Siberian Division of the Russian Academy of Sciences A. P. Ershov
Contents
1. M Pascal C 2002 INTRODUCTION The methods of compiler and converter as a special case development are sufficiently investigated and described 1 3 There is a number of converters 4 6 from the standard Pascal language 7 to C C languages 8 In this work some translation schemes having the theoretical interest in the authors opinion elaborated for a converter from an essential Pascal extension later M Pascal to C are described The converter has been developed by or der of a large telecommunication company and it must not only translate an in put M Pascal code to the functionally equivalent C code but also meet some requirements M Pascal language extensions and requirements for the converter have essentially influenced on the development of translation schemes In this work the following extensions of M Pascal are considered Variable initialization which is syntactically close to the init2. lt gt amp Here actual parameter a of type AT is being explicitly cast to the type FT before passing it to the procedure p Unwise extraction of a value ofthe type FT from a value of the type AT is taking place and this extracted value is being passed to the procedure This scheme has the right to exist with the assumption that none of the external modules from the outside of the set of files being con verted into C call this procedure Otherwise the following properties of the UNIV mechanism should be considered to propose a proper translation scheme e UNIV parameters in M Pascal are always passed by address e Ifthe actual parameter is an expression then the value is copied to a temporary and the address of the temporary is passed The following algorithm therefore needs to be implemented 1 Get an address memory location of an actual parameter of the type AT 2 Pass the address obtained to the procedure In the procedure 3 Accept the address of the UNIV parameter 4 Dereference the address treating it as a pointer to the value of the type FT 28 Different implementations of the algorithm are possible The one below 1s rather interesting template class FT int FT formal type of a parameter class UNIV Auxiliary class UNIV define once const void value public Constructor to convert from UNIV lt T1 gt to UNIV lt T2 gt template class T UNIV const
3. SET SIZE MAX SET SIZE public SetRange size t from size t to The class SetValues is used to create an object of the type Set and has a con structor with a variable number of parameters this is possible because the Pas cal sets cannot contain negative elements and the value ENDSET 1 can be used as the end of an initializing list The class SetRange likewise is used to create the temporary set which contains the elements of the range Thus during initialization of a set temporary objects of the type SetValues or SetRange are created first After that a temporary set object is being con 16 structed as a union of the constructed temporary set objects and passed as a pa rameter to the BitSet constructor template size t LOW size t HIGH class BitSet public Set lt LOW HIGH MAX SET SIZE public typedef Set LOW HIGH MAX SET SIZE Base BitSet BitSet const Set 0 MAX SET SIZE MAX SET SIZE gt amp other To correctly initialize M Pascal sets to empty the set the default construc tor of the BitSet class is redefined Let us consider an example of set initialization TYPE T SET OF 1 10 VAR V T 1 3 5 VAR V1 PACKED SET OF char translated to typedef BitSet lt 1 10 gt T T V SetRange 1 3 SetValues 5 ENDSET PackedBitSet lt 0 255 gt V1 default constructor is called 3 3 Records Taking into account the encapsulation request M Pascal records are na
4. BitRef unsigned char byte unsigned long bit BitRef amp operator const T amp value BitRef amp operator const BitRef amp value The BitRef class contains a pointer to the byte which contains the element of the array and the bit offset of the element inside of this byte Access to ele ments of packed arrays is implemented through the interface of this class Array initialization There are two approaches to implementation of class constructors that ini tialize an array e constructor with the variable number of parameters e constructor that accepts an object of an auxiliary initializing class Let us consider an example TYPE array 0 5 of integer var 128 256 Whth the first approach the declaration of the constructor looks like Array int num of params where num of params is the number of elements in the initializing list which in the general case is not equal to the number of array elements In this case the remaining elements are initialized by zeros A V 2 128 256 13 This approach uses the mechanism of working with a variable number of parameters refer to macro va arg According to the latest standard of the behavior of the mechanism is undefined if complex types non POD types 8 are being passed through ellipsis That is such behavior of a constructor highly depends on a particular C compiler used The second approach removes this restricti
