2 1 Function diagram The business functions use application services provided by one or more of the three subsystems of GenDL Designer This is shown in the diagram below The subsystems are e A Google App Engine App which runs on Google cloud infrastructure e A web client which runs in the browser on the user his computer e The so called code client which runs on the user his computer c writes a used by used by User
163. g similar software from scratch for a problem only slightly different or for the same problem by someone else in the same organization is waste and should be avoided as much as possible Also creating and sharing engineering automation software can help with the creation and distribution of how to knowledge within an organization since engineering automation software necessarily contains the knowledge required to execute an engineering task 3 Unfortunately the overall level of sharing and reuse in daily engineering automation practice is currently low 4 6 Engineering automation applications tend to be non transparent black boxes of varying quality 1 4 This makes it hard to adapt them to the constantly evolving knowledge in engineering work When not adapted the software will become increasingly obsolete and eventually will have to be discarded 7 In contrast transparent and high quality engineering automation applications can be adapted and therefore modified shared and reused rather than discarded This eliminates a source of waste in the engineering process Producing reusable applications proves to be difficult for engineering automation developers Being regular engineers they are not trained to develop software Their experience with abstraction and formal languages helps them to get started and get results quickly but as their software grows they find it hard to manage the increasing complexity They are unfamiliar with ba
164. get an example input file the method and the output by running the example It s very practical you can see how it is applied o 14 Self education is a good exercise for engineers it helps to develop the skill to gain more knowledge in something when you feel you need some The bad bit is that you re not always able to pick up best practises because your objective is to solve the problem o A2 What I found most disturbing about Lisp was that there were lots of keyboard shortcuts Very often I would get lost in the controls and I couldn t find out how I had to do it right Current Software Development Practise General Can you draw a timeline activity diagram of a typical project 12 13 Al A2 don t have a formal or defined process I1 uses a strategy where the software grows organically and iteratively functional from the beginning 14 uses a formal company specific process for large projects o 13 Pve never been formally introduced to version control If I write code within a day I find it pointless to use version control I can remember what changes I made How do you approach the requirements All participants except 14 only write down the main requirements mostly implicitly as reseatch project goals in non software exclusive research project documents More detailed requirements are not written down because you are supposed to know what you want your software to do and because detailed r
165. ggers the creation of documentation Instead it will only establish whether it increases the documentation quality and quantity The most trustworthy Part Thesis article 21 validation method remains deployment in industry with realistic use cases Such an experiment would be the next step 5 3 Metrics For the verification of GenDL Designer two metrics were developed They are based on the inconsistency between the design and the code The first is the overall consistency at the end of the project and is an indicator for correct and complete documentation The second is the flow and indicates whether the design was made before the code was written This metric is derived from measuring changes in inconsistency and the user activity that caused the change 5 3 1 Measuring inconsistency Each inconsistency notification is assigned a weight depending on the gravity of the inconsistency The weight includes the weight of sub inconsistencies such as missing attributes when the whole class is missing in the first place Weights for each type of inconsistency were chosen on a scale from 1 to 5 and are shown in Table 2 Table 2 Weights for different kinds of inconsistencies Inconsistency Weight Missing class 3 sub inconsistencies Missing function or method 5 Missing child 5 Missing attribute 3 Missing or different type of child Missing superclass 2 Different attribute kind i e settable or not 1 The total inconsistenc
166. gineering knowledge MOKA methodology for knowledge based engineering applications Professional Engineering Publishing London 2001 P H Speel and M Aben Preserving conceptual structures in design and implementation of industrial KBSs Int J Hum Comput Stud vol 49 no 4 pp 547 575 1998 P J Lovett A Ingram and C N Bancroft Knowledge based engineering for SMEs a methodology J Mater Process Technol vol 107 no 1 pp 384 389 2000 P Bermell Garcia A metamodel to annotate knowledge based engineering codes as entreprise knowledge resources Cranfield University 2007 C van der Velden C Bil and X Xu Adaptable methodology for automation application development Adv Eng Informatics vol 26 no 2 pp 231 250 2012 S Edwards Using software testing to move students from trial and error to reflection in action ACM SIGCSE Bull vol 36 no 1 p 26 Mar 2004 Part Thesis article 38 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 International Business Machines Corporation IBM The Rational Software Architect Family 2013 Sparx Systems Enterprise Architect software 2013 The ArgoUML Open Source Project Contributors ArgoUML software 2013 M Blaha and J Rumbaugh Object oriented modeling and design with UML Pearson Education 2005 T van den Berg Harnessing
167. gineers than a bespoke description based on software objects and object oriented modelling concepts Bermell Garcia 2007 p 74 Many people find it helpful to think through the design process in a sequential manner even though the design process is not directly represented in the final KBE application Using the process as a basis can help to ensure that the Product Model instance is complete Stokes 2001 p 242 The notation for describing the relation between the product model and the design process model see Figure 5 4 proposed in Stokes 2001 is however too verbose to be practical for anything but lab demonstration models Part Il Initial study 37 lt lt Feature gt gt Hole 7 a al Diameter Determine hole diameter To be evaluated Product model i Product model Bi Process model _ lt lt Feature gt gt Hole Determine screw thread lt Diameter 5mm Evaluated Figure 5 4 Relation between Product Model and Design Process Model Source Stokes 2001 Multiple levels of abstraction Models at higher levels of abstraction provide a clear overview by hiding lower level details Ideally the conceptual model at the highest level understandable by engineers must map to the design and the implementation component to component for increased understandability and maintainability Speel amp Aben 1998 Specifically in tasks which automation of design the reusabi
168. gram module displays an interactive diagram with toolbar and saves the diagram modifications to the server It saves both an image snapshot to the server which overwrites the old snapshot and the individual entities which are added updated and removed as necessary The diagram module doesn t communicate with the server directly the diagram_manager and entity_manager components are used instead The actual diagram is rendered with the ideaDiagram component This module loads ideaDiagram with a configuration for GenDL mainly the so called type_config It also hooks into the events of ideaDiagram to trigger saving modifications IdeaDiagram ideaDiagram creates mxGraph based drawing interfaces configured for drawing GenDL UML and UML Activity Diagrams ideaDiagram was originally created as the formal diagram editor for the IDEA environment Later it was reused and extended in GenDL Designer but it remains compatible with both A typical request is to add a new kind of element to the ideaDiagram vocabulary How to do this is described in appendix C 7 3 7 Class dialog module This module allows showing a class dialog with editable class data attributes methods It is used by the Diagram module The class data format was described earlier in section 4 2 1 Part IIl Code report 59 This module uses Jqote2 to render the class dialog content The content is shown in a dialog with the Bootstrap modal feature The contents are mad
169. h run your app All instances have their own Python runtime but communicate to the same datastore and memcache You only get enough instance hours to run 1 instance all day but that proves to be mote than sufficient so far even with dozens of simultaneous users 6 2 3 Template engines GenDL Designer uses templates extensively to generate HTML code for the web interface and GenDL code for the user Because of its importance and widespread use throughout GenDL Designer the concept is explained here before diving in the different subsystems Part IIl Code report 38 Introduction The concept is best explained with an example Consider the following template lt ul gt for project in projects lt li gt project name lt 1li gt endfor lt ul gt This is equivalent to print lt ul gt for project in projects print lt li gt project name lt li gt print lt ul gt and would produce for example lt ul gt lt li gt Project Bravo lt 1i gt lt 1i gt Renovation main building lt 1i gt lt ul gt which is the HTML expression for e Project Bravo e Renovation main building GenDL Designer generates code on the client side for code to show in code dialogs and the server side for code to put in the bootstrap package Since the generated code should be the same on both sides it would be preferred to have a single template for the server and the client which run on Pyth
170. h were not found in the code Together with the inconsistency the application tries to suggest one or more ways to resolve the issue In this case you can copy paste or even download a code snippet which will resolve the problem GendI Designer Demo Pieter Jan gendidesign appspot com proje Code snippet for wing in package demo pieter jan define object wing documentation author lt name gt lt username gt lt organization gt com description input slots fuel tank volume nil todo default value computed slots O robjects flap type high lift device Part Ill Code report Appendix A Project settings tab Synchronizing your lisp code files to the server is done with the code mirror tool a small Dropbox like utility available through the settings tab gendidesign appspot com projects 5681726336532480 html B cP V y G v E demo pieter jan f gt Main Diagram Consistency Design Code Main diagram gt E wing high lift device v Popular built ins geom base box gt ii geom base General gt ii surf gt ii gw Project settings Name Demo Pieter Jan G Package demo pieter jan Synchronization Before you can compare the design with the code you need to synchronize your external code base with the serve Type Downloads Local Current monitor path D Users Pieter GenDL files Jan Desktop tmp demo pieter jan Last detected code change 7 mi
171. have been shown to be extremely effective in finding defects and have also been shown to be more economical than e g testing One reason is that inspection can find defects testing can t e g unclear error messages and hardcoded values Only prototyping and high volume beta testing find more defects 6 2 2 Attitude of the participants Code reviews ate about criticizing and improving the code not about getting at the author Both the author and the reviewers need to reflect this in their attitude The author must be prepared to hear criticism no one writes perfect code Rather than taking it personal or trying to defend the code the author should be happy with discovered bugs it makes the software better In more subjective matters it is sufficient for the author to acknowledge remarks Later he can decide in private whether or not the remark is valid and whether the code needs to be changed There is no need to argue extensively about remarks during a code review itself Reviewers need to focus on discovering defects rather than criticising the author A useful technique for this is to address the code rather than the author the code does this rather than you always do this The review should be a positive learning experience Comments that are inappropriate in that respect should be flagged immediately No management should be present during the inspection the work isn t finished yet and it wouldn t be fair to m
172. he change Clearly novices or experts without a conceptual model of the considered piece of code can save time when the design is available as documentation to construct the conceptual model rather than having to reverse engineer the conceptual model from an overwhelming amount of details in the code 4 2 1 4 Supporting implementation code generation Despite the indirect advantages of creating a design described in the introduction Engineering Automation developers rarely create a software design because the justification for the effort is not directly apparent to them or to their supervisors The design is skipped and the source code becomes the only accurate description of the software McConnell 2004 Tool support for software design work as opposed to pen and paper design enables code generation With code generation it is immediately apparent to Engineering Automation developers that the design work reduces the subsequent code work Further advantages of code generation ate reductions in errors and improved consistency between the design and the code A possible downside is the need to express the design in a formal or at the very least a semi formal notation Bennett et al 2010 Tools for software engineers to create software designs already exist commercially and free some of which support code generation for general purpose programming languages Kelly 2007 argues however that the domain independent solutions from the Softwar
173. he keyboard delete button does not always work in a project with multiple diagrams The right click menu with delete option does work at all times and can be used with the problem occurs 3 4 Tree collapses and expands for no apparent reason The project tree can sometimes completely expand or collapse after it was updated automatically while it should preserve its state 3 5 Line numbers The line number indication is not always accurate it can be off a couple of lines Part IIl Code report 14 Section Il System Administration Part IIl Code report 15 1 Introduction The second Section of this report describes how to install and administrate the current version GenDL Designer Essentially GenDL Designer is offered as a Software as a Service SaaS solution GenDL Designer is built as a Python web application for the Google App Engine GAE infrastructure a cloud computing service with some free quota 2 The choice for SaaS and Google App Engine SaaS was chosen to make it users as easy as possible With SaaS users do not have to install software or updates change configuration settings deal with operating system compatibility or resolve conflicts with other installed software Additional advantages particularly interesting from the research perspective are the possibility to update the software frequently yet have everyone always working on the latest version and the possibility to gather usage statistics in real time
174. heir own professional goals While they are very sophisticated in their field of expertise they have received little Software Engineering education or training Engineering curricula nowadays commonly include one or more courses on particular programming languages Useful as these courses are they rarely treat Software Engineering in dept due to time constraints The definition of professional end user developers given above therefore still applies to engineers who followed these courses 2 4 Examples Examples of Engineering Automation are a production cost spreadsheet a heat transfer simulation model in a graphical modeling environment like Simulink or Labview pre and post processing scripts for an aerodynamic analysis an experimental finite element method implemented in a high performance language such as Fortran and a script to automatically draw a common part feature within CAD software Examples of software that are not included in the definition are top of the market simulation packages CAD software and PLM solutions because that software is typically implemented by or with extensive support from software engineers rather than regular engineers Part Thesis article 4 3 Engineering automation practice An initial study was performed to understand the needs and culture of engineering automation developers From literature a general view on Engineering Automation was developed Subsequently interviews with experts were held
175. ight click and drag with the right mouse button Deleting elements can be done with the delete button on the keyboard or through the right click menu The delete button does not work when multiple diagrams are used The right click menu always works Note that only when an element is no longer on any diagram it will be removed from the project The diagram is saved automatically after every change except for merely dragging around boxes or lines The save button is mainly provided for peace of mind When saving that button becomes disabled for a fraction of a second as visual feedback When hovering the button displays the time that has elapsed since the last save operation 2 2 Project tree The project tree shows all design elements in the project as well as available built ins from GenDL and GDL packages amp gendidesign appspot com pro v aa demo Main diagram Main diagram Z Detail of lifting surface a a gt B wing gt lifting surface v Popular built ins E geom base base object i E surf box solid 1 amp E surf cylinder solid 1 surf extruded solid 1G gs geom base ae SUIT gt Hawi The top package which is by default named after the project contains all design elements in any diagram in the project The other packages contain built ins Each of these built ins has a link to a documentation page The icons in the tree indicate the type of the element na Package f Diagram Class
176. ights in Engineering Automation development Part Il Initial study 58 8 References Abran A amp Moore J 2004 Guide to the software engineering body of knowledge Angyal L Lengyel L amp Charaf H 2008 A Synchronizing Technique for Syntactic Model Code Round Trip Engineering In 154 Annual IEEE International Conference and Workshop on the Engineering of Computer Based Systems ecbs 2008 Ieee pp 463 472 Avison D E et al 1999 Action research Communications of the ACM 42 1 pp 94 97 Basili V R amp Rombach H D 1991 Support for comprehensive reuse Software engineering journal 6 5 pp 303 316 Baskerville R 1999 Investigating information systems with action research Communications of the AIS 2 October Bennett J Cooper K amp Dai L 2010 Aspect oriented model driven skeleton code generation A gtaph based transformation approach Science of Computer Programming 75 8 pp 689 725 Van den Berg T Schut E J amp La Rocca G 2011 A heuristic approach to developing a wiring harness production support application In International Conference on Evolutionary and Deterministic Methods for Design Optimization and Control Bermell Garcia P et al 2012 A framework for management of Knowledge Based Engineering applications as software services Enabling personalization and codification Advanced Engineering Informatics Bermell Garcia P 2007 A metamodel to annotate knowledge bas
177. ill you might have broken something you aren t aware of A1 When I received the tool my supervisors assumed that it worked and that I could simply extend it Supervisor after finding out that wasn t the case Oh well then I must have read a very good report Separating theory and implementation errors was not felt to be an issue by the interviewed participants Rather they attribute errors to implementation errors on their part 3 3 2 Why Engineering Automation software is developed the way it is Clients and supervisors ask for engineering solutions and answers not software Delivering the answer is where most resources are devoted to The software and along with it Software Engineering practises receive the minimal amount of attention I2 Clients don t really care how you did it they even don t want to be told in a lot of cases Although they do want to know about the methods and assumptions used but they don t want to use the code A1 My supervisor even doesn t look to my code A2 Pm quite sure my supervisor never looked at the code of me or my predecessor What I do in the code doesn t matter to anyone so I don t have a reason to document or write really good code Getting the software to work prevails over reuse considerations There is little to no incentive to prepare for later reuse Instead activities such as documenting compete with more urgent work 13 I would
178. imself rather than to contract or hire a software engineer Part Il Initial study 1 e The development of engineering models and their implementations are usually highly coupled since the latter is often used to verify validate and improve the first This coupling makes it difficult to assign the model development and software development to different actors The choice between assigning the software development to a software engineer or a regular engineer boils down to a trade off between someone who is skilled in developing software but doesn t understand the engineering background and someone who does understand the engineering background but isn t particularly experienced in developing quality software efficiently In any case the engineers must be actively involved in the development and their evolving knowledge should be reflected quickly in the software Bermell Garcia et al 2012 1 3 Reuse of Engineering Automation To reduce the effort required in development it would be beneficial to have an extensive ability to reuse Engineering Automation development efforts from the past However practical experience in both industry and academia shows that reuse of both the code and the underlying ideas is problematic This is what motivates the current research From the perspective of industry practical arguments for reuse are the ability to do more engineering work with fewer engineers and the ability to deliver innovation faster
179. imulation data for visualization developing knowledge based engineering applications and data mining knowledge rules Two participants were graduating engineering students faculty of Aerospace Engineering Flight Power and Propulsion chair TU Delft selected because of their experience with a project involving the reuse of and collaboration on an application for conceptual aircraft design Part Thesis article 5 The interviews results will not be valid for the entire engineering automation community this would require a far larger amount of interviews than can be conducted within the scope of this project At most these interview results will confirm or nuance general findings from literature 3 3 Interview method The interviews were semi structured The same topics were discussed with each participant by using 22 fixed questions sent in advance Follow up questions allowed the interviewer to go into detail and understand each of the answers better The interviews took about 1 30 hour each and were recorded with the participants permission The interviews were processed in two steps In the first step the recording was transcribed and slightly summarized The summary was reviewed by the participant to correct any misunderstandings and filter out confidential information In the second step the answers of all participants were aggregated and compared per question Where applicable the most prominent answer was distilled by counti
180. in of many classes the mixin is defined in a diagram of its own and all classes who have it as a mixin only have a small uncluttered reference to the mixin class in their diagram An example of such a situation is shown above where lifting surface has a diagram of its own 2 4 Consistency checking Fundamental to GenDL Designer is the capability to check the consistency between the design and the code i e the UML diagrams and the GenDL GDL code For consistency checking your local code must be uploaded to the server This is done automatically by the code mirror tool The code mirror tool monitor exe can be downloaded via the settings page The tool is also included in the bootstrap package The code mirror tool does not need any configuration it is preconfigured just before downloading It does not need to be installed either After double clicking monitor exe the code mirror tool icon will become visible in your system tray as shown below and start uploading the lisp files it can find in the folder and subfolders of the folder it is in a a il meee O 15 46 Part Ill Code report 8 Right click the icon to pause resume or quit the code mirror tool A blue sign indicates that the tool is uploading files while a red warning sign indicates a connection problem The tool keeps monitoring the folder until it is quit Typically it takes about 5 seconds for the changes to propagate from your local file system to the server a
181. ine elements and other diagrams can reference them Shortcut elements can be added to a diagram to quickly navigate between diagrams The consistency pane shown in Figure 3 lists all inconsistencies with references to where the inconsistency was found Most consistency checks verify whether for each element in the design and the code a corresponding element with the same name can be found in the code and the design respectively As a result inconsistencies are typically reported as a missing element The user has full control over the resolution of inconsistencies they are not resolved automatically This ensures that the design tool will not block the workflow of the user even when some notifications would be incorrect or irrelevant for any reason This makes the tool more robust towards the future It also assures to new and suspicious users that their design and more importantly their code is safe building trust Below each inconsistency in grey possible solutions are suggested Where possible GenDL Designer provides links to resolve inconsistencies automatically or semi automatically e g by generating code skeletons as shown in Figure 4 The code skeletons are empty since their content is not part of the design The generated code snippets contain todo markers at the position where the user is expected to add detailed code The settings pane finally allows the user to adjust project settings download the code mirroring t
182. ing Principles and methods Data amp Knowledge Engineering 25 1 2 pp 161 197 Tao F et al 2005 Managing the semantic aspects of learning using the knowledge life cycle Fifth IEEE International Conference on Advanced Learning Technologies ICALT05 pp 575 579 Tri D Q amp Tho Q T 2012 Systematic Diagram Refinement for Code Generation in SEAM 2012 Fourth International Conference on Knowledge and Systems Engineering pp 203 210 Van der Velden C Bil C amp Xu X 2012 Adaptable methodology for automation application development Advanced Engineering Informatics 26 2 pp 231 250 Verhagen W J C et al 2012 A critical review of Knowledge Based Engineering An identification of research challenges Advanced Engineering Informatics 26 1 pp 5 15 Wachsmuth G 2011 Compiler Construction course TU Delft Wang Y amp Shao J 2003 Measurement of the cognitive functional complexity of software The Second IEEE International Conference on Cognitive Informatics 2003 Proceedings pp 67 74 Available at http ieeexplore ieee org Ipdocs epic03 wrapper htm arnumber 1225955 Williams K 2006 Migrating to Rational Systems Developer Part 1 Rational Rose Models Available at http www ibm com developerworks rational tutorials r migtorsd r migtorsd pdf pdf Ye Y 2006 Supporting software development as knowledge intensive and collaborative activity In Workshop on interdisciplinary software engi
183. ing and reuse by improving understandability and validity As there is little doubt on the positive effects of understandability and validity on reuse the current work will focus on whether understandability and validity were actually improved For this user feedback will be gathered 5 2 Academic experiment A first large scale experiment will be held in the context of a graduate course at the TU Delft on Knowledge Based Engineering KBE As part of the course about 50 students learn to program in GenDL The final assignment is to develop a KBE application using GenDL in teams of two The system and its documentation have to be defended during a review session This course has been held for several years and the general experience of the tutors is that students struggle to pick up GenDL programming that they structure their code poorly and that they do not create a design beforehand even though they were advised to do so Instead they draw diagrams as documentation shortly before the review session since it is a required deliverable They do not fully embrace the diagrams as a design tool The boundary conditions of the course are similar to general engineering automation boundary conditions a stringent deadline and limited experience with the software language and software development in general What is different however is the up front requirement for documentation The experiment will therefore not establish whether GenDL Designer tri
184. ing and then you could consider its usefulness in later projects Recognized by all participants Activities with long term objectives such as documentation activities and training activities are under resourced Recognized by all participants However some participants indicate that they don t find it necessary to spend more time on documentation and training You are rewarded for publications and results not for software Recognized by all participants Software developers are at the bottom of the research ladder The career aspirations of many involve moving away from developing software The situation is recognized but not necessarily accepted by all participants except 14 Interestingly the participants that had personal motivation to write software 13 and A1 don t associate themselves with a career that moves away from software instead their aspiration is to keep writing software as their career develops There is a high lapse of developers Recognized by some participants but none has the feeling it has affected them such that it became problematic Maintainability is not considered until it is already a problem Mixed response Only I4 seems to be aware of what can be done to avoid maintainability problems Part Il Initial Study Appendix B Current Practise in Engineering Automation Reuse To what extend are applications currently shared and reused In the case of the industrial participants 11 12 13 14 code i
185. ing information readable maintainable transparent etc Bermell Garcia 2007 Repositories should describe the context and how the information is to be applied how to knowledge Markus 2001 An engineering app store has the advantage that the knowledge is connected to an executable process which shows the context and how it can be applied In addition it is reported that a process view facilitates comprehension an opinion which was also frequently encountered in the interviews performed A practical suggestion to encourage reuse is to index design knowledge according to the issues in which it is useful and its role in the design process Bermell Gatcia 2007 Networking Earlier it was discussed how the view on knowledge has changed over time from a possession that can be captured to a socially embedded phenomenon For example in the Nonaka Takeuchi knowledge sharing model this is facilitated by the socialization phase where tacit knowledge is shared among researchers This socialization can be supported with an engineering app store by introducing a networking aspect Statistics mined from usage data can drive social features such as an expert finder peer finder tool finder and workflow suggestions Murphy et al 2008 App package The engineering automation applications are stored in the engineering app store as app packages The first part of these packages is the software and related development artefacts such
186. ion Tests The data processing functions are very suited for unit testing in part the module was structured into small functions for easy testing The unit tests are in test test_dataprocessing 6 3 8 webhash webhash is a simple module to create hashes from files and data This is used by both the code client and the server if the hash of a file is no longer equal the file was changed Obviously the hashes have to be generated in identically the same way Also when transferring the hashes encoding and decoding should not modify the hashes Webhash simply takes the md5 hash of data and encodes the hash in base64 for safe transmission The test scripts for webhash can be found in test test_webhash 6 3 9 time_ago time_ago is a simple module to generate strings like 6 minutes ago from a timestamp by compating the timestamp to the current timestamp This functionality is used in various places and can possible be reused in other projects as well Therefore it was placed in a separate module Part IIl Code report 50 6 4 Serverapp configuration and loading 6 4 1 app yaml configuration GenDl Designer s app yam1 declares a handler for static files simply the contents of a folder and one for dynamic pages the Webapp2 application serverapp app Note that handlers here are not the RequestHandlers inside Webapp2 apps There is also a handler for the website icon in particular The entire webclient is made available unde
187. ion to Reproducibility impeded knowledge under pressure Part Il Initial study 23 Figure 3 1 Underlying problems and consequences in Engineering Automation 3 5 Conclusions The interview results largely confirmed and also clarified the earlier findings from the literature review on Engineering Automation Sharing and reuse of Engineering Automation software takes place on a small informal scale within teams The two most pressing issues that impede reuse are understandability and validity When these issues are resolved raising the level of reuse further will require scaling up the internal team interaction on which reuse now relies This is shown in Figure 3 2 Also several non technological obstacles related to the current Engineering Automation culture were identified These must be accounted for when introducing change Social context Individual context Understandability Code quality amp documentation Towards more reuse Validity Social interaction Testing amp doc validity Beyond team Figure 3 2 Schematic problem representation obstacles that impede reuse The understandability of the code is low due to the lack of high level documentation and due to unclear code High level documentation is desired to help understand the structure of the code and the flow through the code The absence of systematic testing undermines the validity of the code Discrepancies between the docume
188. ipter automating amp sharing how to knowledge in the enterprise In Proceedings of the twenty sixth annual SIGCHI conference on Human factors in computing systems pp 1719 1728 Lieberman H et al 2006 End user development An emerging paradigm In End user development Springer pp 1 8 Lovett P J Ingram A amp Bancroft C N 2000 Knowledge based engineering for SMEs a methodology Journal of materials processing technology 107 1 pp 384 389 Markus M L 2001 Toward a theory of knowledge reuse Types of knowledge reuse situations and factors in reuse success Journal of management information systems 18 1 pp 57 94 McAfee A P 2006 Enterprise 2 0 The dawn of emergent collaboration Management of Technology and Innovation 47 3 pp 20 28 McConnell S 2004 Code complete Microsoft press M ry D amp Singh N K 2011 A generic framework from modeling to code Innovations in Systems and Software Engineering 7 4 pp 227 235 Misra S 2011 Cognitive complexity measures an analysis Modern Software Engineering Concepts and Practices Advanced Approaches pp 263 279 Mutphy C et al 2008 genSpace Exploring social networking metaphors for knowledge sharing and scientific collaborative work 2008 23rd IEEE ACM International Conference on Automated Software Engineering Workshops pp 34 41 Neumiiller C amp Grtinbacher P 2006 Automating software traceability in very small companies A ca
189. is mainly attributed to tension between research goals and software engineering goals the attention tends to shift to the first The software is in the first place considered as a research tool to address immediate research needs As a result activities related to software quality and reuse i e long term usefulness of the software are under resourced These activities include documenting distributing and supporting software and following Software Engineering training In the engineering domain the secondary role of software goals is confirmed by Elgh 2008 and La Rocca 2011 who warns for ad hoc development for short term gains the limited durability will result in rework Howison amp Herbsleb 2011 adds that distributing and supporting reseatch softwate is in fact a time consuming activity The underlying cause identified by Howison amp Herbsleb 2011 for the scientific software community is an incentive problem academics are rewarded for publications not for software The incentive problem is also easily recognized outside academic environments Engineering Automation software is developed for its output The incentive to increase the maintainability only appears later in the life cycle of the software when maintenance activities already turn out to be problematic Elgh 2008 Another explanation is that software developers are often at the bottom of the research ladder Investing in software skills is not in line with their career as
190. ith the new render style will be created by the user Also describe the data format expected Typically name string is sufficient Normally the name attribute is used as cell label If this is not sufficient you have to modify setup_labels Add render logic in creating main js Add logic to _render_cell that renders the cell and possibly subcells The existing code should be a good example of what is possible It can be as simple as creating a single vertex but custom logic can include the rendering of subcells In that case it is recommended to move this code to a separate file Add an entry to cell_updater_creators for the new render style cell_updater_creators has for each render style a function that takes a cell and returns an update function The update function takes data and updates the cell For simple cells default_updater_creator can be used For more complex cells with subcells updating will be more difficult and that logic should be in a separate file Part IIl Code report Appendix C
191. ivity that doesn t directly support this objective receives little attention As a result little is done to make it convenient to reuse the development effort put in Engineering Automation software Interviews with 4 professional engineers at EADS Innovation Works and 2 engineering students at the TU Delft AE FPP yielded similar findings and provided further insight in how engineering automation software is developed The two most pressing issues that impede reuse are understandability and validity When these issues are resolved raising the level of reuse further will require scaling up the internal team interaction on which reuse now relies Also several non technological obstacles related to the current Engineering Automation culture were identified These must be accounted for when introducing change laborious tasks are skipped where possible there is only an incentive for answers Software Engineering experience is limited and the development is highly iterative and incremental Four solution concepts for more reusable software were explored a graphical design tool an engineering app store code reviews and coding policies These concepts were evaluated for their potential to improve reuse judged from their alignment with the two main issues and the four non technical obstacles found earlier A trade off identified the graphical design tool as the most promising The project will implement a prototype of a graphical design tool for Gen
192. lability of abstraction mechanisms Secondary notation extra information in means other than formal syntax 9 Closeness of mapping closeness of representation to domain 10 Consistency similar semantics are expressed in similar syntactic forms 11 Diffuseness verbosity of language 12 Hard mental operations high demand on cognitive resources 13 Provisionality degree of commitment to actions or marks 14 Progressive evaluation work to date can be checked at any time ROS BON a As a guideline 6 generic activities are suggested incrementation transcription modification exploratory design searching and exploratory understanding Exploratory design is the most demanding designers continuously make changes at many levels from details to fundamental design changes The suggested system properties which are most important for such an activity are low viscosity reduced premature commitment high visibility and high role expressiveness Part Il Initial study 34 Frequently a trade off must be made between several dimensions improving one dimension tends to deteriorate another For example introducing abstractions can reduce viscosity by providing a single place to edit a common value but at the same time this tends to introduce a hidden dependency Typical trade offs are shown in Figure 5 1 can increase ee viscosity lt secondary notation need for lookahead can reduce can increase abstractions can increa
193. le and that usage statistics were gathered They were also encouraged to contact the experiment coordinator when they encountered issues Only a few users contacted the researchers mainly for reporting a bug that blocked them Most users preferred to get along themselves In a few instances the coordinator visited or contacted the participants to check up Sometimes users had indeed questions and opened the GenDL Designer application to discuss In those instances it was sometimes possible to give the user extra tips regarding less obvious features of GenDL Designer such as the possibility to create several interlinked diagrams Overall the researchers tried to remain neutral and stressed that participants had to use the software the way they found it most useful even if that meant not using it at all Part Thesis article 28 7 Discussion In anticipation of the results of the actual experiment the results of the trial runs are already discussed to arrive at preliminary conclusions 7 1 Consistency metric The users who started from scratch share the same tendency for high consistency This is encouraging since this implies the creation of design documentation Without GenDL Designer this is usually not created The driving force behind this tendency is most likely the consistency list all users watched the consistency pane regularly The users apparently felt the need to eliminate the notifications in the list One reason
194. level documentation and due to unclear code High level documentation is desired to help understand the structure of the code and the process steps Validity The validity of the code is undermined by the absence of systematic testing Discrepancies between the documentation or reports and the actual software further reduce the trustworthiness In addition to identifying technological issues the literature and interviews also revealed important non technological obstacles in the current engineering automation culture that must be accounted for when introducing change Four such obstacles were found Skipping laborious tasks where possible Software activities other than software construction itself are perceived as time consuming and not a necessity This applies to Part Thesis article 12 activities like designing documenting testing integrating and supporting software and software engineering training Only incentive for answers Engineering automation developers are not rewarded for those non construction activities They do these activities only as far as they can justify these activities because it helps them to get an answer to the engineering problem they are trying to solve Effort to make their software more reusable comes at their own expense Limited Software Engineering experience Engineering automation developers have limited experience with Software Engineering and the basic practices used by software engineers They are
195. like to describe the engineering knowledge also for myself because I like to keep track of those things but generally there is no time The codes that are written are just there to do the job get a value out Part Il Initial study 19 A2 I now have already plenty of work so I won t spend more time on making things pretty But I think that if they had insisted on me taking care of the code rather than giving me more work my work would have been more valuable for the following students There is very limited attention for Software Engineering practises which would increase the quality of the software and improve its reusability This includes training development methodologies testing and documenting I3 There is no formal training program We wouldn t even talk about that It s almost assumed in the team that people can program They don t expect you to do a brilliant job but they expect you to do a good enough job 14 Tve heard of MOKA when I was in the KBE team but it was perceived as a fairly theoretical overhead A1 I know it exists things like Systems Engineering and I know they can be very valuable I stay critical enough towards my work to arrive at good results I consider it disproportionally much effort to use Systems Engineering I know that s not a very good argument Little resources are dedicated to them and the enthusiasm for devoting more resources is generally
196. lity of software components can be linked to the reusability of modular design activities Bermell Garcia 2007 Hiding lower level details benefits the developer in several ways From the perspective of Knowledge Management the layered presentation of information allows to locate comprehend and use information in a timely fashion which is required for successful knowledge collaboration Ye 2006 In addition modularity and information hiding are the cornerstones of modern Software Engineering Goodliffe 2007 McConnell 2004 Consistency Having a model split in multiple views and levels increases the need for links between elements in different views or levels and for consistency checking Consistency can either be enforced or the user can be notified of inconsistencies Peckham amp MacKellar 2001 Enforcing consistency is not always desired the shortest path from one consistent state to another state might involve inconsistent states 5 2 4 Unified Modelling Language The Unified Modelling Language UML was created by Rumbaugh Booch and Jacobson and is based on their previous work with object oriented modelling and design Essentially UML provides notations to describe a class model a state model and an interaction model The class model describes static structure It consists of class diagrams showing classes optionally their attributes and methods and relations between classes An example is shown in Figure 5 5 The state model de
197. med that the actual development is performed by software professionals and the user can only change pre defined configuration options or request functionality changes Examples of End User Development are spreadsheets recording macros in word processors customizing email filtering and processing rules but also low entry level scripting interfaces embedded in certain applications In Lieberman et al 2006 End User Development is formally defined as A set of methods techniques and tools that allow users of software systems who are acting as non professional software developers at some point to create modify or extend a software artefact A major driver for End User Development is the diverse and changing nature of requirements With conventional development cycles software professionals have to keep up with the ever changing requirements of all their users This makes the development process slow time consuming and expensive This situation is avoided if users themselves would be able to continuously adapt the systems to their needs Lieberman et al 2006 2 3 3 Professional End User Development A major target group in current End User Development research are professionals in diverse areas outside computer science such as engineering medicine business and more Some of the tasks they need to perform are suitable for automation yet they might be so specific that a commercial solution is not readily available Lieberman et
198. might be that the notifications are presented as errors something is wrong and needs to be fixed As long as the errors are there the solution is not ready and the problem is not solved Another reason might be that at least initially the users adhered to the usage pattern they had seen in the short demonstration and or the quickstart manual This pattern focuses on resolving inconsistencies that emerge Two of the three users who documented existing code documented a very large portion of their code one only documented a small portion Interestingly the 3 users who used GenDL Designer to create documentation did not create more documentation that the 4 users who did not use GenDL Designer for the sake of creating documentation 7 2 Flow metric Users who started from scratch design more in the beginning than in the end the flow tends to decline This is what you would expect once users are in the code small problems are easier fixed there It is interesting to note that different users working on the same toy project share the declining flow evolution pattern but still differ significantly in average flow users 1 and 2 versus user 3 flow 0 54 0 65 and 0 27 respectively To some degree these numbers quantify the personal preference for working top down or bottom up i e structure rigorously first or let the structure evolve out of working software This personal preference has a notable effect on how GenDL Designer is used and maybe
199. might work for their input but it can break yours When things like that happen you quickly loose interest in pulling changes from others 3 4 4 What more reuse of engineering automation would require Literature Activities related to software quality and software reuse are under resourced because research goals prevail over software goals These activities include documenting Part Thesis article 9 distributing and supporting software and following software engineering training Distributing and supporting research software is in fact a very time consuming activity 4 6 Reuse is further impeded by the instability of the communities of practice Software developers are at the bottom of the research ladder and career moves such as graduation or promotion cause a high turnaround and associated knowledge loss 4 Interviews The competition for resources between working on current research and preparing for reuse in future research was clearly present in the interviews Getting the software to work prevails there is little to no incentive to prepare for later reuse Even though sharing and collaboration might be overall beneficial to the organization it introduces overhead which comes at the expense of the one that must facilitate reuse One participant said I now have already plenty of work so will not spend more time on making things pretty But think that if they had insisted on me taking care of the code rather th
200. n I found is that no one reads it people rather look over their desk and ask directly o 13 Creating a diagram In this case it served two purposes e It was primarily for me to remember how it all came together so I needed to map it out I do this more often now after seeing how much it helped me keep focussed on what I want the program to do and how it all fits together e Show all team members and stakeholders that the numbers were not drawn out of thin air They wouldn t look into the code but this way they could understand what my calculation methods were o 13 I would like to describe the engineering knowledge also for myself because I like to keep track of those things but generally there is no time The codes that are written are just there to do the job get a value out o A2 An activity diagram would be handy so you know what happens The connection with the actual code must be clear however You need to see how elements in the diagram map to what you have to do in the code o A2 I now have already plenty of work so I won t spend more time on making things pretty But I think that if they had insisted on me taking care of the code rather than giving me more work my work would have been more valuable for the following students If relevant How do you manage the relations traceability between code documentation design documents and other software artefacts With a version control with comments N