5. FreeBSD Pascal To C converter ftp ftp freebsd org pub FreeBSD ports i386 packages 4 stable lang ptoc 3 50 tgz MSC Pascal To C C converter http lib buryatia ru cgi bin win SOFTWARE ptoc README txt Jensen K Wirth N Pascal User Manual and Reports Springer Verlag 1978 Program language C standard ANSI ISO TEC 14882 http www techstreet com cgi bin detail product_id 49964 51 B C P A C B 92 25 12 01 15 05 02 60 x 84 1 16 1 8 2 0 50 OOO 630090 6
6. any string var x longint p any string len integer begin p ptr s len strsize s x ptrdiff ptr s nil writeln procla p strings is 1 len endproc 33 var p string string 20 abcd test string begin 1 string endprog code looks like this typedef MString lt 32767 gt any string static void procla MSubString s MSubString is used instead of VAR String type parameter int x 0 any string p NULL short len 0 s Ptr any string len s StrSize cwrite lt lt 1 p strings is lt lt p 1 1 lt lt endl extracting a sub string MString lt 20 gt p_string abcd test string int main void 1 string return 1 4 NESTED FUNCTIONS AND PROCEDURES In contrast to the C language M Pascal allows us to declare nested func tions and procedures Though some C compilers permit nested functions GCC extension it is really hasty to rely on extensions of clever implementa tions For simplicity only nested functions will be considered all further dis courses are correct for procedures as a special case of functions Let us consider the following example FUNCTION outer INTEGER TYPE t INTEGER VAR a b t FUNCTION inner t BEGIN inner inner atb ENDFUNC end inner BEGIN outer 34 ls b 2 outer inner ENDFUNC
7. stmt This means that in the case of straight translation of M Pascal for statement to the above loop will become infinite after compilation by a C compiler To preserve the source M Pascal semantics the following translation scheme Is proposed M Pascal VAR var T FOR var init val TO final val DO stmt T var var init val T last var for last var final val var lt last var stmt For implementation of the DOWNTO for cycle the loop variable increment is replaced by decrement 26 The above scheme does not work in the cases when the loop variable runs the values from the lower to upper value of a type Let us consider the example M Pascal type fc range 0 255 var i fc range for i lower fc range to upper fc range begin functions lower and upper returns correspondingly the low 0 and upper 255 value of the type fc range typedef char fc range const fc range LOW fc range 0 const fc range HIGH fc range 255 fc range i LOW fc range fc range last var for last var HIGH fc range i lt last var i In this example the loop translated will become infinite because semantics of the for statements in M Pascal and C is different When the value of the loop variable i becomes 255 the loop body executes but after the succeeding increment of the variable i its value becomes 0 which satisfies the for condi
8. code forces us to use C classes to simulate the types mentioned Fur thermore it would be desirable to secure the internal data of these classes making it private this would conflict with the requirement b on using the list of initializers 10 e There is no possibility to specify an interval range in a list of initializ ers in C This means that initialization of sets 1 VAR a SET of char a z will require a special constructor to be defined this would conflict with the requirement a on using the list of initializers e M Pascal records are naturally converted into C classes structures Taking into account that in the general case a record can contain a field of type SET which has a constructor it comes out that this record can not be initialized by a list of initializers being of a non aggregate type The following example illustrates this struct Set Set class Record Set field r Set error non aggregates cannot be initialized with the initializer list Since we are initializing the C classes which correspond to the M Pascal records by constructors all the fields of these classes should have a constructor at least a default one That is for every structured type in M Pascal a C class is being introduced All such classes have the following properties e Internal representation of data is closed e A convenient and similar to the origina
9. 4 x unsigned char c 5 T n access methods attribute packed sizeof T1 6 bytes 18 It is just enough to add an unnamed field to preserve data layout class T1 byte offset byte size unsigned char a 0 1 char 8 1 1 int b js 2 5 unsigned char c i 6 1 access methods attribute packed sizeof T1 yt 7 bytes The packed records without variant parts are almost similarly translated The type T6 from Example 1 6 can be translated into C in the following manner class Tolshort at3 short b 3 char 2 of 2 bits Short er3 short short e 3 access methods attribute packed 19 bits The size of T6 without attribute packed is 4 bytes char 2 unnamed bit field Data Layout Preservation for records with variant part From Example 3 3 2 one can see that the sizes of the type T9 in M Pascal and in C are different In this case attribute packed does not help us since the fields of non integral types in are allocated in separate bytes see DLP chapter One of possible solutions is to move the UNION construction to the external level class T9 1 union bit offset struct boolean a 1 0 boolean b 1 1 unsigned char c 6 Ls 2 unsigned char d 7 8 attribute packed Struct i boolean 1 s 0 boolean 1 1 unsigned char 6 2 unsigned char e 7 ye 8 _attribute__ packed a
10. 7 byte Array record is word aligned but the size of a record can have an odd num ber of bytes i e word alignment of a record type is executed only if this type is used in other record or array types and only if it is not the last field in record types Example 1 2 T2 record a integer b char end sizeof T2 3 bytes T21 record a b T2 sizeof T21 7 bytes T22 array 1 2 of T1 sizeof 722 8 bytes C The fields of classes are allocated in successive bytes from low to high memory addresses and follow the same alignment and size rules as their corre sponding types require Any field that requires more than one byte and less than or equal to 2 bytes e g short is always located at an even offset from the start of the class Example 1 3 struct T3 byte offset byte size char a E 0 1 Ny short b A 2 2 char c 4 d sizeof T6 6 bytes Any field that requires more than 2 bytes is always located at an offset which is a multiple to four from the start of the record Example 1 4 struct 4 byte offset byte size char a 0 1 int b ie 4 4 char ic 8 1 sizeof T7 12 bytes Packed data types M Pascal No field will be allocated starting in other than bit 0 of a byte 1f this allocation would cause the field to extend into the next byte Example 1 5 T5 packed record a b c d e 0 7 end 19 bits Sizeof T5 3 bytes The field c e is entirely c
11. PARSER M Pascal files SUPREVISOR PASFEO amp PASFE structures IR graph PASFEO amp PASFE generation methods Supervisor works with a configuration file command line and OS environ ment and calls other components Parser consists of lexical syntactic and semantic analyzers Parser transforms text files containing M Pascal modules into the internal representation IR graph IR graph is an acyclic graph with nodes which are instances of PASFEO 46 and PASFE structures The program parse consists of the following parts the main algorithm which is a syntax analyzer reads lexemes from the specified input stream by invoking the lexical analyzer algorithm and parses M Pascal declarations and statements according to some rules declared in y files For coding the parser stage of the converter GNU Bison and Flex utilities were used The input grammar of M Pascal was written as the rules of the utility Bison and then were converted into the C code to build it into the whole converter Parsing a current statement this algorithm also processes semantic attributes For example the semantic attribute of a constant declaration construction is a con stant value PASFEO module The PASFEO module abbreviation PAS denotes Pascal abbreviation FEO denotes Front End Objects 1s a set of structures intended to supply all types data and functions necessary for checking scopes visibility
12. UNIV lt T gt amp a memcpy this amp a 4 Constructor that gets an address of an actual parameter template class AT UNIV const AT amp a value amp a Type cast operator used to cast from UNIV lt T gt to T type operator FT amp return FT value void p UNIV lt FT gt Casting from UNIV lt gt to UNIV lt FT gt is performed FT a a Dereferences the value member p UNIV lt gt a Initializes through the value member of UNIV lt gt with an address of a Besides the comments in the code there are the following important notes e Semantics of passing UNIV parameters is preserved here e A temporary variable a is used to accept the address of the actual pa rameter Then it is being dereferenced initializing a local variable a In the case when it is a VAR UNIV parameter this initialization will be converted to void UNIV lt FT gt FT amp a a reference sign added To finalize the discussion the following caution is quite appropriate UNIV parameters are unsafe and often unnecessary Of course they are left for back ward compatibility However we would not recommend using this technique in the subsequent development based on the converted code 29 3 7 Strings and Sub Strings One of the differences between M Pascal and Standard Pascal is a double no tation of the type String M Pascal holds two concepts
13. extend grammar if we agree to restrict the form of construc tions enclosed in CCCs but in our case the source code would need a lot of changes In our approach we slightly restrict the use of CCCs but the grammar is left unchanged All CCCs are divided into two types translated and not translated One can manually change not translated CCCs so that they become trans lated Here is an example of the CCC which cannot be translated A Sendif Sendif will be successfully parsed but an error message will be given when C is parsed Definition 1 Let us call a simple CCC the one of the kind if boolean expression any text 1 program fragment without CCC Example contains two simple CCCs 40 Definition 2 In CCCs of the kind if boolean expressionl any textl 2 if boolean expression2 any text2 program fragment without CCC any textl and any text2 are alternative to each other if 1 they are derived from the same non terminal or grammatical sequence in the same grammatical construction or 2 any_textl and any_text2 are declarations and they declare the same identifier Definition 3 Two CCCs of kind 2 in which any _text1 and text2 are alternative are called CCCs with alternative CCCsA Two CCCs in which any textl and any text2 are not alternative are called successive simple CCCs Example 2