201. n is distinguished from mainstream Software Engineering because it is created by engineers designers and researchers with limited training and or experience in Software Engineering if any Unfortunately many insights from Knowledge Management and Software Engineering are not applied in Engineering Automation Starting with Knowledge Management from a technical point of view documentation is insufficient so that knowledge remains in the head of the original developer From a social point of view there is a lack of sharing and collaboration networks Next engineering automation developers are unfamiliar with practises common in Software Engineering There is a lack of requirements and systematic testing Formal methodologies for designing applications e g MOKA are perceived as overhead and therefore not used Basic tools such as version control systems and debuggers are overlooked Sharing and collaboration is limited as later reuse is not adequately supported the software quality suffers from low understandability and the absence of systematic testing The underlying problem is an incentive problem engineers and researchers are professional end user developers who need results and answers Software activities other than software construction e g designing documenting testing improving code comprehensibility integrating and supporting are mostly unrewarded and often also perceived as time consuming There is no doubt that Knowledge M
202. n than source code comments and perhaps a thesis or paper I3 I usually share my code with people who work very close to me We save a lot of time by speaking to each other rather than writing things down The problem of documentation I found is that no one reads it people rather look over their desk and ask directly 3 3 4 What more reuse of Engineering Automation would actually require The practical experience of the two academic participants who had to extend existing software showed two reuse traps it was difficult to understand the code because of its low quality and poor documentation and there were hidden assumptions and flaws under the hood Part Il Initial study 20 A1 Emphasizing the contrast between the code he got and his own code Over a couple of weeks months time I found out that much of it wasn t correct or accurate enough I know that what is in my report corresponds to the code Even though sharing and collaboration might be overall beneficial to the organization it introduces some overhead which comes at the expense of the one that must facilitate reuse This overhead can be the need to tidy up the code provide support communicate or integrate I3 admits that he isn t keen on sharing his code outside the team then he would need to provide support and need to better clean up his code This interferes with the regular activities for which I3 is contracted A1 about a fellow student He
203. nager object it was given The modules are shown below in a diagram depicting the direct dependencies depends on and in a diagram depicting indirect dependencies uses Direct dependencies l l I l l fr a n 1 Ld tree_display tab_manager diagram_manager consistency_display settings L L E Ld L L H 4 f _ uto_update_tab auto_update_ generic_tree_display diagram consistency resolution C C T T aaa Lo Depends on class_dialog code_dialog code_generator L Lo L The dependencies are in the first place related to user interface elements The start point for loading modules is the HTML page itself The HTML page creates an entity manager which allows sending and retrieving individual elements of the design to the server Next it loads the interface the project tree the tab manager and the default tabs The single entity manager object that the HTML page creates is shared among the project tree diagram tabs and consistency resolution tabs Part Ill Code report 57 Indirect dependencies objects that use other objects L diagram_manager L T CI tab_manager C consistency_resolution V C entity manager C The indirect dependencies are related to functionality e A diagram saves modification to the design in a diagram with the entity manager e A diagram can open a new diagram with the diagram manager
204. names Having to guess or look up variable names over and over again is time consuming and error prone In many cases a good naming convention can help The use of lowercase and uppercase letters to make a distinction between types variables and constants is an example e g CamelCase for types camelCase or no_camel_case for variables and UPPER_CASE for constants Another example are modifiers such as total average Knowing that you and your team always add the modifier to the end removes the confusion between costsTotal and totalCosts Favour clear code over commented code Some research suggests that the most readable programs have one line of comments per 5 10 lines of code However adding comments to meet this metric doesn t necessarily make the code more readable Such a policy would address the symptom of programmers not writing clear code but it doesn t address the cause Comments are a last resort Rather than writing more comments put the extra effort into making the code itself more readable and clear Let comments express indent and overview There is no point in repeating the code in the comments Comments should only express what the code itself can t Comments should express why the code is the way it is This is especially important when cryptic performance optimizations have been applied The use of flags in comments should also be standardized e g todo or FIXME so that the flags are
205. nd giving you fewer things to worry about McConnell 2004 Similar to coding policies that relate to the code one can formalize agreements with respect to testing version control and committing designing documenting reviewing etc Coding policies can steer important decisions to protect against known pitfalls For arbitrary choices such as layout coding policies eliminate the need for pointless decision making In a team coding policies encourage a uniform code style so that developers can focus on important aspects of other developers their code rather than keeping track of arbitrary decisions or being distracted by insignificant variations 6 3 2 Introducing a coding policy Although there is general agreement about what practises are good and bad every developer will have his own idea about that is more aesthetic The point is not to get caught up in the unimportant details Focus on things that really matter and give the biggest improvements for your team s code There must be a balance between sufficiently detailed and still having wide consensus support Leave rare but tedious cases to individual taste if they don t make much of a difference and allow the rules to be broken for genuine cases Pushing down a coding policy from the management generally doesn t work and threatening people in order to make sure they will use the policy will only make them feel more negative about it Instead when introducing a coding policy th
206. nd the browser Once code has been uploaded the current list of inconsistencies can be viewed in the Consistency Design Code tab shown below Main diagram Consistency Design Code Settings Help cen Design Code Description i The UML attribute lift ofthe UML class lifting device inthe design doesn t have a corresponding slot lift in i define object lifting device inthe code in sourcellifting device lisp 3 code snippet gt The UML method xfoil analysis ofthe UML class lifting device inthe design doesnt have a corresponding function xfoil analysis in define object lifting device inthe code in sourcel lifting device lisp 3 code snippet A B wing has different documentation in the design and in the code in source wing lisp 3 i Design Code The tab displays the list of inconsistencies found Each notification has the same structure e Two icons indicate the type of element to which the notification applies and whether that element was found in the design the code or both e The inconsistency is described typically in a wording which shows which elements do not map properly onto each other or which element has a missing mapped element Where possible a link to the source code file and a line number are provided e Advice is provided on what you can do to resolve the inconsistency In some cases automatic or semi automatic resolution options are available These are provided as clickable blue links
207. ne Starting from the top level function compare_functions you can follow the trail to the set of functions that the design and the code have in common For that set the description of the corresponding functions is already checked Below that check you could add the argument checking logic To keep the checking logic understandable it is strongly advised to adhere to the practice of defining compare functions rather than putting in the logic directly These functions are easier to understand and easier to test than monolithic functions that compare a full class and everything in it Compare functions typically have this structure def compare_Xes parent_element_name parent_element_data Xes_in_design Xes_in_code inconsistencies warnings errors an Ae Compare Xes in the design and code param parent_element_nam str not always needed e g name of the class when comparing slots param parent_element_data dict not always needed for file name and line number param Xes_in_design list of X dictionaries Format param Xes_in_code list of X dictionaries Format param inconsistencies lt inconsistency gt list to append inconsistencies to param warnings string list to append warnings to param errors string list to append errors to Pik es in_design_only in_common in_code_only partition_by_name X design Xes_in_design code Xes_in_code erro
208. needed results for him it wasn t interesting to make sure that his adjustments worked in all situations A central repository to facilitate sharing and collaboration could not only be used to share code but also act as a learning resource for the knowledge embedded in the code I3 It would be good to have a central repository where high quality code is available but I think mostly as a learning resource Actual reuse will be difficult because it is unlikely that what you need is exactly available and because to know all the assumptions you need to go through lots of code or lots of documentation if the author had the time to write it 14 He was interested in how I solved the problem at the design level which modules and how P S they interact This was transferred in an informal way as a slideshow This discussed the input the business requirements algorithms steps of the process etc To determine whether the software could serve the intended goal functionality what the softwate does and how well and example input output are the most important Support references to supporting material and high quality code are appreciated but not mandatory For actual reuse requirements are not expected What is expected is design documentation that shows a clear structure i e the program broken down in modular blocks and an understandable flow also referred to as story storyboard or engineering process
209. neering research WISER 06 New York New York USA ACM Press p 15 Part Il Initial study 61 Appendix A Interview agenda and questions Review of the A cette spices sascaiatenisneiiissa writbeiarat tint aiabaaotar a aa naaa 1 Aimof ENG Tnterview 5 Min zetonen a a a a a aniston awrite 1 Context questions 1 titi atuscrssscninatiit A a AE A A A AA 1 C trent S ftwafte Developmen t PracHse iicuinatintectesiasiuiaaetiacaccin anaa a 2 General OO mH a a a EE a E A A T RS 2 D cument ti n 5 MN aeeoe aare NANESE OKENE Ea a anvo Ea sanstaasseaiedsdibade cal ERESSE SESSA EAE ST usted 2 Motivation Incentives Attitude towards software 10 Min s sessessessssssessesressesstesteseeseersrssrssresresees 2 Current Practise in Engineering Automation Reuse 15 MIN s sssssssssesssesseesseesreessressresrreseresereseeeseeeneess 3 Requirements for Engineering Automation Reuse s ssessesessseressrieseesseesseesseeereesstessresreesereneeenerenresnresnrese 3 Relevance Assessment OMN snan ee a a a ea AN REN Sa eaa aE 3 Act al Reuse 10 acc Men err rer epe ter or ere kaa E E E E ere 3 Review of the Agenda Total time 1h 30 estimated to be adjusted if necessary Aim of the Interview 5 min Engineering Automation software is software written by engineers which automates an engineering task This includes spreadsheets Matlab scripts pre and post processing simulation data etc Much if this software is not very future proof it is hard to ad
210. nerates There is a global event handler that listens for clicks on these The links have a command and inconsistency index associated with them That s how the global listener knows what to do with the click on the link Actually resolving inconsistencies when possible is done with the consistency_resolution module Some inconsistencies are things the user cannot modify in the design the body of slots and child constraints These are correctly reported as inconsistencies but the user shouldn t be bothered with them In fact the design should just display the values given in the code Just before rendering the consistency tab filters out these inconsistencies and applies them directly with the consistency_resolution module 7 3 9 Consistency resolution module This module supports automatic or semi automatic resolution of inconsistencies There are three sub modules corresponding to the different resolution commands that can be given add_to_design show_code_snippet and update_design_data Other modules don t use the sub modules the main module provides resolvers These are functions that take the inconsistency type and inconsistency data as they were retrieved from the server and resolve the inconsistency They return whether the operation was successful add_to_design and update_design_data use the entity manager to manipulate the design When calling the add and update methods of the entity manager they pass on an even
211. ng how many of the participants referred to that answer and by taking into account the importance indicated by the participants This was published in 17 3 4 Results of literature review and interviews The findings from literature and interviews are presented side by side per focus question 3 4 1 How engineering automation software is currently developed Literature Based on experiences of her own and of colleagues 18 states that the average academic engineering automation researcher is far removed from the software engineering world Common development practices from Software Engineering are not adopted requirements are not explicit software design is done minimally testing is done ad hoc documentation often skipped and the traceability between the engineering knowledge and the application code is missing 1 5 9 Between plan based traditional development methods and agile development methods engineering automation developers choose the latter but they adopt them only selectively communication and flexibility are embraced they work highly iteratively incrementally and interactively but in the areas of requirements and testing agile practices are not adopted 5 19 The software development practices that are used are error prone For example it is common to passively look for incorrect behavior rather than actively gathering evidence of correct behavior 5 Part Thesis article 6 Interviews Thes
212. ns there is little interest in serious Software Engineering training Consequences The problem of limited reuse can now be structured from root problem to consequences as shown in Figure 3 1 The root problem the incentive scheme explains why it is hard to understand current Engineering Automation software and why there is a lack of systematic testing and validation These in turn are found to trigger three consequences 1 Contribution to knowledge under pressure lack of transparency makes applications difficult to audit The possibility of errors or differences between the described method if any and the implementation undermine the validity of the conclusions 2 Reproducibility impeded not enough information to reproduce a part of the calculations with reasonable effort 3 Limited support for future research reuse unprofessional attitude where other s research is used but reuse of your own research is not facilitated properly For any organization that aims for sustained research these consequences ate severe and provide additional reason for an investigation into measures to promote reuse Rewarded for answers publications and results Root problem not for software the computation process or intermediate results Software problems rd Hard to understand Lack of systematic and rigorous current EA software testing and validation Consequences Limited support for future research reuse Contribut
213. ntation generation GenDL Designer was developed with Engineering Automation developers in mind and therefore differs significantly from general software engineering tools with similar objectives The success of a software process improvement project largely depends on non technological aspects 14 It was key to understand the culture and needs of engineering automation developers to gain commitment and avoid resistance Engineering automation practice was investigated with a literature review and with expert interviews The gained understanding was subsequently applied in GenDL Designer There are two main contributions to be found in this article First it presents a literature review on Engineering Automation and the results of a set of interviews with engineering automation experts These show which issues impede reuse most at the moment and explain the most prominent non technological obstacles Second it presents preliminary results from the deployment of a support system for Engineering Automation This experiment addresses the potential and feasibility of incremental code and design documentation generation for engineering automation development specifically Part Thesis article 2 2 Definition of Engineering Automation Engineering Automation was defined before as the software implementation of engineering models and solutions by engineers themselves to automate their own engineering tasks In this section the definition
214. ntation or reports and the actual software further reduce the trustworthiness The main challenge is not so much finding strategies which promote reuse but rather to find strategies which engineers are able and willing to adopt Therefore the strategies will have to deal with four non technological obstacles that became apparent in the literature review and the interviews Firstly software activities other than software construction itself are perceived as time consuming and not a necessity This applies to activities like designing documenting testing integrating and supporting software Secondly Engineering Automation developers are not rewarded for these activities They do these activities only as far as they can justify these activities because it helps them to get an answer to the engineering problem they are trying to solve Making their software more reusable by doing more comes at their own expense Thirdly Engineering Automation developers have limited experience with Software Engineering and the basic practises used by software engineers They are unaware of how Software Engineering practises like designing and testing or even basic tools can fix the problems they are experiencing Finally Engineering Automation software is developed iteratively incrementally and interactively It is common for a project to start vague without many requirements and without Part Il Initial study 24 much of a design Developing the calculation
215. nutes ago File count 0 Code mirror tool The code mirror tool is a small dropbox like utility that monitors lisp files and uploads them to the server when they are changed It does not make changes to your code and does not require installation Simply put it in the root folder that contains the project code and run double click it A 0 E M M Q oy w O a w kej O fm Code bootstrap A zipfile to get you started with the package as package soon as possible The zipfile contains a default folder structure generated code files and the code mirror tool When starting your project you can also download the code bootstrap package That is a zipfile which contains a default folder structure generated code skeletons for the design as it stands now and the code mirror tools You typically only download the bootstrap package once Code mirror tool The code mirror tool is a small dropbox like utility that monitors lisp files and uploads them to the server when they are changed it does not make changes to your code and does not require installation Simply put it in the root folder that contains the project code and run double click it gm Code bootstrap A zipfile to get you started with the package as package soon as possible The zipfile contains a default folder structure generated code files and the code mirror tool Part Ill Code report Appendix A Download the code bootstrap package and un
216. o add the element to your design 7 Gendl Designer Demo Pieter Jan gendidesign appspot com projects 5681 80 html v HE demo pieter jan Main Diagram Consistency Design Code Settings Main diagram gt E wing Design Code Description high lift device 2 r i i z The object fuel tanks in define object wing inthe code in source wing lisp 19 doesnt have a corresponding child fuel tanks ofthe UML class wingi v Popular built ins in the design E geom base box i To resolve this problem you can add the object fu gt HE geom base gt surf gt EE gwi poddns g y2eqp 4 PAN In this case a child object was added to the code When you add it to the design with the provided link it will be added to the tree on the left You might need to unfold the tree branches of the parent class to see it It will be marked with a star to indicate it is new From the tree you can drag it into the diagram You can now continue to make design or code changes The consistency tab will show the difference and try to help you in resolving the inconsistencies discovered Part IIl Code report Appendix A Advanced features The following features are not covered in this quickstart guide You most likely won t have any trouble discovering them on your own Project management After you logged in you see all your projects From here you can manage them e You can create a new project e You can remove projects The b
217. o web based processes thereby relieving experts from supporting non experts CoScripter consists of a development tool for end users a macro recorder in a web browser and a wiki for sharing and editing the recorded web based processes The findings of this experiment were Oo Sharing took place a lot but collaborating very little Even tagging rating and commenting facilities were barely used o It was also found that adoption was higher among more technical end user developers o There is a need for early adopters critical mass The generalization of scripts to make them more widely usable is time consuming o The sharing model must be clear users can find this to be an obstacle in sharing O Part Il Initial study 53 6 2 Code review Because code review was not selected as a solution concept that will be explored in this project it will only be discussed briefly based the discussion on code reviews and inspection in McConnell 2004 and Goodliffe 2007 6 2 1 Purpose of code review The primary goal of code reviews is to increase the software quality and reduce the defect rate The review improves the quality directly by pointing out weak spots but also indirectly because code that will be reviewed will receive more care Secondary goals of code reviews are facilitating knowledge interchange facilitating learning facilitate mentoring and encouraging implicit or explicit code conventions Formal reviews in particular
218. of Engineering Automation is related to Software Engineering and Professional End User Development Finally the definition is clarified with examples 2 1 Software Engineering The IEEE Computer Society defines Software Engineering as the application of a systematic disciplined quantifiable approach to the design development operation and maintenance of software and the study of these approaches that is the application of engineering to software 15 Similar approaches might suit engineering automation developers as well and are certainly worth studying Regular engineers who develop software are not software engineers but end user developers and professional end user developers in particular One can expect that a regular engineer will encounter different problems than a software engineer will where the regular engineer lacks the skills to develop quality software efficiently the software engineer lacks an understanding of the application domain Different problems most likely require different solutions Therefore the scope of this work was limited to regular engineers and Engineering Automation was defined accordingly 2 2 End User Development In 16 End User Development is formally defined as A set of methods techniques and tools that allow users of software systems who are acting as non professional software developers at some point to create modify or extend a software artifact Examples of End User De
219. of formal reviews the defect list is used to update the checklists and to assess the effectiveness of the review process This continuous improvement and self assessment of the process is in fact what puts software organizations in the top level of the Capability Maturity Model CMM In other words it is what makes you excel in disciplined predicable high quality software development An example checklist for formal review can be found in appendix C Part Il Initial study 55 6 3 Coding policy Because coding policies were not selected as a solution concept that will be explored in this project it will only be discussed briefly based the discussion on coding policies standards and conventions in McConnell 2004 and Goodliffe 2007 6 3 1 The purpose of a coding policy Coding policies are an explicit version of best practises with respect to code presentation language use variable naming commenting etc They increase the code quality and make software development safer They make the code easier to work with and pick up protect for known hazards compensate for language weaknesses and reduce the complexity and variability in the code so that there is less to worry about The key is that any convention at all is often better than no convention The convention may be arbitrary The power of naming conventions doesn t come from the specific convention chosen but from the fact that a convention exists adding structure to the code a