14. mented in some compilers The second mode is necessary for debugging pur poses only since it is impossible to obtain any debug information in the third mode Additionally it takes much time for the converter to format the output C code because of complicated requirements stated This essentially increases the converter run time on a large number of input files So an additional option has been introduced into the converter which allows us to switch off the formatting thus reducing translation time for debug purpose 50 CONCLUDING REMARKS We have described some translation schemes from extended Pascal to C which are of the most interest in our opinion The schemes satisfy the require ments which are essential for preservation of functionality of a large software hardware system after replacing a program language and or a part of hardware We tried to show different approaches to the development of these schemes and we hope that it may be useful for other developers REFERENCES Aho A V Ullman J D The Theory of Parsing Translation and Compiling Vol 1 Parsing Prentice Hall Englewood Cliffs N J 1972 Aho A V Sethi R Ullman J D Compilers Principles Techniques and Tools Addison Wesley Reading Mass 1988 Kasyanov V N Pottosin LV Methods of compiler development Novosibirsk Nauka 1986 in Russian GNU Pascal To C converter http www ibiblio org pub Linux devel lang pascal ptoc_3 34 tar gz
15. of named program objects like functions and types This module serves for storing and retrieving the name scope information during the source code parsing and for holding information about the named program objects used in M Pascal pro grams Parser calls it when a new object in the program has to be created or ex isting one is queried by name All PASFEO structures are inherited from one parent PASFEO Object Each PASFEO object has a name member storing the name of this object It is possible to determine which construction represents a certain PASFEO object For this purpose a PASFEO object has a set of member functions used by the syntax analyzer PASFE module The PASFE module abbreviation PAS denotes PASCAL FE denotes Front End is a set of structures intended to supply all types data and functions necessary for so called Front End processing i e for generation of Internal Representation IR from the parse tree The module PASFE describes all possible kinds of IR graph nodes and information contained in them The IR graph is a directed acyclic graph in simple cases a tree obtained after parsing the source M Pascal program code Most of the nodes of this graph or tree cor respond to some clauses syntax constructions of M Pascal comments and some compile time facilities of the M Pascal language environment Thus there are nodes for type label constant variable declarations procedure and function dec larations statements
16. on the whole clarify the functionality of nested functions 5 CONDITIONAL COMPILATION 5 1 Translation scheme M Pascal provides several language features not to be found in the standard Pascal One of them which is required to be kept at translation is conditional compilation To arrange Conditional Compilation M Pascal has directives if and Send Sif boolean expression any text 38 Any program code and other directives may be enclosed within these two di rectives Let us call the pair of conditional compilation directives and a program code between them Conditional Compilation Clauses CCC In addition there is a set directive Yoset identifier expression which is used to assign a constant value to a compile time identifier The difference be tween M Pascal conditional compilation and C C one is in the following there is no preprocessor in the compiler of M Pascal the compiler calcu lates a boolean expression and compiles or not the code inside CCC e as a result in expressions of CCC both the CCC variables and program constants including constants of enumerated types can be used For ex ample const a 1 Sif a a is a program constant set b 10 b is a CCC variable set true a is a new CCC variable its value will further be used in expressions of CCC and only in them Send The M Pascal conditional compilation directives are translated directly into
17. operators etc and for STACKSIZE INCLUDE IF END and SET compile time facilities Aside from nodes corresponding to syntax con structions there are some auxiliary nodes regarding to the code generator needs Each node represents a structure with pointers to other nodes and or program objects Nodes also contain information obtained at the scope resolution stage particularly the pointers to named and unnamed program objects i e types vari 47 ables constants functions etc The same node may correspond to different syn tax constructions depending on the value of one of its members Codegenerator All classes corresponding to IR graph nodes have a com mon parent the class OM The class OM encapsulates information about an object such as its location type of lexeme etc It implements memory allocation functions to allocate memory in a more definite and effective way This class has a virtual method virtual void Iterate Translate amp msg which performs an action over an object The action is described within msg function object The msg parameter holds the translation message output file name buffers and so on The method is virtual and 1s overridden for every class of objects individually A set of Iterate methods is Codegenerator itself If the object contains other objects 1 e it is a container the method applies msg to these nested objects too thus propagating the action through all levels of the ob ject st
18. pointers to them As the pointer type is the POD type and has the fixed size 4 byte such changes can resolve the problem with macro va_arg In general terms this new approach complicates implementation a bit and increases the cost of initialization t10 boolean variant a va_list Marker va_start Marker a variant_a switch variant_a case true b va_arg Marker short c va_arg Marker uchar switch c case 0 d va_arg Marker boolean break default break case false f va arg Marker uchar break default va_end Marker The constructor call looks like the following 24 fdefine INIT type value auto ptr type new type value get T10 var rec TlO true INIT short 1 INIT char 0 INIT boolean true 3 4 Bound The BOUND attribute in M Pascal indicates that the variable will be associ ated with another allocated variable or memory reference at run time through the use of the built in BIND procedure No storage is directly associated with the BOUND variable except a pointer to the variable to which it is bound Only local variables of procedures or functions can be declared with the BOUND at tribute After the bind procedure has been executed any references to the BOUND variable will reference the store occupied by the variable it was bound to This variable may be of any type The actual BIND ing may appear anywhere in the procedure bo
19. public Construct MSubString from literals MSubString MSubString const char amp ch MSubString const char str Construct MSubString by binding it to an existing string MSubString const MSubString amp other MSubString const char address long from long length substrings MSubString operator long index long sub length assignments template int gt MSubString amp operator const MString M amp other any string MSubString amp operator const MSubString amp other any sub string MSubString amp operator const char str any literal MSubString amp operator const char ch any char aggregate comparison template int M bool operator const MSubString M amp other template int M bool operator const MSubString M amp other Concatenation MSubString operator const MSubString amp other any string MSubString operator const char s any literal MSubString operator const char amp ch any character Pascal functions const size t StrSize const long StrScan const MSubString amp pattern The class MSubString denotes the type of a string sub string whose actual size 1s not possible to be calculated at compile time For example if a string 1s passed into a routine as a VAR parameter then it is being cast to the MSub 32 String type to provide a safe mechanism of manipulation with the string at run time MSubSt
20. restrictions on readability of the target C code 35 e Nested constants moved out as in the case of types can result in the name conflict which is avoided by the same methods as above An al ternative would be passing the constants as additional parameters of constant reference type as compared to variable passing e The syntax of a nested function call changes The additional parameter list 1s required which is not a problem for the converter but becomes a problem for future maintenance of the C code Usage of C namespaces for the nested structure preservation The structure of nesting is described in terms of C namespaces which can be nested in turn There is a namespace defined for each function which auto matically includes the namespace of the parent function At the beginning of the namespace description every function is declared using its own namespace with the help of using directives short outer namespace namespace outer typedef short t t a b t inner namespace namespace inner namespace outer t namespace outer inner using namespace namespace inner return atb short outer using namespace namespace outer 1 2 return inner Regardless of the fact that namespaces allow us e preserve the logical structure of the source code e with the name encapsulation to avoid any name conflict which is al most inevitably originating from de nesting e to achie