220. oftware design tool Finally started to assess whether the approach works and whether it has any chance of being adopted in daily practice A working solution is one thing industry adoption another It is not new to use diagrams in software development and it certainly isn t new to design and think a solution over before implementing it What is new is the research on how to make software design feasible for engineers The first results are encouraging but further validation is necessary before strong conclusions can be drawn The main output of my thesis is the article in part It describes the problem and the solution in sufficient detail to understand what have worked on for more than a year The initial study performed and which was the basis for many of the decisions taken later is included in part Il Finally part III contains the code report with the technical details of the software design tool developed As the code report might show tried to follow the guidelines advocate to others with respect to carefully crafting documenting and testing code Whether lived up to my own standards is left to the judgment of the reader Delft The Netherlands Pieter Jan Dewitte January 2014 Acknowledgements I m most thankful to dr T van den Berg for supervising my thesis work for his support and most of all for his eagerness to think along and ahead also thank dr P Bermell Garcia of Airbus Group Innovations for supp
221. on and Javascript respectively This also reduces the maintenance cost as only 1 template needs to be maintained Choice of template language and engine The template language of choice was Jinja2 because of its intuitive yet powerful syntax and the confidence the author already had with Jinja2 from previous projects Unfortunately this template language is not available in Javascript only in Python The only template language available for both Python and Javascript that was found is the Django template language The Python implementation is well maintained and feature rich but the Javascript implementation is stuck at v0 9 misses important features is quite buggy has terrible error reporting and documentation is no longer available on line Nevertheless for code generation these issues could be worked around Because of the problems with the Javascript implementation of Django another template language was selected for javascript template tasks other than code generation Jqote2 was selected because its expressiveness equal to Javascript itself and prior experience of the author with the package Each of the languages is reviewed next I recently discovered pwt jinja2 http pythonhosted org pwt jinja2js which implements a large part of Jinja2 in Javascript I haven t tested it though It also requires compilation in Python Part IIl Code report 39 Jinja2 Jinja2 is a template language implemented in Python Th
222. on when it becomes meaningful at least to some by processing and or relating to other data It no longer exists on its own it has context What is information to one i e has context and meaning can be meaningless and thus data to someone else The aggregation of information such that it can be acted upon and used leads to knowledge This may be in the form of discovered patterns Reijnders 2012 There is ongoing debate on what constitutes Knowledge Management The definition cited by Bj rnson amp Dingsoyr is a method that simplifies the process of sharing distributing creating capturing and understanding of a company s knowledge Interesting in this definition is the inclusion of the organizational aspect While learning is often considered an individual activity it is argued in Knowledge Management that an organization can learn as well This is not only reflected in the memory of the participants but also in the institutional mechanisms of the organization policies processes A change of these mechanisms can be seen as a form of learning 2 1 2 Knowledge Management theories There is no single theory in Knowledge Management that governs knowledge its handling and reuse but rather a diverse set of theories that focus on different aspects and levels of knowledge learning and reuse The view on knowledge has changed over time from a possession that can be captured to a socially embedded phenomenon The first per
223. one of the participants maintains traceability with external documentation they created if any Some participants use comments to indicate the origin of formulas 13 A1 Motivation Incentives Attitude towards software What determines whether it is worth to develop some software What motivation and incentive is there In the case of the participants from industry 11 12 13 14 a client asks for engineering solutions and answers If software is the easiest way to get there then software is used Another reason to develop software is the desire for an automated capability e g for reduced lead time or to repeat error prone calculations 11 12 14 For Al and A2 software was inherent to their assignment as an instrument to perform research Only I3 and A1 say explicitly that they have personal motivation to write quality software Some participants also provide reasons for ot writing black box software A2 notes that doing calculations by hand allows you to spot unrealistic values which indicate there is Part Il Initial Study Appendix B something wrong with the method 13 finds that you can learn from the calculation process and doing it manually is very suited for tasks where you are less sure of what to do For these cases spreadsheets are more suited also because you can quickly make interactive software 13 The clients we work for don t typically ask for software they ask for an answer So we usually don t delive
224. ood and consolidated Detailed methodologies have been proposed to design and model Engineering Automation systems such as CommonKADS and MOKA see section 2 3 5 The high potential benefits are commonly acknowledged but multiple researchers Speel amp Aben 1998 Lovett et al 2000 Bermell Garcia 2007 van der Velden et al 2012 report that these academic frameworks are sometimes perceived as too complex or difficult to use by small research teams which lack the time and resources or willingness to obtain training in these methodologies A striking lack of systematic and rigorous testing is frequently encountered Segal 2008 Sletholt et al 2012 Four identified causes are e The developer is the user leading to an attitude where the entire usage period is considered an iterative testing and improvement period Segal 2007 e Adoption of a testing attitude where one is passively looking for incorrect behaviour rather than actively gathering evidence of correct behaviour Segal 2008 e In some cases hard to test theory independent from implementation since the implementation serves to test the theory Sletholt et al 2012 Segal 2008 e Lack of formal requirements to test against Sletholt et al 2012 Finally the traceability link between engineering knowledge such as formal and informal models and the application code is often missing Verhagen et al 2012 Adoption Software Development Methodologies and Software Process 2 2 3
225. ool and generate all code that can be generated at once to start up a project Firefox Gendi Designer Demo Pieter Jan gendidesign appspot com projects 5631 Code snippet for wing in package demo pieter jan define object wing documentation author lt name gt lt username gt lt organization gt com description input slots fuel tank volume nil todo default value computed slots O sobjects flap type high lift device Figure 4 Code snippet generated from design documentation Part Thesis article 17 4 3 Graphical notation for diagrams The graphical notation is based on the class diagram notation of the Unified Modeling Language UML 32 but was simplified to be more accessible for Engineering Automation developers and it was adjusted to fit better with KBE languages such as GenDL The notation emerged among GenDL developers at the TU Delft as a practical variation on UML and has recently been defined in 33 GenDL classes do not have attributes but have input slots computed slots and child slots instead Input slots and computed slots are shown in the diagrams where attributes would be expected Child slots are visualized as a rectangle outside the class block connected to the class with a composition link This emphasizes the tree structure of the model and provides space to display the input slot values or rules the child might have Fur
226. or supporting software development with code generation after completing the formal model Because of the diversity in target platforms for KBE systems additional information might be required before the formal model can be translated to platform specific code MOKA proposes to use a platform specific editor to adjust and complete the MOKA formal model Once complete a draft version of the KBE application can be generated This draft version can then be developed further into the final application The main critique to MOKA is that its focus is mainly on the Capture and Formalize steps two steps which are typically performed by a so called knowledge engineer Other roles such as the domain expert who provides the domain knowledge and the end user which uses the developed system to design products are not thoroughly considered There is also no clear strategy for the maintenance and reuse of knowledge Verhagen et al 2012 A critique especially relevant to Engineering Automation stems from the projected use of code generation to support application development Applying code generation followed by manual modification as proposed in Stokes 2001 conflicts with the iterative nature of the application lifecycle and software development in general Further elaboration on the approach is needed to avoid losing manual changes when te generating code in a next iteration Finally it is observed that some designers find it difficult to use the structur
227. org TR rdf primer World Wide Web Consortium W3C SPARQL Specification v1 1 2013 Online Available http www w3 org TR 2013 REC sparql11 overview 20130321 Part Thesis article 39 PART II INITIAL STUDY Development and Reuse of Engineering Automation Software Literature and interviews P J A R Dewitte B Sc December 3 2013 Literature Review and Interviews Development and Reuse of Engineering Automation Software Pieter Jan Dewitte Abstract Increasingly engineers create software to support their daily engineering activities This self written software referred to as Engineering Automation helps them to solve similar problems faster by automating parts of their engineering work It is desirable to reuse that software for similar applications and share that software with colleagues rather than re creating it over and over again developing Engineering Automation software requires time and resoutces Unfortunately this literature review confirmed the general perception that the level of sharing and reuse of Engineering Automation software is low Engineering Automation applications tend to be non transparent black boxes of varying quality This makes them hard to adapt them to the constantly evolving knowledge in engineering work The underlying cause for the lack of reuse is the incentive scheme for Engineering Automation which directs the efforts to providing answers to engineering questions Any act
228. ormatics 22 4 pp 445 456 Gessenharter D amp Rauscher M 2011 Code Generation for UML 2 Activity Diagrams Towards a Comprehensive Model Driven Development Approach pp 205 220 Goodliffe P 2007 Code craft the practice of writing excellent code No Starch Pr Green T R G amp Petre M 1996 Usability analysis of visual programming environments a cognitive dimensions framework Journal of Visual Languages C Computing 7 2 pp 131 174 Herbsleb J et al 1997 Software Quality and the Capability Maturity model Communications of the ACM 40 6 pp 30 40 Part Il Initial study 59 Howison J amp Herbsleb J 2011 Scientific software production incentives and collaboration In Proceedings of the ACM 2011 conference on Computer supported cooperative work New York NY USA ACM pp 513 522 ISO TEC 15504 1998 Information Technology softwate process assessment Kelly D F 2007 A software chasm Software engineering and scientific computing Software IEEE 24 6 pp 119 120 Kelly S amp Tolvanen J P 2008 Domain specific modeling enabling full code generation Wiley Klein R 2000 Knowledge Modeling In Design The MOKA Framework Proc Artificial Intelligence in Design 00 pp 77 102 Lauder M et al 2010 Model driven systems engineering state of the art and research challenges Bulletin of the Polish Academy of Sciences Technical Sciences 58 3 Leshed G et al 2008 CoScr
229. orting this research by all means he had available I d like to thank Ir D Steenhuizen for his feedback on the draft of this paper This work would not have been possible without the interview participants and test users who devoted some of their time to me the engineers at Airbus Group Innovations and students at the TU Delft Last but not least I m most indebted to my family for their love and support in all past 24 years of my life Delft The Netherlands Pieter Jan Dewitte January 2014 Summary Increasingly engineers in for example Aerospace Engineering create software to support their daily engineering activities This is referred to as Engineering Automation A prime example of Engineering Automation is Knowledge Based Engineering It is desirable to reuse and share this software rather than to discard it soon after its creation Unfortunately the overall level of sharing and reuse in daily engineering automation practice is currently low Producing reusable applications proves to be difficult for engineering automation developers An initial study comprising a literature review and expert interviews showed that the two main issues are the understandability and validity of the software and documentation The study also provided insight in the current Engineering Automation culture The most important aspect identified is the lack of incentives for software activities other than coding itself Based on the initial study
230. ortunately Basili et al Basili amp Rombach 1991 observed that reuse is less institutionalized in Software Engineering than in any other engineering discipline This in contrast to the common opinion that reuse should be considered early on e g during system design Blaha amp Rumbaugh 2005 Part Il Initial study 7 2 3 Engineering Automation 2 3 1 Introduction Engineering Automation distinguishes itself from mainstream Software Engineering by authorship it is software implemented by regular engineers rather than software engineers It therefore falls within the domain of Professional End User Development The scope of this work excludes software developed by software engineers because due to their different background software engineers and regular engineers are likely to encounter different problems and need different solutions This section starts with a discussion of End User Development and Professional End User Development Next Design Automation and Knowledge Based Engineering are discussed two areas within Professional End User Development specific to engineering The section is concluded with a discussion on the adoption of Software Engineering practises in Engineering Automation 2 3 2 End User Development End User Development is a subfield of Software Engineering which researches the possibility for users to develop further a software system This is in contrast to mainstream Software Engineering where it is assu
231. osed of data from the datastore e data not stored in the datastore because maintaining that data is costly and the system can continue gracefully without the data should it be evicted from the memcache This section gives an overview of what is stored in the memcache Memcached models User Team Project FileGroup These models are stored in the memcache entirely This means that their properties are available in the cache not their children Part Ill Code report 44 Project Aggregated ModelEntities The complete design model is required for the project tree in the web client and to compare the design to the code in the consistency checker Rather than requesting all entities from the database over and over again an ageregated data structure is stored in the memcache Project Last design and code change timestamps The system keeps a timestamp of the last change to the design and the last change to the code This has two purposes e With this information the system can detect switching from design sessions to code sessions or vice versa This is important for the logging of the consistency evolution only at switches we need a consistency snapshot More snapshots are not practical and do not add much value It would mean that a new snapshot is taken every time the user saved a file or adjusted a diagram e The system can show to the user when the last recorded change was With this information he can assure himself that his
232. other files contain 1 class and are not a suited place for generally used utility functions class or function name lisp A lisp file containing one class define object temp An empty directory tests An empty directory README txt A readme file template file monitor exe The monitor exe for the given project and filegroup monitor info txt Information about the monitor FAQ etc At the time of writing the svg diagrams and the off line viewer in documentation are not included yet They can be viewed online Part IIl Code report Appendix B Appendix C Adding a style to ideaDiagram To add a style you need to modify the ideaDiagram library itself A style tells the system how to render data For data that is no more than a name usually a box is all the style really is For more complex data such as a class diagram element rendering the data is not that simple and more logic is required In a nutshell Add the style to styling js Add value information to value js Add render logic in creating main js Add the style to styling js Add an entry to the style object Create a cell style definition in setup_styles and put it in the stylesheet with putCellStyleQ See the docs of mxGraph for what styling can be applied Modify the rest of styling js as needed should be self explanatory Add value information to value js Add default data to default_values needed if new empty objects w
233. ound is highly iterative and interactive There is no explicit process however and in practice the development process is better labeled as ad hoc than agile 3 4 2 Why engineering automation software is developed the way it is Literature The prime explanation given in literature for the ad hoc development is tension between research goals and software engineering goals attention tends to shift to the first The software is in the first place considered as a research tool to address immediate research needs i e short term goals 6 20 21 The underlying problem is an incentive problem academics are rewarded for publications which lead to funding engineers in industry for the output of their software for both the software work itself is normally not rewarded 6 20 Furthermore literature offers explanations for several aspects of engineering automation development Part Thesis article 7 The exploratory nature of research and design makes requirements elicitation difficult It is argued that because an engineering automation developer is both the developer and the domain expert this is partially compensated for 19 The available methodologies for designing and modeling engineering automation software such as CommonKADS 22 and MOKA 23 are perceived as too difficult and complex especially for small teams 24 27 Several explanations were found to explain the lack of systematic testing Since the develope
234. pare Name n Eq Num return 1 Return value Num return fib n 1 fib n 2 If Compare Name n Eq Num 1 Return value Num 1 Return BinOp Call Name fib BinOp Name n Sub Num 1 l Add Call Name fib Bin0p Name n Sub Num 2 Figure 5 6 example of source code and an equivalent AST 5 3 4 Parsing reverse engineering The first step of reverse engineering lower level code text to a higher level model is parsing the code into an AST which should then be processed into the higher level model The scientific foundation of software languages and parsers for them is well established Patsers ate categorized according to the complexity of the language to be parsed regular context free context sensitive or unrestricted in increasing order of complexity The existence of Part Il Initial study 41 a reasonably efficient parser is guaranteed for languages in the first two classes Regular languages have a fundamental limitation they cannot handle nested structures In practise most parsers recognize context free languages and the additional complexity of the language beyond context free must be handled after parsing That is the parser will accept the program but subsequent processing might find out that the program wasn t valid after all Wachsmuth 2011 To reduce the effort required to build a parser several parser generators have b
235. per is the user so that the entire usage period is an iterative testing and improvement period It is not always known what the output should be when testing rather the software is run and it is checked whether the output is sensible In some cases hard to test theory independent from implementation since the implementation serves to test the theory Defining a test is difficult because there is no formal list of what has to be tested e Do you use one or more formal methodologies when developing Engineering Automation software development methodologies knowledge engineering methodologies o Do you and to what extend do you use these Incremental development Clear iterations with clear goals A prioritized TODO list with features and bugs Small sub task descriptions with example inputs and outputs Coding standards policies conventions Are they made available in the first place e Do you find yourself effective in producing software Effective in the sense that you can get it to the level you find sufficient within the time you find reasonable e What do you find most time consuming in software development Documentation 5 min e To what extend and on which levels do you document your software e If relevant How do you manage the relations traceability between code documentation design documents and other software artefacts Motivation Incentives Attitude towards software 10 min e
236. pirations Segal 2007 Support mechanisms for collaboration and reuse along both the technical and social axis seem to be missing On the technical axis documentation is insufficient due to several causes Besides the low priority of documentation Segal Segal 2007 Segal 2008 notes that it is very common that software is developed adhoc and passed on from researcher to researcher resulting in problematic software artefacts Also code comprehensibility does not appear to be a significant issue for professional end user developers The ad hoc nature of development and insufficient documentation and traceability are also reported by Verhagen et al 2012 as major issues which make reuse hard On the social axis reuse through communities of practise is hampered due to their instability Segal 2007 the desire to move on to new research positions combined with other factors such as graduating students can cause a high turnover of developers and the associated knowledge loss Willingness to participate in communities of practise is usually present though This discussion on the adoption of Software Engineering practises in Engineering Automation software development is concluded with two interesting remarks Howison amp Herbsleb 2011 note that the Software Engineering community proposes the scientific software community to adopt techniques without encouraging understanding of what is the cause of the problem they are trying to fix Segal
237. ple for the lexer is given below An integer is recognized and converted from string to integer for further processing later on def t_INT t EV I 97 O 9 t value int t value The argument t is a token with a value and a type The value is by default the recognized string the type is derived from the name of the function Both can be altered An example for the parser which recognizes and re concatenates qualified symbols like a b c is given below def p_qualified_symbol_doublecolon_symbol p qualified_symbol qualified_symbol DOUBLECOLON SYMBOL p O p l tig piz 39 99 The argument p is a list The first element should be set to the processed value of the symbol recognized The subsequent elements contain the recognized tokens and symbols in the same order as in the right hand side of the grammar rule The token values in the list are the values as processed in the lexer functions The symbol values in the list are the result of other earlier applications of the grammar rules In the example p 1 is such an earlier result i e what was earlier assigned to p 0 in an earlier reduction Note the naming convention where tokens are written with capitals and symbols in lower case From the grammar description the parser can be generated In fact PLY generates parser tables which are fed into a generic parser system a bit like discs you have to put in a game console PLY generates the tabl
238. ply The most stressed attribute is functionality This is both about what the software does and how well it does it Example cases test cases for the whole program also rank very Part Il Initial Study Appendix B high all participants indicate the need for them Functionality and example input output ate the main considerations to determine whether the software could serve the intended goal The availability of support is invariably appreciated but not considered mandatory References to papers or external documents explaining the methods and engineering knowledge used in the code are desired by most but considered less important by others The same applies to the quality and complexity of the code Surprisingly the availability of documentation is considered fairly unimportant e A1 Problems when reusing another tool e Difficult to understand e Not programmed very well e Not documented clearly e It turned out that for the design studies I wanted to perform the requirements on aerodynamics were stricter than what the original tool was designed for e It turned out that some things just weren t right There were many unacceptable assumptions So the program runs and there is a result but it s just not right Actual Reuse What determines how well you can work with someone else s source code Both characteristics of the source code itself and other factors The answer varies largely among participants Mention