21. the C ones The set directive is translated into define directive set debug TRUE define debug ccc TRUE The postfix ccc will be used to distinguish program constants from CCC variables since the preprocessor having executed define directives would oth erwise replace the program constants The value of a CCC variable may be changed by a later set command To avoid the warnings of GCC about rede fining constants in define directives the scheme of translation can be slightly changed Sset debug TRUE E fdefine debug ccc TRUE set debug FALSE E fifdef debug ccc fundef debug ccc endif define debug ccc FALSE 39 The directives 3if and ena are translated accordingly into if and endif directives if boolean expression any text fendif 5 2 Translated and not translated CCC Directives of the conditional compilation are external constructions of the source language but not members of the input grammar However for preserva tion of CCCs in the target code it is necessary to parse all CCCs and to build their internal representation The extension of the source language grammar con sidering an arbitrary arrangement of CCCs is impossible inside CCCs there can be enclosed not only a complete syntactical construction derived from a non terminal but also any sequence of terminals Example 1 Sif debug if a 1 then begin end b c d e Sif debug end end It is possible to
22. variant_a tag value receiving switch variant_a according to the tag value the number and the order of receiving of the last parameters is determined in switch operator case true b va_arg Marker short c va_arg Marker uchar switch c case 0 d va_arg Marker boolean break default break case false 23 f va arg Marker uchar break default va_end Marker Within this approach the beginning of the parameter list relates to the fix part fields and to the tag field Further all parameters are treated by C mecha nism of working with the variable number of parameters The use of this con structor requires that initial values for all elements of a record are to be specified explicitly in the constructor call For this purpose each incomplete initial list in the source code is extended by the default values of corresponding elements dur ing translation to C The approach described above just as in the case of array initialization uses the mechanism of passing the variable number of parameters va_arg macro whose behavior is undefined when passing objects of non POD types according to the C language standard In the general case the record can contain any number of fields of such types In order to eliminate dependence on a C com piler implementation the scheme of translation can be slightly changed It is pro posed to pass not the initializing objects themselves but the
23. E Boolean OF True name STRING 10 False ENDREC one true ONE farray Ce ABRAY 1 55 OF char set D PACKED SET OF 1 10 1 3 5 It should be noticed that the objects both global and local not initialized ex plicitly will be set at zero implicitly by M Pascal compiler There are two major possibilities of object initialization in C using an ob ject class constructor and using a list of initializers this feature is rather an in heritance of C Generally an object of the class T may be initialized using the list of initializers if only a T does not define constructors b all members of T are public c T does not have base classes d T does not have virtual functions Therefore only the classic C structures can be initialized by the list of initial izers Moreover according to the standard of C 8 only the global scope ob jects are initialized by zeros while the nested scope objects initial value 1s unde fined if not specified explicitly The syntax of the list of initializers in M Pascal and C is much alike which is an essential argument for the constructor less approach However there are a number of notes which make such a scheme impracticable e presence of the following structured types in M Pascal o variant records sets arrays o strings and sub strings and demands to preserve Data Layout semantics and appearance of
24. SIF MODE MAIN table infor table type SEND SIF MODE DEBUG table infor ARRAY debug files OF select record SEND SIF MODE RELEASE table infor ARRAY release files OF select record SEND These CCCs are CCCsA because they declare the same identifier ta ble infor Example 3 table infor IF MODE MAIN lst CCC table type SEND SIF MODE DEBUG 2st CCC ARRAY debug files OF select record SEND SIF MODE RELEASE 3st CCC ARRAY debug files OF select record SEND 41 These CCCs are CCCsA because they are derived from the same non terminal type denoter in the same grammatical construction vari able declaration see below Example 4 SIF MODE DEBUG wait time EXTERNAL DEFINE relative time 0 0 6 0 SEND SIF MODE MAIN help pid process id SEND These CCCs are two successive simple CCCs because they declare different identifiers The main problem 15 to recognize if any _text1l and any text2 in 2 alternative to each other or not It is not a problem to translate the CCCs similar to Example 1 Example 2 and Example 4 if there are no CCCs similar to Example 3 Example 3 is bad because parser does not know if the second third CCC is alternative to the first second one or it is a new CCC declaration successive CCCs It may be solved for some cases but in the general case especially in expressions it is impossible grammatical