239. r http lt server gt assets All other URLs are routed to server_loader app server_loader py is the Python file next to app yam1 which loads the Webapp2 app from the serverapp sources server_loader py is discussed in the next section All files in the codeclient folder are uploaded to the static file servers of GAE All other files are uploaded to the application servers of GAE except the ones that match a skip files rule Each rule is a regular expression Currently the following files are skipped e Yaml deployment configuration files e Automatically generated backup files compiled Python files e Files in a folder starting with a dot e g hg the version control system Test documentation inspiration playground and _disabled folders e The Dojo files which for which we use a CDN The development files of mxGraph see web client chapter The code client is actually skipped automatically because it is deployed to the static file servers not the application servers of GAE e experiment folder which contains off line project tools To test the regular expressions a test script has been written which lists for each file whether or not the file is skipped test_app_yaml_skip_files regex py app yaml also enables the appstats built in functionality appstats records the usage of Remote Procedure Calls RPC such as datastore and memcache calls and displays the statistics when visiting http lt server g
240. r understandability and maintainability GenDL Designer can also be used as a documentation tool and as a learning tool Documentation can be created for existing GenDL code Novice GenDL developers can benefit from seeing how design elements map to GenDL code The typical workflow when using GenDL Designer starts with drawing the high level structure of a GenDL application in UML like diagrams in the web based application From these diagrams code can be generated and downloaded which then must be completed with details not in the diagrams The on line diagrams and the local code can be modified independently A small upload utility uploads local code files as they change The web based application continuously compares the diagrams to the code lists all inconsistencies and helps to resolve them This code report is written with three readers in mind system users system administrators and system developers This report is divided in three sections which correspond to the three readers system usage system administration and system development The sections built on top of each other Before skipping to a next section it is recommended to scan the preceding sections Part IIl Code report 2 Section System Usage Part IIl Code report 3 1 Introduction The first section of this report describes the system from the perspective of a user First of all what is GenDL Designer for Basically GenDL Designer helps you with designing
241. r is also the user using the system is considered testing too Other contributing factors mentioned are having to test both theory and implementation at the same time 5 19 and the lack of formal requirements 19 Agile software development methods fit well with scientific software development because they share an emphasis on responsiveness to change on collaboration and on an iterative nature 5 19 Interestingly J Howison and J Herbsleb 6 note that the software engineering community proposes the scientific software community to adopt techniques but without encouraging that community to understand what is the cause of the problem these techniques are trying to fix This might explain the limited adoption of software engineering techniques Interviews The interviews show the tension described in literature all software work except the coding itself has to compete with more urgent work One participant said I would like to describe the engineering knowledge but generally there is no time The codes that are written are just there to do the job get a value out Clients and supervisors ask for engineering solutions and answers not software The software and along with it software engineering practices receive the minimal amount of attention and enthusiasm for devoting more resources is generally low Best practices are not picked up due to a lack of training One participant said We wouldn t even talk abou
242. r software to them we develop software to get the answer A1 My supervisor even doesn t look to my code When I received the tool my supervisors assumed that it worked and that I could simply extend it Supervisor after finding out that wasn t the case Oh well then I must have read a very good report A1 again I know that what is in my report corresponds to the code A2 Pm quite sure my supervisor never looked at the code of me or my predecessor What I do in the code doesn t matter to anyone so I don t have a reason to document or write really good code A2 It is very normal to read each others thesis but is very strange to read each others code While 90 of your time you spent on the code which no one will have a look at 14 e Developing software reduces the lead time of the task You can reduce the time to complete a task from weeks to hours That is the key driver e You can also increase iteration when you decrease lead time it gives you a chance to try different things e Finally automating the task gives you an opportunity to integrate with other tools The function your tool performs is then one link in a chain of functions In literature some common views of engineers on software are described To what extend do you recognize these O Software is a research tool which is developed in the first place to get some output needed by the research at hand First you must get it work
243. ration i e project URL on the fly see section 6 3 6 Part IIl Code report 67 9 Ideas and directions for improvements During the development choices were made regarding which features to implement Features that weren t necessary for the research and required significant effort to implement were left out Yet some of them would be valuable features useful if GenDL Designer moves into the direction of a commercial grade product They are presented here 9 1 Support other languages than GenDL Since GenDL is a niche language supporting multiple languages would greatly increase the usefulness of GenDL Designer This is not too difficult given that the other language has similar semantics to GenDL In practise if the current design model fits with the new language it is sufficient to replace the code generation and code parsing components The code generation components are templates and therefore fairly straightforward to replace Code parsing might be more difficult The availability of an existing parser would help Afterwards the general parser output has to be processed to the class data format and the function data format used by GenDL Designer If the designer part of GenDL Designer has to be adjusted as well adding the new language might prove to be more difficult In that case GenDL Designer will be more an adjustable example of how to implement a design tool 9 2 Export diagram as SVG Currently the only way to export
244. ratively and incrementally without a formal method Requirements elicitation software design testing and documenting are largely skipped Part Thesis article 10 Rather than thinking their software solution through first engineering automation developers quickly start coding and keep tweaking the code until the overall application output fits with their expectation For any software system of realistic size this is not the fastest way to develop software nor can the software be expected to work correctly for inputs other than the ones manually checked 28 In fact giving sufficient thought to a problem before embarking on a solution and spending sufficient resources on systematic validation are two general guidelines applicable to any engineering discipline The lack of systematic testing is particularly problematic the code is fragile when introducing changes or even downright incorrect 3 5 2 Why is engineering automation software developed the way it is Underlying engineering automation development is the need to provide answers not software Any activity that is not part of getting the answer is perceived as time consuming It was found that this applies to software design documenting testing systematically integrating with existing software and supporting reusing developers but also to software engineering training 3 5 3 Howis engineering automation software currently shared and reused The overall level of sharing an
245. re Microsoft Windows APIs These are documented on http msdn microsoft com libra In fact the purpose of the module is to take care of the cryptic details and provide a high level interface to create system tray applications The menus to display on right or left click are described with nested MenuItem instances Sub menus are supported Part Ill Code report 66 8 4 Building the executable 8 4 1 Introduction The code client is preferably an executable that just runs if the user doesn t need to install and or configure it the barrier to using the code client is lower In practice this means that the Python code needs to be turned into an executable and the executable has to be pre configured for the project for which it is downloaded This is done with a configuration file This file contains the URL of the project It is not possible to build the executable on the fly on the Google App Engine server with PyInstaller the executable has to be compiled on the executable target platform In addition it would take costly server resources to build the executable over and over again Generating the pre configured executable is therefore done in two steps First an un configured executable i e with an empty configuration file is generated locally and deployed to Google App Engine along with the rest of the application Then an executable with the right configuration file is derived from it on the server 8 4 2 Building the
246. reas tacit knowledge is only available in the expert s head and he might even not be aware of its existence Part Il Initial study 5 2 2 Software Engineering 2 2 1 Introduction The IEEE Computer Society defines Software Engineering as the application of a systematic disciplined quantifiable approach to the design development operation and maintenance of software and the study of these approaches that is the application of engineering to software Abran amp Moore 2004 Software Engineering principles apply to all software including Engineering Automation software Software Engineering emerged to deal with the increasing complexity of computer programs software development needed to become a disciplined engineering practise Software Engineering can be seen as the application of Systems Engineering to Software Development Studer et al 1998 van den Berg et al 2011 Software Engineering is a broad field The current discussion is limited to sketching a basic framework to discuss Engineering Automation later in this report 2 2 2 Software requirements design and testing As with Systems Engineering a Software Engineering project should start by establishing the project objectives and requirements Then the system can be designed at an increasingly finer level of detail The use of modelling languages such as the Unified Modelling Language is very common As in any engineering discipline testing should be performed
247. requirments document source code file X Objective and scope X X X Context e g origin X X Required background knowledge X X Description of the process X Detailed validity application domain assumptions etc to help others X X X to assess the applicability to their problem Use X Input description X X Example output for example input X X Training materials e g tutorials X Test cases X Quality level of readability correctness defect detection rate user X friendliness Information on how and how easy the package can evolve Related X to quality Contact details sources of help X X Feedback method X X Alternative ways of using the application X Interfaces with other packages X Dependencies and related documents X Decision history what worked and what not X Solution domain what SE method was applied X Scope at what level of the architecture and in which stage of the X development is the object used Table 6 1 Proposed meta data to describe reuse candidates The meta data most often mentioned is o Objective and scope 3 o Detailed validity application domain assumptions etc to help others to assess the applicability to their problem 3 o Context 2 o Required background knowledge 2 Part Il Initial study 52 Input description 2 Example output 2 Contact details sources of help 2 Feedback method 2 Information on how and how easy the package can evolve 1 5
248. route has an associated request handler class Hence all similar resources will match the same request handler class thanks to their similar URL e The request handler class has the methods get put post and or delete to manipulate the resource referred to by the URL read create write and or delete data respectively loosely speaking The body of handlers is usually quite short even though they contain the main application logic This is because the heavy lifting is moved into generic reusable components For example get methods typically request data based on the URL and output the data in the requested format Getting the data is done very easily with the interface of the datamodel package discussed in the previous section Formatting the data is made convenient with the generic handler types discussed below Because the handlers are so concise they were easy to make self documenting with only a few comments you simply read what they do when reading the source code The handlers are organized in sub packages according to the route tree Apart from three more clarifications the source code speaks for itself generic handler types access control and password handling are discussed below Generic handler types Two custom RequestHandler classes are provided in server_app request_handler to subclass from These handle two common situations WebRequestHandler formats data with a Jinja2 template and is used for routes that end with
249. rs for X_data in in_design_only inconsistencies append X in_plan_only X name X_data name X data X_data Part Ill Code report 49 for X_data in in_code_only if IMPLEMENTATION_DETAIL in X_data description continue inconsistencies append X in_implementation_only X name X_data name X data X_data file function_data file line function_data line for X_in_design X_in_code in in_common compare_description X_in_code name X_in_code X_in_design X_in_code inconsistencies warnings errors Do more checks here The compare functions do not return values they simply append values to inconsistencies errors and warnings lists Integrity checking Integrity checking is an additional tool to use during development and testing to check whether the design model is still valid This concerns the structure of the data of ModelEntities but also the reference they keep to each other such as class child relations The tool can also fix errors should they occur The tool is accessible from the admin panel in the web interface This is only for development and testing It allows quick fixing of the database after fixing a bug that messed up the data so that a new test can be conducted It is not intended for production use data might be lost In production the design model simply should not get corrupted and if it does it mandates serious investigat
250. rs and peer interaction There will however be a need to enforce the policy if the policy prescribes activities such as testing which do introduce extra work This requires organizational change and is the weak point of this solution concept The solution concept has no conflict with the limited amount of software engineering training or with an iterative development style 4 3 Solution concept selection The selection process rates the solution concepts with respect to potential to improve reusability and applies penalties to solutions which conflict with the non technological obstacles Each of the solution concepts is rated twice on a scale of 1 to 5 for its potential to improve reuse once for understandability and once for validation A rating of 1 corresponds to hardly any potential while 5 corresponds to very high potential This gives a maximum score of 10 A penalty of 4 points is applied per conflict with an obstacle Each conflict makes the introduction of the solution concept harder since some organization or culture change will be necessary This is undesirable due to the short duration of this master thesis project and because the solution will be adopted less widely when more change is required All ratings penalties and scores are shown in Table 4 1 Part Il Initial study 31 Solution Reuse improvement Side Justify Limited Iterative Score concept potential activities directly training style Un
251. s There are basically two ways to handle dependencies They can be copied into the code base and treated as if it is regular code for the project That includes putting them under version control The other way is to only keep a list of the dependencies and the required version numbers under version control The dependencies then have to be fetched separately from the source code Luckily tools exist to automate this The advantage of the first approach is the simplicity you get the code from one place in one operation The advantage of the second approach is that if the dependencies are fetched automatically upgrading to a new version of the dependency is as easy as changing the required version number The code base currently follows the first approach Part IIl Code report 27 For future development it would be wise to investigate the possibilities to switch to the second approach There are however several difficulties e There are dependencies for Python and Javascript Most likely there will be two dependency managers needed Python has the pip system organized centrally by the Python community Javascript doesn t have such a central community Part IIl Code report 28 4 Data models Knowing and understanding the data models that are being manipulated is vital to understanding the application logic of a program Also data is what is exchanged between subsystems The interaction between subsystems can only be under
252. s found to be a universal panacea Every notation highlights some information at the cost of obscuring other information Blackwell 2006 As an illustration of the advantages and drawbacks that graphical and textual representations each have consider the LabVIEW model and equivalent BASIC program in Figure 5 2 and Figure 5 3 respectively which were the subject of a case study Green amp Petre 1996 Blackwell amp Green 2003 Part Il Initial study 35 Distance travelled ET 400000 yer Awe oe F Figure 5 2 Example LABVIEW model Source Blackwell amp Green 2003 Mass 10000 Fuel 50 Force 400000 Gravity 32 WHILE Vdist gt 0 IF Tim 11 THEN Angle 3941 IF Tim 100 THEN Force 0 ELSE Mass Mass Fuel Vaccel Force COS Angle Mass Gravity Vveloc Vveloc Vaccel vdist Vdist Vveloc Haccel Force SIN Angle Mass Hveloc Hveloc Haccel Hdist Hdist Hveloc PRINT Tim Vdist Hdist Tim Tim 1 Figure 5 3 Equivalent BASIC program of the LABVIEW model in Figure 5 2 Source Blackwell amp Green 2003 The case study illustrates how graphical notations excel in their ability to show dependency links and data flows but suffer from the clutter and layouting effort that these links introduce The researchers found that due to the jiggling with boxes and lines it took 8 times longer to Part Il Initial study 36 make the same change in the LabVIEW model than it took
253. s by clicking the logs button in the Google App Engine Launcher For live GAE applications you can inspect the log messages on https appengine google com gt your app gt logs Set the minimum severity to Error to see only error messages Every error indicates a problem The system might have recovered though and users might or might not have noticed Messages at debug info and warning level can usually be ignored 4 4 Available logs To see to what extent the software is used how it is used and which features are used scripts have been created that download process and analyze log files There are three sorts of logs for GenDL Designer Request logs Application logs Consistency logs The request logs are standard Apache server like log files which record every request that has been made to the server Because nearly every action in the web application or on the local lisp files triggers a web request and because the URLs have been chosen such that they are meaningful much information can be mined from the log files For example PUT http localhost 8080 projects 4967730973245440 design data 1383438891255 1 json origin diagram js 3Aevent_CELL_ADDED Part IIl Code report 19 This request indicates that in project 4967730973245449 design data entity 1383438891255 1 has to be created The webclient has sent along an extra piece of information that indicates that command origin is the diagram js module which reacted to a
254. s hardly or not shared across teams According to 13 there is little to no interest of other teams to reuse his code Inside teams legacy code is available personal or from colleagues The academic participants A1 A2 their project was to continue with given software Code is re used by copying code rather than by creating a shared library This seems to be because there is not enough discipline to keep a shared library working code gets moved or other s their changes break the code for you 13 A1 A2 14 used to work in a KBE team where code was shared intensively including collective ownership of some parts of the code This isn t the case anymore 13 admits that he isn t keen on sharing his code outside the team then he would need to provide support and need to clean up more his code This interferes with the regular activities for which I3 is contracted o A1 a shared repository introduces the risk that someone else breaks your code When things like that happen you quickly loose interest in pulling changes from others and rather keep working on your own version What material is typically available to learn about how the software performs its task The participant from industry 11 12 I3 14 usually only have the code I1 and I3 indicate that the lack of requirements design documentation etc is not a problem because you can ask I4 indicates that documentation is used when it is available but it is no problem to reverse
255. s only an incentive for answers Software Engineering experience is limited and the development is highly iterative and incremental Four solution concepts were evaluated for their potential to improve the level of reuse a graphical software design tool an app repository with quality metrics code reviews and code policies The evaluation graded the solution concepts with respect to the two key issues understandability and validity and checked the alignment with the non technological obstacles identified earlier Eventually the graphical software design tool was selected as the subject of the experiment to be conducted Finally the graphical design tool concept was reviewed in depth Literature about graphical programming notations and code synchronisation provided relevant theoretical foundations and practical advice State of the art tools were tested and or reviewed for even more practical insights The most important recommendations for the prototype of the design tool are generate both code and design documentation incrementally and iteratively use a customized UML notation without unnecessary detail provide soft consistency warnings and generate and parse code with industry standard techniques The next step of the research is to apply these recommendations in a prototype of a graphical design tool and evaluate the prototype with an experiment This will establish the feasibility and effectiveness of the approach and provide further ins
256. s unfamiliar with the Software Engineering practises it encourages through proper explanations in the user interface and the concept doesn t restrict iterative development approaches 4 2 3 Code review Code review Goodliffe 2007 McConnell 2004 refers to the regular review of a part or all of the code and the related documents by one or more persons other than the author The review is supposed to reveal defects and maintenance issues in the code such as low understandability Code reviews are a universally acknowledged technique and have been around since people punched their programs into stacks of cards Reviews encourage best practises and discourage practises which lead to maintenance nightmares Besides increasing the software quality and reducing the defect rate of the code code reviews are also useful from a teamwork perspective knowledge about particular pieces of code as well as development best practises is exchanged Code review has a very high potential to improve reuse Knowing that he or she will have to explain his work gives the developer a reason to keep the understandability of the software high and the validation trustworthy The review itself will point out where this failed so improvements can be enforced afterwards Code review does unfortunately conflict with two non technological obstacles Engineering automation developers see code review as an extra non essential activity even though it does Part Il Initial s
257. scribes states and transitions between states of objects in isolation It consists of state diagrams The interaction model describes the communication between multiple objects and or users It consists of use case diagrams which describe the interaction of a user with the system sequence diagrams which describe the communication between objects and activity diagrams which show the workflow performed by the system and optionally the object data flow and state information Blaha amp Rumbaugh 2005 Part Il Initial study 38 speed altitude setEnginesFullThrotle setEnginesidle 1 engines LiftingSurface airfoil span calculateDrag in speed calculateLift in speed fuelTankVolume if engineSetting deflection calculateThrust in speed in temperature Figure 5 5 Example of a UML class diagram Two critiques of UML relevant for a graphical software design tool were found Firstly UML doesn t raise the level of abstraction compared to Object Oriented languages most of the UML language maps one to one to software constructs Instead Domain Specific Modelling must be used to raise the level of abstraction and obtain the associated development performance gains Angyal et al 2008 This is in line with the position of Kelly 2007 stated before professional end user developers need tools that take into account their context rather than domain independent tools from
258. scription SRR me ee The UML attribute 1ift ofthe UML class lifting device inthe design doesnt have a corresponding slot 1ift in define object lifting device inthe code in source lifting device lisp 3 T in lifting device in the code code snippet or remove the attribute lift from lifting device in any diagra gt The UML method xfoil analysis ofthe UML class lifting device inthe design doesnt have a corresponding function xfoil analysis in define object lifting device inthe code in source lifting device lisp 3 blem you can define the function xfoil analysis in lifting device in the code code snippet or remove the method xfoil analysis fron n any diagra A B wing has different documentation in the design and in the code in source wing lisp 3 Design Code Figure 3 Screenshot of the consistency pane 4 2 User interface Figure 2 shows the main interface of GenDL Designer Most prominent is the drawing canvas for the diagrams Tabs allow switching between diagram panes a consistency pane and a settings pane The project tree which contains an overview of all elements in the design Part Thesis article 16 diagrams remains visible at all times Furthermore there is a help center and a feedback button On the drawing canvas diagrams can be created One project can contain multiple diagrams to deal with the complexity of large projects Multiple diagrams can contain the same element so that one diagram can def
259. se can increase hidden dependencies visibili ty complex relationship Figure 5 1 typical trade offs Source Blackwell amp Green 2003 5 2 2 Limitations of graphical notations Based on the intuitiveness of graphical notations it is tempting to think that a graphical notation could be created that is superior to any existing textual notations This thought has been criticised from three points of view The first criticism stems from Software Engineering In his paper No Silver Bullet Brooks 1987 argues that the hard part of building software is the specification design and testing of this conceptual construct not the labor of representing it and testing the fidelity of the representation This puts in perspective the effect a graphical representation can have if the only feature it adds is that it is graphical The second criticism targets the overeager and over detailed use of diagrams Multiple authors warn for death by detail disappointing and ineffective results when using an inappropriate level of detail when drawing diagrams Formality is of less importance than the communication aspect McConnell 2004 Goodliffe 2007 Blackwell 2006 The third criticism is rooted in psychology and criticises the belief that there exists some notation that is universally best in its domain whatever the activity may be Psychological experiments have shown that some visual languages are better for certain purposes but none wa
260. se study and lessons learned In 27s IEEE ACM International Conference on Automated Software Engineering ASE 06 pp 145 156 Niazi M Wilson D amp Zowghi D 2005 A maturity model for the implementation of software process improvement an empirical study Journal of Systems and Software 74 2 pp 155 172 Nonaka I amp Takeuchi H 1995 The Knowledge Creating Company How Japanese Companies Create the Dynamics of Innovation How Japanese Companies Create the Dynamics of Innovation Oxford university press Oldham K amp Kneebone S 1998 MOKA A Methodology and tools Oriented to Knowledge based engineering Applications Proceedings of the pp 1 10 Paulk M C et al 1993 Capability maturity model version 1 1 IEEE Software 10 4 pp 18 27 Paulk M C Konrad M D amp Garcia S M 1995 CMM versus SPICE architectures IEEE Computer Society Technical Council on Software Engineering Software Process Newsletter 3 pp 7 11 Peckham J amp MacKellar B 2001 Generating code for engineering design systems using software patterns Artificial Intelligence in Engineering 15 2 pp 219 226 Reijnders A W 2012 Integrating Knowledge Management and Knowledge Based Engineering La Rocca G 2011 Knowledge Based Engineering Techniques to Support Aircraft Design and Optimization Delft University of Technology Segal J 2008 Models of scientific software development First International Workshop on So
261. seesseees aie oboe abe side we dS PA Wait OAM CHO Myo cies scciceisacesecesseecastsastessds stuns sesedseassceadetsacuassastisonsesa eds AAA ANE hives casas dhatsdueos iN aAA S RA ENNE 53 72 Technologies used rarnana an e NE AE GA I IA dee barn en baad Bek anit 7 3 Components oe S ss ses ere er an aes ae TAs ACtivity GAS tas sxc essa sisareds RER N R lactis Joa Joel sir 8 CODE CLIENT SUBSYSTEM 8 1 Introduction 8 2 Technologies used 8 3 Components ccecceceeeeees 8 4 Building the executable 9 IDEAS AND DIRECTIONS FOR IMPROVEMENTS ooien aa E E EEA EA A EA 68 9 1 Support other languages than GenDL pa ssia a as ods fats 927 Eixporttidiagran as SNG errietara en e o A deat cesdvayaventecqsite REEERE EEE EE 9 3 Off ined cumentatio M sresnsisascspcarosacogasocecra ancicn aana aana anaa 9 4 Version control Support sssssssssssssssssssee 9 5 Cut copy paste and undo in diagrams 9 6 Suppress selected consistency notifications bly i ies a is sets 9 7 Drag Scidtop fot slots ssccciseciklavsleasapeanscssesdoleaduastehuavactenedetssvecagnsinsatunsdenedubcteauisustannipiascecdyssbeundghsishaivesteesdeosetansyiais 9 8 Dynamic classtype MOtatlOn aies ic5 5 ssnssvectssesccetsannnsatvacsasedevesoneseninnastantacselvecosesssinignsiengsosetsttsountentansateassssetovaceantetead 9 9 Automatic layouting and mass import 9 10 Assien colours tO Classes ar E EE EKE EE E ESE E E tes 9 11 Filter text field for the project tte ccccccccsssess
262. sharing based on reputation is already in use for scientific software publications lead to reputation which leads to funding This model appears to be working for software that falls under it while software work that does not fall under it such as maintenance work was found to be under resourced leading to low levels of sharing and collaboration Howison amp Herbsleb 2011 From the knowledge consumer side the executable nature of apps gives an additional means to test validate compare and evaluate the knowledge 4 One can expect that a repository shared between companies requires a financial compensation system for the intellectual property exposed through the apps This is beyond the scope of the current research and this report Part Il Initial study 29 4 2 2 3 Participation costs Compared to regular knowledge repositories an engineering app store has 2 opportunities to reduce the costs of participation First the cost of describing how to apply the knowledge in addition to the knowledge itself is reduced because the Engineering Automation software already describes this to a large extent if proper attention is given to the understandability Second the documentation required in an engineering app store can partially be created as a side product of the software development The graphical design tool is an example of this In such a situation you document for yourself i e create a design with immediate benefits i e cod
263. sic software engineering practices to deal with this complexity such as requirements elicitation software design testing and documenting These practices would help them to implement their software correctly and in the long term faster 5 8 9 Part Thesis article 1 This report addresses the feasibility of introducing software design and design documentation activities into the engineering automation development process A software design is here understood as the high level structure of an application The software design hides implementation details so that the developer can focus on larger issues and understand the system as a whole Creating a software design has several advantages If done before implementing the high level view allows reflection to improve the application structure before it is actually implemented Reflecting improves the quality of the software and prevents costly corrections later on During implementation the design guides the process After implementation the software design is documentation which helps developers to understand the software quicker This makes it easier to modify the software later 10 12 A support system GenDL Designer was developed to make the software design activity as feasible as possible for Engineering Automation developers that use the GenDL framework 13 an open source knowledge based engineering system GenDL Designer is based on incremental code and design docume
264. silo raaa Eesaia TARS En Ee En SS 5 3 2 Mapping between models of higher and lower abstraction 5 3 3 Code generation forward engineering 3 4 Parsing FEVELSELEMPINESLIND oiei inae aE E EE E AE E E E E E O Part Il Initial study 5 3 5 Iteration and round tripping 573 6 Testing iisivsesievsesencartuccnsestesescestein sits at ap 5 4 STATE OF THE ART UML MODELLING TOOLS WITH ROUND TRIP FUNCTIONALITY 5 5 CRITICAL NOTES ABOUT SOFTWARE DESIGN TOOLS cscscscsssssssssssssssssssossssssscscesesssssssescessssssscesssasssescscessssesscstssssssseeasassees 5 6 CONCLUSIONS AND RECOMMENDATIONS sisescszesscycozscucpceisscaysotecscyentensevendanssydndetayoudenvcveedenesveudavs A TRE 48 6 LITERATURE REVIEWS OF UNSELECTED CONCEPTS ccccsscsccssscessssssessessssessessseessessseeseeseees 49 6 1 ENGINEERING APP STORE WITH QUALITY METRICS w sccscssssssssssessesssessessessvcssessvcsucssessscsvessessvsssessessscssesuessecseesarsassssessseaveesees 6 1 1 Theoretical toundations of knowledge reuse 2i 2sscsiexed R A E E E A AEE 0 1 2 Software quality MECS esnai nnana aR NE R A AORE ONA AARRE 6 1 3 Engineering app store aspects discussed in literature 6 1 4 Existing approaches similar studies 6 2 CODE REVIEW ss csessasselecegtteseceatenseccatensecennenseceasensescateeeseasenseseascasceaseaseccaseasescasenseceaucnseceatendeceateaseccatenseceatensecsatensessateaseccatenseeestis 6 2 1 Purpose of code review 6 2 2 Attitude of the p
265. spective is supported by the knowledge lifecycle model see e g Tao et al 2005 The second is supported by two widely referenced theories from cognitive and organizational science which focus on knowledge transfer in organizations Part Il Initial study 4 Wenger theory of communities of practise and Nonaka and Takeuchi s theory of knowledge creation Wenger theory of Communities of Practise considers organizational learning as a social phenomenon on different levels individuals engage in communities communities refine practise and organizations sustain interconnected communities of practise Nonaka and Takeuchi s theory of knowledge creation is based on the distinction between explicit and tacit knowledge Organizational learning takes place as knowledge is refined and spread by passing trough 4 stages Socialization using e g observation and practise Externalization in which the knowledge is converted to an explicit form Combination e g using aggregation and classification and finally Internalization in which the knowledge is converted from explicit to tacit such as when an expert acquires know how This is shown in Figure 2 1 gt Tacit Tacit a Socialization Dialogue Externalization 3 p amp Field Sa Building nate Knowledge 3 5 a Learning by F faternalization Doing Combination et L Explicit Explicit atl Figure 2 1 The Knowledge Spiral Source Nonaka amp Takeuchi 1995 2 1 3 Knowl
266. st have at least 1 output and each slot should be in 1 class The incentive for modeling the process as well might be there already engineers find that a process view is easier than a description based on software objects 26 The process view could be a stepping stone for the class view helping to ensure the completeness of the class view 23 9 3 Application to Systems Engineering The principle of code and design documentation generation could be extended beyond software and applied in Systems Engineering More precisely it can make a contribution to the field of Model Based Systems Engineering MBSE MBSE is the formalized application of modeling to support Systems Engineering MBSE aims to represent all project data for the development phase and later product life cycle phases as interrelated digital models 34 For this purpose SysML 35 was developed a modeling language similar to UML for Systems Engineering Two major tasks for systems engineers are to define the system design and to guard the system integrity The project view on the system design flows down to the subsystems while the practical difficulties and required changes flow up This is especially important in Concurrent Engineering which requires intensive synchronization and consistency checking between engineering disciplines to maintain a shared vision 36 Part Thesis article 34 One of the largest challenges in MBSE at the moment is the seamless integration
267. state of the art tools Three broad classes were identified bi directional model generation one directional change propagation and bi directional change propagation An overview is shown in Figure 5 7 Change propagation across models Bi directional model generation One directional change Breirectional lige 3 7 propagation propagation Conventional round trip engineering Partially editable code Notification generation Batch mode iff ee THREE Saye automatic synchronization Real time Forward model transitions an automatic synchronization Figure 5 7 The main approaches and mechanisms to support propagation of modifications gt Regular languages are related to regular expressions a tool for text parsing available in many programming languages Part Il Initial study 42 Bi directional model generation Bi directional model generation is the simplest approach One repeatedly transforms the entire project to the desired level of abstraction before making changes This is commonly known as round tripping because of the alternating direction of the transformation The drawbacks are possible information loss due to the mapping issue when switching between levels of abstraction and possibly a limitation in how quickly and conveniently one can switch from one level to another van Dijk 2013 One directional change propagation
268. stment if they find themselves effective What do you find most time consuming in software development Most participants I1 13 A1 A2 find debugging the most time consuming i e the iterative debug fix test cycles after a first attempt The least experienced participants I1 and I2 find learning the advanced sides of a new language very time consuming Other time consuming activities are dealing with non engineering infrastructure 11 A1 cleaning up code 13 and translating business requirements to the software world 14 Part Il Initial Study Appendix B Documentation To what extent and on which levels do you document your software For all participants comments in the code is the main documentation They mainly document for themselves It is felt that not much documentation is needed and that it is sufficient to explain informally what is going on If code is to be shared with others documentation is more rigorous A2 indicates that in his case the additional effort for this is not justified by any incentive Frequently time pressure conflicts with the need to document 13 A1 A2 External documents tend not to be used 11 13 Al and A2 were asked to create a user manual though The user manual they received looked good but was flawed o 13 TI usually share my code with people who work very close to me We save a lot of time by speaking to each other rather than writing things down The problem of documentatio
269. stood when the data model has been defined clearly Therefore this chapter describes the data models at various levels and places in the application before the next chapters dive into the subsystems 4 1 Database data model The data model used between the database a Google App Engine datastore and the server application is shown in the class diagram below It matter mainly for the server subsystem but conceptually it affects the other subsystems too name string package string visible bool True when_was_last_design_change Date when_was_last_code_change Date when_was_last_change Date users AE Sth usemame string name string password_hashed string _ ps referenced_in CodeFile get_content_hash string A project consists out of design elements Diagram and ModelEntities code elements FileGroup and Codefile and access right elements Team and User The design elements represent the design stored in the project Diagram elements store an editable diagram document The actual format is of no concern to the server A ModelEntity represents a single element in a diagram such as a class child object or type relation Again the datastore is not concerned with the data format of the model entity apart from the name and the type because these are used in query operations A single ModelEntity can be displayed in many diagrams they are not owned by one
270. systematically rather than ad hoc and testing should be done on different levels starting with small subsystems and ending with full system tests Blaha amp Rumbaugh 2005 Testing is best performed by automated tests non automated testing is boring slow inefficient and prone to human error Automated tests are more likely to be run regularly Goodliffe 2007 Maintaining traceability between the requirements various design documents tests and other engineering artefacts allows justifying and explaining the choices made resulting in increased understanding of the design and enabling impact analysis of proposed changes Dick 2005 Nevertheless traceability is usually perceived as large overhead Rigorous adoption is limited to mainly bigger companies who are forced by some standard Neumtller amp Grtinbacher 2006 2 2 3 Software development methodologies and the software process Software development methods are frequently categorised as either plan based sometimes called traditional methods or agile methods Agile methods have emerged more recently They have a larger focus on anticipating changes and do so by emphasizing informal information exchange incremental development and customer involvement and or feedback Cockburn amp Highsmith 2001 The emergence of agile software development methods with their emphasis on communication seems to coincide with the observed shift in Knowledge Management from a purely technical view to a
271. t _ah stats Besides enabling it in app yaml which makes the statistics page available appstats must also be hooked into the RPC system so it can record the calls This is done in appengine config py discussed in a later section 6 4 2 server_loader py The server_loader py script is placed in the root folder of GenDL Designer so that GAE can find it import it actually server_loader py adds the subfolders of the project with server side Python modules to the Python path After this the modules can be imported In particular server_app app is imported so that server_loader app is available as configured in app yam1 This is necessary because the folder structure of GenDL Designer carefully reflects the three main parts of GenDL Designer the GAE server side app the web client and the code client The Python modules for the GAE app are therefore necessarily in a subfolder Adding the right subfolders to the Python path is done by add_server_to_python_path py This script is also used by appengine_config py discussed next 12 Content Delivery Network see http en wikipedia org wiki Content_delivery_network Part IIl Code report 51 6 4 3 appengine_config py appengine_config py is a Python script loaded when starting up the GAE app Currently it is only used to install the appstats hook The hook is slightly customized compared to the standard method GenDL Designer s hook omits a log statement for every request Those log s
272. t contribute directly to these answers such as designing and testing are skipped where possible Software engineering experience is limited and finally the development is highly iterative and incremental 8 2 Deployment results Based on this knowledge GenDL Designer was developed founded on the principle of incremental code and design documentation generation GenDL Designer s aim is to encourage users to create accurate design documentation and even create this documentation before writing the corresponding code A full scale experiment with GenDL Designer will be conducted in spring 2014 in anticipation of that trial runs were held The data from these trial runs are used in this report for drawing the following preliminary conclusions GenDL Designer encourages the creation of accurate design documentation and encourages designing before implementing This conclusion is based on the observation that when GenDL Designer was used for newly started projects nearly complete documentation was created and a large part was created before writing the corresponding code In regular Engineering Automation projects that would have been unlikely It is however the case that not all users found it necessary to document all code and that some users have a lower tendency to design before code than others GenDL Designer does not eliminate that The principle of incremental code and design documentation generation has the potential to improve the
273. t has the advantages that it is simple to implement and very robust the engineering automation developer his workflow is never blocked by imperfections in the mechanism The downside is the manual work for the user This must be kept as low as possible and should be evaluated together with the prototype users Code synchronisation can make use of standard industry proven techniques for code generation and parsing For code generation the template based approach is the simplest Alternatively pretty printing an AST is possible if templates are not flexible enough For parsing parser generators are available that can generate a parser from a language description Further processing of parsed contents might prove to be difficult since lower level models such as source code typically allow for more variation than higher level models such as diagrams Recommendations for subsequent research When the software design tool prototype has been implemented these recommendations can be considered If the prototype is successful it could be extended with design critics such as in the ArgoUML software In addition to notifications about inconsistencies with the code recommendations can be given on how to improve the software design itself Regardless of the success of the prototype it can only solve a part of all problems Other tasks of professional end user developers should not be forgotten Part Il Initial study 48 6 Literature reviews of unsel
274. t ins E geom base base object amp E surf box solid G EEEE E surf cylinder solid G fuel tank E surf extruded solid 1 gt HE geom base a a surf surf box solid iz gwl poddns g y2eqp 4 A Screenshot of the main interface of GenDL Designer Part IIl Code report Table of Contents ABBREVIATIONS cccccsssscsssscsssecesecesseecsseecsseecessecessecessesesseecseecesseeesseeessecessesceseecesseeessesessescnseeseseeseseeeeses 1 INTRODUCTION ccccsscessscessecesseeceseeceseecessecessesesseesessesessesessescnseecnseeseeeeessesessesceseeseseeseseesenseecesesceseeoens 2 SYSTEM USAGE 1 cccecccsssscessscesesesseecssssceseecessesessesesseeceseesessesesseseeseseessescsessessesensescesseseseesessesenseeesseeseseeoees 3 1 INTRODUCTION isieteeiey E iel esi E atlas E A A ate ates el este ele telat eines E A a 2 FEATURES oaii 2 1 UML drawing 2 2 Project tree 2 3 Multiple diagrams 24 Consistency checkin nonar nn AA AAAA AE AA seeaaiai sesdere tacsaests 8 2 5 Code Generation ae wht ly ie its ab a a 26 Eo k p code secstsicsccissi iiseseexsstectyns A ERE AE a RR a RE R RE 2 0 Project bootstrap PACkAge see ccstsaseveswttiesstexseetetthivareatasestvsdiveelethssantesaivestentadsasabigearesacesosetosdensaststantestesesteroateintess 11 2 8 Downloading and restoring diagrams 2 9 User accounts 2 10 Help center 2 11 Admin panel 3 IK NOWN BUGS E EE EEE Rae Ree
275. t object This is regular javascript object with the keys origin and optionally is_last origin signals to other modules where this change comes from and signals the origin in the log files is_last indicates whether that entity manager operation is the last one one add_to_design operation can invoke multiple entity manager operations e g add a class also adds the children and the relations with Part IIl Code report 60 the children is_last is used by other modules to know whether they should update the UI or wait 7 3 10 Code generator and dialog modules The code generator module generates code snippets for classes class entries and functions Class entries are attributes and methods ie GenDL input slots output slots objects and functions Django templates are used to generate the code These templates are the same templates as the ones used on the server side to generate code for the bootstrap package The data that must be inserted in the templates was described earlier in section 4 2 The code dialog module allows showing a code dialog with optionally a file download button It is used by the consistency_resolution module This module uses Jqote2 to render the class dialog content The content is shown in a dialog with the Bootstrap modal feature 7 3 11 Auto update tab module Both the consistency tab and settings tab are auto refreshing when viewed they refresh themselves every X seconds to show the latest sta
276. t of inconsistencies between the design and the code In addition the server subsystem is also the main hub from where both the web client and the code client are retrieved by the user the web client is viewed through a browser the code client is downloaded The server subsystem runs on the Google App Engine GAE which will be introduced in the next section The server subsystem mainly consists of a GAE app which is written in Python The app handles HTTP requests from the web client and the code client and uses the GAE Datastore and the GAE Memcache to persist data and retrieve persisted data This is shown in Server side GAE infrastructure GAE Data Store O O server_app web server plygendi lisp parser Design modifications Inconsistensies Code modifications The level of Python expertise required for the server subsystem is moderate Advanced topics and tricky aspects of the Python language few as there are are not needed mainly due to the excellent frameworks that are available The only relatively advanced aspect of the Python language used in the GAE app is the decorators facility When starting it is sufficient to know what they are and how to use them Part Ill Code report 34 6 2 Technologies used Three groups of existing technologies are used a parsing framework PLY the Google App Engine and template frameworks These technologies are introduced first together with their references before
277. t test E S PROJECT_DIR_NAME server stc E S PROJECT_DIR_NAME server vendor E S PROJECT_DIR_NAME server test PyDev PYTHONPATH e The final PYTHONPATH used for a launch is composed of the paths defined here joined with the paths defined by the selected interpreter Mercurial Pijet pies 2 External Libraries String Substitution Variables Project References External libraries source folders zips jars eggs outside of the workspace PyDev Interpreter Gramrr PyDev PYTHONPATH When using variables the final paths resolved must be filesystem absolute PARSEE etsy Changes in external libraries are not monitored so the Force restore internal info Run Debug Settings should be used if an external library changes Task Tags Validation C Program Files x86 Google google_appengine C Program Files x86 Google google_appengine lib webob 1 2 3 C Program Files x86 Google google_appengine lib webapp2 2 5 2 C Program Files x86 Google google_appengine lib yaml 3 10 C Program Files x86 Google google_appengine lib fancy_urllib Final step is to try out the local deployment of your copy of GenDL Designer For this see the steps in Section II 3 3 Part IIl Code report 33 6 Server subsystem 6 1 Introduction The server subsystem main responsibilities are to provide a storage service for the design and the code and to generate a lis
278. t that It s almost assumed in the team that people can program In line with the main explanation found in literature only few consider it feasible to write detailed requirements in advance Only one participant was aware of design methodologies like MOKA which he said was perceived by his team as a fairly theoretical overhead One participant explained why he did not make a detailed design of his software I found it difficult to write down beforehand what I m about to do It was never taught how to do that for programming while we did learn how to do this for say mechanics The participants have the feeling that they can get by with on the go testing as they develop and use their software They think systematic testing takes too much time or will not pay Part Thesis article 8 off Separating theory and implementation errors or the lack of formal requirements was not felt to be an issue by the interviewed participants One participant illustrated what happened as a result of a lack of testing When received the tool my supervisors assumed that it worked and that could simply extend it The supervisor after finding out that was not the case Oh well then must have read a very good report Some feel that external documentation will not be read Instead people ask questions directly which saves time Going through the calculation process step by step and seeing the output leads to new insigh
279. tatements were clogging up the log files appengine_config py also needs to set the Python path right This is done with the same script as in the server_loader py script Part IIl Code report 52 7 Webclient subsystem 7 1 Introduction The webclient provides the main interface of GenDL Designer It provides the environment where the user creates and modifies the design and where he sees the inconsistency notifications The web client is hosted by the server subsystem and runs in the browser The web client is a regular web app written in the normal web languages HTML CSS Javascript and communicates with the server subsystem over HTTP The most important components of the web client are shown below The web client has two main areas the user can be in the design area with diagrams and the consistency area In the design area the user can modify the design These modifications are instantly saved to the server In the consistency checking area the user can see the inconsistencies which are continuously retrieved from the server For some inconsistencies the user is given the option to automatically modify the design in the design area The consistency area itself doesn t save design modifications this is the task of the design area Design modifications Inconsistensies Web browser a a a ao a ee Ee eee ee eee ee ee eee a eee ewe wee wee ee ee ee Design area Consistency area The level of Javascript expertise required
280. ted while a flow of 1 would indicate that the design was only created to Part Thesis article 23 document existing code A higher flow indicator would generally be better but a perfect score of 1 is unrealistic in real life projects and by no means required Part Thesis article 24 6 Experiment trial runs 6 1 Introduction Individual testers were given access to GenDL Designer in preparation of the full scale experiment that will be conducted in spring 2014 These trial runs uncovered bugs triggered new feature requests and allowed refining the support material It also gave the opportunity to develop and test the log processing facility with realistic user input Some testers used GenDL Designer to set up their project from the beginning Others used it to document already existing code written by themselves and or others A final group used GenDL Designer to perform a small assignment toy project purely for the sake of testing GenDL Designer In total 7 users participated in testing The remainder of this section will present the data that was extracted from the server logs the calculated metrics and the user feedback that was gathered Although the amount of users in the trial runs is too small to draw solid conclusions the data will be processed the results will be discussed and preliminary conclusions will be presented as an example of how it will be done when the full experiment will be conducted 6 2 Serv
281. terface of IDEA were created with jQuery Later it became clear IDEA had to evolve into a general knowledge editor which could not be aligned with the ideas for GenDL Designer incremental code and design documentation generation Instead GenDL Designer was developed as a separate application from the ground up but some GenDL Designer user interface elements already created for the showcase were reused The second reason is more practical there exists a django templating module for dojo This allows using the same django templates for code generation on the server and in the browser Apart from the templates the most visible aspect of dojo in the system is the module loading system it uses AMD AMD is not part of dojo but dojo uses this convention As it is good practice the rest of GenDL Designer was also structured with AMD modules 7 2 6 Bootstrap Bootstrap is an open source CSS styling library developed by Twitter With very little effort standard HTML markup can be turned into a sleek good looking website or web application On top of this Bootstrap provides a couple of minimal plugins for things like displaying modal dialogs Part IIl Code report 55 Most of the interface of GenDL Design apart from the diagramming component consists of Bootstrap styled and Bootstrap rendered elements 7 2 7 jQote2 jQote2 is a templating plugin for jQuery It was discussed earlier together with other templating engines in section 6 2 3
282. the XML file with the design gets replaced with another earlier file When this happens the design on the server should update to the state just checked out This would make switching between branches easy Part IIl Code report 68 9 5 Cut copy paste and undo in diagrams These features were disabled because they are error prone Copying for example copies all data including the unique id which is then no longer unique Undo is a great feature but some modifications should not be un done e g assigning an id in the background The smallest bug in these features might lead to data integrity problems and eventually data loss One would have to do extensive testing preferably with real users but with unimportant data a difficult combination 9 6 Suppress selected consistency notifications In some cases the user can see consistency notification of which he knows they can be ignored In order not to distract him of other more important notifications there should be an ignore button on every notification 9 7 Drag amp drop for slots While re designing just like when coding it takes some time to figure out where in the class model which information should be available Slots are shifted around and duplicated passed down between parent objects and mixins until eventually the puzzle fits It would be very convenient if this could be done by simply dragging and dropping slots This feature would increase the effe
283. thermore the notation uses the generalization specialization connector for both mixin relations super classes and child type relations 4 4 Application architecture GenDL Designer is built as a web based application Users draw diagrams and compare these to their code inside the web based application while they still write code locally on their own system like they used to do This setup requires no changes to the existing workspace of engineers and therefore keeps the barrier to get started with GenDL Designer low The link between the web based system and the local system is provided by a small utility that runs locally and mirrors the code to the web based system This allows the web based system to know the current state of the code The design that is represented visually in diagrams is internally also stored as a traversable graph with nodes and edges This graph the design model and the parsed contents of the code the code model are both transformed to a similar data structure An algorithm compares these and generates a list of inconsistencies Nearly every action of the user invokes a request to the server All requests are logged and can be analyzed to derive usage statistics Because GenDL Designer is offered as an on line service this can be done continuously to monitor usage and react to issues as soon as they appear long before the end of the research 4 5 Alternative GenDL Designer usage patterns While intended as a
284. thod is triggered check_access_to_project will consult the session attribute and check whether the username in there belongs to a user with sufficient access rights If there is no username the response will be a redirect to the login page and the get method will not be triggered Password handling At first sight the password handling might seem unnecessarily complex the system does not simply check for equality with a previously stored password The user passwords are hashed SHA 256 with a salt before storing them in the datastore a basic security precaution The same thing is done every time the user wants to log in if the end result is the same the password was correct This approach improves security because hashing is a one way transformation of the password It prevents hackers should the database ever fall in their hands from taking users their passwords and trying those on say their webmail or banking account Still so called rainbow tables are available that map hashed values to often used passwords with that hash By adding the salt a fixed sequence of random characters this kind of attack is also prevented admin is a special user which passes every security check The admin password is not stored in the database like regular users their password It is a configuration setting This is a common setup to ensure the admin password can be checked even when the database cannot be accessed To change it see Section
285. thor To this side of the spectrum one finds knowledge repositories newsletters and web portals With traditional knowledge repositories users can change content but cannot change structure This Knowledge Management strategy fails to react to unforeseen circumstances New strategies focus on practises and outputs instead of on knowledge These platforms such as blogs and wikis do potentially a better job in representing how the work really gets done Knowledge repositories require discovery mechanisms Examples of such mechanisms are search links tags automatic recommendations and notifications e g via email The main risks for repositories are that too few people might feel the need to author and that the delicate balance between strong quality assurance and low entry level for contributors is missed 6 1 2 Software quality metrics Software quality metrics and ratings give an indication of the quality of software Simple metrics are e g how much tests and comments are present in the code compared to the total size of the code More advanced metrics indicate problematic parts of the code by measuring software complexity the degree to which a system or component has a design or implementation that is difficult to understand and verify Several complexity measurement procedures have been proposed Commonly they score code components by taking into account the amount of control structures branching iteration function calls etc
286. time ago message when visiting a project after one day the timestamps were not used for some time so the memcache servers removed the timestamps from memory If last changed timestamps are important to users it might be worth it to use the database to store the timestamp Part IIl Code report 70 Appendix A Quickstart manual GenDL Desiger quickstart guide v1 1 26 January 2014 Introduction This is a manual to get you stared with the GenDL Designer quickly It gives a quick tour with lots of screenshots for clarity and gives some important information The GenDL Designer is a web based application to support the design and implementation of GenDL code This should result in better and easier to create KBE code You can find the application on http gendldesign appspot com Start screen When you browse to the application you will see the login screen Firefox GenDL Designer gendldesign appspot com amp gendldesign appspot Design amp Implementation Support for GenDL Projects Login You need to login before you can continue yoddns 8 Deqpe24 a All projects are private you need to log in before you can see and edit them Part Ill Code report Appendix A Once you ve logged in you will see all your projects GenDL Designer e gt gendidesign appspot com c B P Se Logged in as pjardewitte Logout Design amp Implementation Support for GenDL
287. tment of Flight Performance and Propulsion dr ir T van den Berg Delft University of Technology Department of Flight Performance and Propulsion ir D Steenhuizen Delft University of Technology Department of Flight Performance and Propulsion dr A Bacchelli Delft University of Technology Department of Software Engineering dr ir P Bermell Garcia Airbus Group Innovations Preface Software development is becoming more and more important in aerospace engineering research Reducing the weight of a load carrying structure by optimizing its topology the best method for rescheduling aircraft flights to minimize the delays for passengers investigating the feasibility of a circular landing strip by simulating take off and landing all these aerospace problems can and have been researched by writing software In a way writing software has become a research technique a way to find answers to engineering problems like applying calculus to solve differential equations or performing chemical experiments in a lab In contrast to the widespread use of software development among engineers there are widespread difficulties encountered by engineers writing software Software development is a relatively new technique and is still evolving fast Our understanding of software development in engineering research has not yet crystallized into clear and widely recognized conventions and best practices This is where software development
288. to improve understandability and validity in order to contribute to more reuse GenDL Designer aims to improve understandability by encouraging developers to create design documentation and to create this documentation for a large part before actually writing code This encourages a well thought and sensible application structure just like thinking through the problem and possible solution alternatives is beneficial in any engineering discipline GenDL Designer aims to improve the validity too by checking that the design documentation is complete and still corresponds to the code In summary GenDL Designer tries to accomplish what is now hardly done among engineering automation developers that applications are documented properly and that developers create a design before writing the corresponding code To accomplish this GenDL Designer takes into account the four non technological obstacles related to the culture of engineering automation developers identified before By generating code skeletons from the design and diagram elements from the code designing becomes a directly value adding activity and documenting becomes less time consuming GenDL Designer does not require much training apart from reading a short manual or watching screencasts And finally GenDL Designer is designed from the ground up for iterative development by generating code and design documentation incrementally GenDL Designer was developed for Engineering Automation de
289. tor but it is strongly recommended to use an Integrated Development Environment IDE which supports editing Python HTML CSS and Javascript code and provides an overview of the project files You ll be switching back and forth between different files and different languages regularly hence the practical advantage of such an IDE I personally recommend using Eclipse with plugins for Python Pydev and web development The remainder of this chapter gives instructions for that IDE but other IDEs could be used as well and the procedure will be similar 5 2 Required software First install e Python 2 7 32bit see Section II 3 1 e Google App Engine see Section II 3 1 e Eclipse is a downloadable zip file no installation required Requires Java though e PyDev is installed through the Eclipse plugin system e Eclipse Web Tools is installed through the Eclipse plugin system If you intend to change the code client you will need to install additional software to build the executable This and the rest of that build process are explained in the code client chapter and can be skipped for now 5 3 Creating an Eclipse project When starting up Eclipse you choose or create a workspace which is a folder on your file system A workspace contains projects which are simply folders in the workspace with some project meta data inside Start by creating a new Python Pydev project You can leave it entirely empty Next check o
290. ts and triggers iteration This responsiveness to change and built in iteration matches with agile development methods One participant explained why he did not use a formal method he admitted that even though he in general acknowledged the value of formal methodologies he could not get himself to look out for those and use them for his software work 3 4 3 How engineering automation software is currently shared and reused Literature Segal 4 and J Howison and J Herbsleb 6 report a limited level of sharing and collaboration among professional end user developers and in the scientific software community respectively Software is passed on from researcher to researcher resulting in problematic software artifacts Interviews Among the interviewed participants reuse takes place on a small scale within teams Knowing about existing code is done through internal team interaction there is no central repository Reuse occurs by copying legacy code and modifying it The ideas behind the software are shared informally if one asks for it rather than writing them down One participant said We save a lot of time by speaking to each other rather than writing things down The problem of documentation found is that no one reads it people rather look over their desk and ask directly Currently code is copied rather than turned into a shared library because that easily gets broken or is moved One participant said Their change
291. tudy 30 save time by avoiding problems later on Justifying the time spent on code review requires that the organization supports or even enforces code review If not code review will not be taken seriously and fail Code review does not require software engineering training for the developer whose work is reviewed in fact the review is training for him The reviewers don t need to be highly trained either though they do need to know what good code looks like Applying code review iteratively is no problem and is in fact highly recommended 4 2 4 Coding policy Coding policies Goodliffe 2007 McConnell 2004 are an explicit version of best practises with respect to code and what it looks like to make it easier to work with In professional software organizations it is normal to have a coding policy or house style Coding policies would have a positive effect on reusability They make it easier to reuse software by encouraging a uniform code style that makes code more understandable Other developers can then focus on important aspects of the code rather than keeping track of arbitrary details It can also prescribe testing practises that improve the validation of the software and therefore the reusability Regarding the non technological obstacles the standardization that a coding policy prescribes doesn t add or remove any work except for the need to familiarize yourself with the policy This can be done gradually through e g poste
292. turns the inconsistencies but also warnings and errors encountered during consistency checking All three are sent to the web client to be displayed Analysis output format The inconsistencies are a list with elements of a fixed format lt element_type gt lt problem_type gt element_data_1 some_data Part IIl Code report 48 element_data_2 some_data file element__data file line element_data line Element type can be class function slot etc Problem type typically is in design only or in code only The element data is used to generate the inconsistency warning messages and possibly a code snippet The data varies per element type and problem type Errors and warnings are simply lists of strings which will be shown on top of the inconsistency list in the consistency tab Consistency checking algorithm Because the design and the code are expressed in the same unified model see section 4 3 Unified Design Code data models comparing them is relatively straightforward The top level functions compare_classes and compare_functions cascade down to compare_slots compare_description etc These in turn use generic low level functions which e g compare sets of names The different modules in server_dataprocessing consistency reflect this approach Altering the comparison is straightforward As an example suppose you want to compare the arguments of defun statements a check currently not do
293. tus on the server When switching to another tab refreshing needs to be suspended temporarily Since this is common behavior the logic that determines when to update was moved to a separate module the auto_update_tab module This module doesn t do the update itself instead it calls a provided callback function whenever an update should be performed auto_update_tab takes care of suspending these calls when the tab is not active Part IIl Code report 61 7 4 Activity diagrams The relation between the different modules is clarified with activity diagrams 7 4 1 Subscribe mechanism The subscribe mechanism is demonstrated through an example where the user adds an element to the diagram The design model on the server is then updated and the project tree refreshed This involves the modules diagram js entity_manager js and tree_display js ideaDiagram lt CELL_ADDED gt diagram js event_CELL_ADDED entity_manager js em add n gt HTTP PUT lt project gt design data lt id gt json entity_manager js em notify_subscribers tree_display js lt anonymous function gt subscribe tree_display js query_and_refresh_entities l tree_display js refresh HTTP GET lt project gt design data tree json Part IIl Code report 62 8 Code client subsystem 8 1 Introduction The code client monitors files on the user his computer and reports changes to the
294. uctions on where to unpack the files to If you do need to change them see Section III of this report for build instructions 3 3 Local deployment e Open the App Engine Launcher installed as part of the SDK e Click File Add Existing Application and navigate to the root folder of the source code This is the folder that contains app yaml e Adjust the ports if necessary know which ports have already been occupied on your system If you are sure you have never installed any local server the default is OK e Run the application and wait until the status icon becomes green e Browse to http localhost 8080 adjust your port if you did not take the default You should see the login screen of GenDL Designer e Test downloading the code mirror tool and try uploading a valid and invalid lisp file with it If the consistency tab displays the contents of the correct lisp file as missing in the design and also shows an error message for the invalid file the whole system is installed correctly 3 Mercurial or Hg is a distributed version control system very similar to the git version control system Hg and git are alternatives to older version control systems like SVN and CVS which require and are limited by a central version control server Part IIl Code report 17 Warning although the SDK development server seems to be working very well and reasonably fast Google never implied any suitability to run production applications Data can
295. unaware of how software engineering practices like designing and testing or even basic tools can fix the problems they are experiencing Iterative and incremental development Engineering automation software is developed iteratively incrementally and interactively It is common for a project to start vague without many requirements and without much of a design Developing the calculation process or seeing output triggers new insights and revisions of the entire software solution These conclusions show that to promote reuse several issues must be resolved GenDL Designer which will be described next was developed to tackle the first few and to further refine and solidify the understanding of engineering automation development Part Thesis article 13 Part Thesis article 14 4 GenDL Designer 4 1 Scope GenDL Designer is a graphical design tool with code synchronization for GenDL an open source object oriented knowledge based engineering system GenDL developers can create software design diagrams and generate code skeletons from them Afterwards both the diagrams and code can be changed independently GenDL Designer continuously analyses both and lists inconsistencies between the two Optionally it helps to resolve inconsistencies in both directions by generating additional code skeletons or diagram elements This will be referred to as incremental code and design documentation generation GenDL Designer s overall aim is
296. unctionality A variety of software tools to support design and subsequent implementation are available to software engineers of varying quality Goodliffe 2007 These tools originate from both academic and commercial backgrounds Although these tools are not geared towards professional end user developers some serve as useful examples of state of the art technology in software development support tools According to Tri amp Tho 2012 the most famous UML modelling tools include IBM Rational Rose Enterprise Architect and ArgoUML Further discussed are the latest developments in the IBM Rational product line IBM Rational Rose Rose was initially developed by Rational Software Corporation the cradle of UML It is currently in the high end of the market with a yearly license fee of several thousand dollars per seat Rose supports round tripping with the partially editable code approach followed by batch mode automatic synchronization Special comments mark the areas which can be edited before re importing the code into the design model However IBM the software vendor of Rose acknowledges that with this mechanism changing the source code more drastically than changing simple implementation details can easily become problematic up to the point where the design model validity is affected Williams 2006 IBM Rational Software Architect family The Rational Software Architect family of products is the successor of Rose The licence fee
297. understandability of applications and the validity of their documentation GenDL Designer reduces chaos adds structure and provides overview The validity of the application itself is also positively influenced through the increased understandability which uncovers defects Part Thesis article 31 It is feasible to introduce incremental code and design documentation generation in an engineering automation context GenDL Designer demonstrated that it is suitable for engineers because it handles the incentive and deadline pressure by not being a burden because it does not require software engineering training and because it fits with the usual iterative and incremental development style However some potential users will not see a reason to adopt the approach because on the short term the time the approach saves them is about equal to the additional time it takes Further improvements in the user interface might make the approach a net time saver also in the short term and convince more potential users These findings are promising but preliminary The full scale experiment will point out whether these conclusions are indeed justified 8 3 Limitations An important characteristic of GenDL Designer is that the employed design language maps closely to the code language The main difference is the level of detail For situations where this is different incremental code and design documentation generation might not be feasible code synchroni
298. users design before writing code is measured by calculating to which degree information flows from the design to the code and from the code to the design The modifications are grouped into sessions of successive design or code modifications At the end of each design session and each code session the inconsistency change during that session is calculated Al s inconsistency D Cs inconsistency D _1 Cs_1 s0 5 4 Here s refers to the session for which the change is calculated The flow of information is positive if information flows from the design to the code This is the case when the inconsistency increases during a design session or decreases during a code session The design is then leading with new information the code catches up with information that was already in the design The set S of sessions with positive flow and the set S with sessions of negative flow are defined as St s Sp Al s gt 0 U s Sc Al s lt 0 s 0 5 5 S s Sp Al s lt 0 U s Sc Al s gt 0 In these equations Sp and Sc are the sets of design and code sessions respectively Summing the flows in each set yields the total positive and total negative flow flowt ee A s 5 6 flow Doe A s Finally the overall flow indicator is calculated as a number between 1 and 1 flow flow no flowt flow a A flow of 1 would indicate that the design always perfectly foresaw what had to be implemen
299. ut the source code repository to a temporary location see Section I 3 2 When finished move all files and folders including the he folder to the Eclipse project folder on your file system Then refresh the project in Eclipse select the project and press F5 You should now see all files and folders in Eclipse Then configure the Python path for this project in Eclipse This is needed when you want to run Python code outside Google App Engine e g for automatic testing It is also handy because it enables auto complete and code analysis features Right click the project and choose properties In the dialog that appears go to PyDev PYTHONPATH and configure the path as shown in the two screenshots below gt Part IIl Code report 32 PyDev PYTHONPATH Resource Builders The final PYTHONPATH used for a launch is composed of the paths defined here joined with the paths defined by the selected interpreter Mercurial Project Facts Source Folders i External Libraries String Substitution Variables Project References Project Source Folders and zips jars eggs PyDev Interpreter Gramrr PyDev PYTHONPATH When using variables the final paths resolved must be workspace relative Refactoring History E S PROJECT_DIR_NAME experiment src Run Debug Settings E S PROJECT_DIR_NAME codeclient src Task Tags E S PROJECT_DIR_NAME codeclient vendor Validation ES S PROJECT_DIR_NAME codeclien
300. utomation software An overview of the participants is given in Table 3 1 To ensure anonymity participants will be referred to with an anonymous ID rather than with their real name Type ID Position Work experience and typical use of Engineering Automation Industrial 11 Intern Industrial experience 4 months in current position previously placement in Eurocopter Typical use Post processing simulation data for visualization or to derive results Industrial 12 PhD Student Industrial experience 1 year in the company Typical use Automate the application of data mined engineering knowledge Industrial 13 Contractor Industrial experience 1 year and 9 months in the company Typical use Automate conceptual design and simulation Part Il Initial study 17 workflows Industrial I4 Contractor Industrial experience 14 years in the industry 12 years in the software company consultant Typical use Develop Knowledge Based Engineering applications Academic Al Graduate student Industrial experience Typical use Aerodynamic side of Multi Disciplinary Optimization Academic A2 Graduate student Industrial experience Typical Engineering use Automate conceptual design and simulation workflows Table 3 1 Overview of the interview participants The interviews were conducted and processed in a uniform manner It was chosen to have semi structured intervie
301. utton will ask for confirmation so don t worry if you accidentally clicked the Remove button Once removed this is irreversible e You can download the current design and later upload again This is good for making a local backup or to keep versions Multiple diagrams To manage the complexity of your project you can split your design into multiple diagrams A design element can be in multiple diagrams at the same time Its data will be shared There are two ways to create a new diagram The most obvious way is to click the button in the diagram toolbar Alternatively you can create a diagram link from another element such as a class A shortcut element then appears Double click the shortcut to navigate to the new diagram The first time you use the shortcut it will ask you for a name The most common use for multiple diagrams is the following pattern objects fuselage Export images Exporting images is currently not supported but there is a button in the toolbar to toggle the background grid Without the grid you can easily take screenshots Part Ill Code report Appendix A Issues feedback errors feature requests Finally notice the Feedback amp Support button on the right of every page When you click it you can submit ideas feedback bugs but also vote for features and issues Or you can use the button to contact me directly Gendl Designer Demo Pieter Jan
302. vel ones rather than creating the models directly Although powerful this approach introduces a cognitive barrier rather than describing a model instance one has to describe a more abstract transformation process that can generate a variety of models depending on the input model Tri amp Tho 2012 Bi directional change propagation The third identified approach is bi directional change propagation although this could also be considered a more advanced form of round tripping The individual evolution of models causes loss of consistency between models Two mechanisms to regain consistency are notification generation and automatic synchronizing Notification generation reports inconsistencies and leaves the actual resolution of inconsistencies to the user Because the user has full control over both models his workflow is never blocked due to limitations or errors in the consistency checking system Implementing consistency checks is usually easier than implementing automatic change propagation The workflow suggested in Lauder et al 2010 to build such a tool is 1 Define a meta model for each of the models involved 2 Define an integration scheme between the meta models 3 Create platform specific integration code The drawback is the need for manual synchronisation activities although when different engineers work on different models manual synchronisation might in fact be preferred because of the increased control and awareness Lauder
303. velopers and what is needed when they have to reuse code Topic Currently done Desired Understandability Low code quality but usually with comments Simple clear code with comments No external documentation other than High level documentation about the reports structure and the steps Validity No systematic testing Verifiable correctness of code Documentation not entirely consistent with Documentation consistent with code code 3 6 Conclusions Literature and interviews show that the level of reuse in Engineering Automation is limited Sharing and reusing software is done informally within teams The two most pressing issues that impede reuse are understandability and validity When these issues are resolved raising the level of reuse further will require scaling up the internal team interaction on which reuse now relies Also sharing with others and supporting them will need to become easier and or more rewarding These issues are shown schematically in Figure 1 Individual context Social context Understandability Social interaction Code quality amp documentation Beyond team Towards more reuse Validity Testing amp doc validity Must become more rewarding Support reusers Required for sharing within team Required for sharing with others Figure 1 Schematic representation of obstacles that impede reuse in Engineering Automation Understandability The understandability of the code is low due to the lack of high
304. velopers and therefore differs from existing general software engineering tools with similar objectives such as the IBM Rational Software Architect products 29 Sparx systems Enterprise Architect 30 and ArgoUML 31 Most importantly it uses a simple modeling language tailored to GenDL rather than a full featured modeling language suitable for general software development Part Thesis article 15 The lack of options to choose from makes the tool easier to pick up GenDL Designer does not support automatic synchronization as well as the commercial solutions from IBM and Sparx Systems but on the other hand does provide more control over synchronization than those solutions do Help center Diagram tabs Consistency tab Settings tab Firefox Gendi Designer demo gendidesign appspot com projects 5640 vi Bene Main diagram Consistency Design Code Settings Main diagram Detail of lifting surface a Qa aga x e gt wing gt E lifting surface v Popular built ins geom base base object amp E surf box solid amp surt cylinder solid amp surf extruded solid fuel tank volume nil gt E geom base gt amp suf surf box solid EE gwl objects Project tree Drawing canvas Feedback button Figure 2 Web based User Interface of GenDL Designer Main diagram Consistency Design Code Settings Help centi Design Code De
305. velopment are spreadsheets recording macros in word processors customized email filtering and processing rules but also scripting interfaces embedded in applications with a low entry level A major driver for End User Development is the diverse and changing nature of requirements It is hard for external software professionals to keep up with their users needs This is avoided if the users themselves are able to continuously adapt the systems to their needs 16 Part Thesis article 3 2 3 Professional End User Development Some end user developers develop software in pursuit of their professional goals in areas outside computer science such as engineering medicine business and more Some of the tasks they need to perform are suitable for automation yet they might be so specific that a commercial solution is not readily available 16 These professionals tend to have a highly technical field of expertise in which case they are used to formal languages and abstraction and hence tend to have few problems with coding per se which sets them apart from regular end user developers 5 The following definition for professional end user developers is based on a description given in 5 and is similar to the definition given in 16 except for the constraint on technical professions Professional end user developers are people working in highly technical knowledge rich professions who develop their own software in order to advance t
306. view which also incorporates a social perspective Bjornson amp Dingsoyr 2008 acknowledge that knowledge management relies primarily on explicit knowledge in plan based methods and primarily on tacit knowledge in agile methods For example requirements documents are more formal and extensive in plan based approaches while agile approaches would emphasise iterative customer interaction There is a gap between the demand and ability to produce high quality software cost effectively Basili amp Rombach 1991 Effectively managing the software process is an enabler for software quality improvement and Software Process Improvement SPI is the most widely used approach Niazi et al 2005 The two most common SPI models ate CMM CMMI Paulk et al Part Il Initial study 6 1993 Software Engineering Institute 2002 and the ISO TEC 15504 Standard SO TEC 15504 1998 often referred to as SPICE CMM short for Capability Maturity Model Integrated is a reference model for assessing and improving software process maturity in an organization along an evolutionary path from ad hoc chaotic processes to mature disciplined software processes The CMM is organized into five organizational maturity levels For each level the key process areas that will help reaching the next maturity level are identified Herbsleb et al 1997 SPICE is a suite of standards on software process assessment In contrast to CMM CMMI SPICE grades each process separately
307. ware Engineering background to make a design The concept can be implemented such that it remains convenient to use throughout iterations of design and code changes 4 2 2 Engineering app repository with quality metrics 4 2 2 1 Introduction The engineering app repository intends to be a central virtual portal where automated engineering solutions apps can be found and evaluated An app is built around an executable process but also includes related documents e g to know where and how to apply the app or to gain understanding of how the process works Like any knowledge repository the purpose of the engineering app repository is to lift knowledge out of the minds of individuals and make it accessible throughout the organization or perhaps even beyond Among the means to evaluate apps are software quality metrics ratings and indicators developed within Software Engineering that point out possible flaws App developers might pro actively react to the feedback provided by the metrics because they know the metrics are visible to colleagues Compared to regular knowledge repositories an engineering app store has advantages in the incentives and costs of participation in knowledge repositories 4 2 2 2 Participation incentives Giving visibility to all software activities and software quality will encourage knowledge providers to engage in software activities now neglected and encourage to keep delivering quality A reward model for
308. whether dropping the blue arrow will work The current relations are e Class is composed of child object diamond link Child object has type class triangle link e Class has mixin class triangle link e Function is followed by function arrow link e Class uses function dashed arrow link e class link to diagram diagram reference dashed link One can also drop the blue arrow in the void on the canvas after which a new element will be created automatically If there are multiple options a menu with choices is displayed Dropping in the void is in fact the most common method to create new elements because it conveniently creates both a new element and a relation Editing the title of an element is done through double clicking the title The class slots and functions can be edited by double clicking them The dialog shown below then appears Details of wing Description This object represents a wing of the conceptual aircraft The wing is a swept wing with 1 kink at 40 of the span Input and computed slots Kind Name Description Required t Computed slot E fuel tank volume Fuel capacity in m3 L Remove Add Methods object functions Cancel Save changes 2 These are LISP functions not GenDL functions i e defun instead of define object functions Part IIl Code report 6 Zooming and panning moving around the diagram is also done with the mouse To zoom scroll the mouse To pan r
309. ws a fixed list of questions ensured that the same topics were discussed with each participant while at the same time the freedom of follow up questions allowed the interviewer to dig deeper and understand the answer better In total 22 fixed questions were asked see appendix A The interviews took about 1 30 hour each The questions were sent in advance Each interview was scheduled at a time convenient for the participant during multi hour simulation after a deadline on a Friday etc so that they had the time to answer the questions fully The interviews were recorded Afterwards the recording was transcribed and slightly summarized Small talk and confidential information such as partner company names were filtered out The summary was reviewed by the participant to correct any misunderstandings Also permission was asked to publish the summary anonymously In a last processing step a qualitative analysis step the answers of all participants were aggregated and compared per question Where applicable the most prominent answer or answer aspect was distilled by counting how many of the participants gave that answer or referred to that answer aspect and by taking into account how important participants found them Importance was indicated by the participants or inferred from the context The aggregated answers are available in appendix B 3 3 Results With the aggregated answers the main questions as stated in the introduction can be
310. y how Engineering Automation software is currently reused and what more reuse would actually require This should result in a set of constraints and opportunities to base my strategies on Type ID Position Work experience and typical use of Engineering Automation Industrial I1 Intern Experience 4 months in current position previously placement in Eurocopter Typical use Post processing simulation data for visualization or to derive results Industrial 12 PhD Student Experience 1 year in the company Typical use Automate the application of data minded engineering knowledge Industrial 13 Contractor Experience 1 year and 9 months in the company Typical use Automate conceptual design and simulation workflows Industrial 14 Contractor Experience 14 years in the industry 12 years in the company software Typical use Develop Knowledge Based Engineering applications consultant Academic A1 Graduate student Experience Typical use Aerodynamic side of Multi Disciplinary Optimization Academic A2 Graduate student Experience Typical Engineering use Automate conceptual design and simulation workflows Context questions What is your position in the company What is your professional history What do you use Engineering Automation for See table Do you develop individually or in a team All participants develop individually i e not working on a share
311. y is given by inconsistency Dj Ci Deis m W n 5 1 That is the total inconsistency between the design model D and code model C at moment i is the sum of the weights W for each notification n in the inconsistency notification list L at moment i As an example consider the situation where the design and the code differ by one class that is only in the design with three input slots and two child slots Given the weights in Table 2 the inconsistency is calculated as inconsistency 1 Welass 3 Wattribute 2 Wenhild 39 433425 5 2 22 Part Thesis article 22 5 3 2 Measuring completeness and correctness of documentation The degree to which GenDL design accomplishes the first objective complete and correct documentation will be measured by calculating the consistency between the design and the code at the end of each project relative to what it could have been inconsistency Dena Cena inconsistency Do Cena inconsistency Dena Co consistency 1 5 3 In this equation subscript end refers to the state of the models at the end of the project while subscript O refers to the state at the start of the project i e when the model was empty The consistency will be a number between 0 and 1 0 means there is a complete lack of correct design documentation while 1 indicates that the design documentation corresponds to the code perfectly 5 3 3 Measuring the level of design before code The second objective whether
312. ypes of knowledge reusers and their needs It is argued that 4 different reuse types exist personal and own team reuse reuse by other teams expert seeking novices and secondary knowledge miners e g data mining The present discussion will focus on the first three Common for all types are the stages of knowledge reuse defining the question locating the source human or document selecting information and applying the knowledge What is different between reuse types is the information needed in each stage of knowledge reuse For personal and own team reuse notes can be informal and case specific omit all domain knowledge and can focus on short term issues For reuse by other teams the notes would have to be sanitized The information is often de contextualized to be no longer case specific Some domain knowledge is added but still a working background knowledge is assumed and it is unlikely the authors can be so complete that their presence is no longer required For novices case specific context would in fact be helpful Their main challenge is to judge the relevance of knowledge Tailoring information for another reuse type requires effort How much depends on the knowledge distance between the knowledge producer and consumer To facilitate tailoring intermediaries and incentives are used which ate discussed next Intermediaries Intermediaries are human or technological means to alleviate the workload of the reusers They are r
313. zation becomes harder and cannot be as complete This undermines the incentive provided by code and design documentation generation The boundary conditions used for the development of GenDL Designer the non technological obstacles deadline pressure lack of training etc will hold for most professional end user developers Given that an appropriate design language can be constructed for their particular tasks incremental code and design documentation generation can be applied for them too The findings do not apply to software engineers however Their level of training is so high that either they already use generic software engineering tools with the same scope of GenDL Designer or that they are so experienced that they prefer coding directly Part Thesis article 32 9 Recommendations The recommendations for further research concentrate themselves on three levels GenDL Designer should be developed further to facilitate broader industrial validation GenDL Designer should be extended with testing aspects and a process view to cover more issues in Engineering Automation development and the principle of incremental code and design documentation generation can be extended beyond software and applied to Systems Engineering 9 1 Developing GenDL Designer further GenDL Designer addresses several issues identified in the initial study To further investigate them the user interface of GenDL Designer should be further improved project

Thesis - Repository TU Delft

Contents

Download Pdf Manuals

Related Search

Related Contents