25. Siberian Division of the Russian Academy of Sciences A P Ershov Institute of Informatics Systems V A Markin S V Maslov R M Novikov A A Sulimov EXTENDED PASCAL TO C CONVERTER Preprint 92 Novosibirsk 2002 In this work the main translation schemes elaborated for a converter from an essential Pascal extension to C are described The converter has been devel oped by order of a large telecommunication company and it must not only trans late an input M Pascal code to the functionally equivalent C code but also meet some requirements M Pascal extensions and requirements for the converter have essentially influenced on the development of translation schemes A P Ershov Institute of Informatics Systems 2002 II 92 2002 KOH
26. ambiguity Let us consider the following grammatical rules variable declaration identifier list type denoter initial value list identifier list identifier identifier initial value list constant initial value initial valuej initial value expression index expression expression conditional statement if statement case statement if statement IF Boolean expression THEN statement ORIF Boolean expression THEN statement else part else part ELSE statement case statement CASE case index OF case list element case list element OTHERWISE statement sequence END In each rule all non terminals and grammar sequences can be divided into 42 tive and non iterative Iterative non terminals grammatical sequences enclosed in curly braces t Examples of iterative non terminals grammar sequences are identifier in identifier list initial value ORIF Boolean expression THEN statement grammar sequence case list element Assertion 1 If two or more Conditional Compilation Clauses are alternative and contain program fragments derived from iterative non terminals grammatical se quences then they can be translated into C CCCs successfully 2 Any simple CCC containing any program fragment can be translated into C CCC successfully 3 CCCs meeting the syntactic property or
27. ccess methods sizeof T9 2 bytes 19 bit orr orre size a ua 7 Wf f Some remarks e attribute should be used exactly as shown e number of unnamed fields in structures can be reduced by increasing their size e g in the second structure three unnamed fields can be re placed by one char 8 e there are no nested unions all enclosed variant parts are transferred to the top level For example 10 record CASE a boolean OF TRUE b integer case c 0 1 of 0 d boolean 1 e char FALSE f char end T10 is translated into the following class T10 union struct 4d boolean a 7 char 8 additional field for word alignment short b unsigned char c boolean d struct char Filler 5 five bytes unsigned char e struct char 8 unsigned char f access methods If in a target hardware the most significant bit has number different from that in a source one the fields contained in each byte should be enumerated in 20 inverse order to preserve the bit order in a byte For example the type T6 will have the following form class T6 char 2 short b 3 short a 3 first byte char 2 short d 3 short c 3 second byte char 5 short e 3 third byte hus access methods attribute packed So by using e attribute packed when a record has a field whose alignme
28. d outside any CCC statement an empty CCC stack In this case the converter issues a warning and chooses the first found object In our example if there are no CCC directives around the call statement of the procedure proc the converter produces a warning on this line because it can not find any object with the name A and empty CCC Set However it will continue assuming that nothing depends on the type of A In the general case this may lead to erroneous under some conditions set output code It should be recommended that a named object be used under the same set of conditions as it is declared In our example it must be something like the following 45 Sif exprl var A integer 1 tend Sif expr2 var A longint 2 tend Sif exprl proc A Sif expr2 proc A Send routine call with object 1 send routine call with object 2 The approach described allows us to realize all the theoretical issues of translation and to translate the majority of the source M Pascal code containing CCC directives 6 IMPLEMENTATION OF THE CONVERTER 6 1 The general structure of the converter The converter consists of the following modules supervisor parser PASFEO structures PASFE structures and code generator The modules of PASFEO and PASFE structures can be considered as libraries configuration file configuration file parser command line parser command line i
29. dy and is valid until execution reaches the next BIND or returns from the procedure Thus a BOUND variable can be translated only to a pointer rather than to reference since a reference may be initialized only once The following translation scheme is proposed M Pascal bvar BOUND type type bvar Further each call BIND bvar ref Is translated into bvar type amp ref Example TYPE SomeType RECORD f INTEGER END VAR bvar BOUND SomeType BEGIN BIND bvar recl WITH bvar DO SomeProc f END Is translated to class SomeType short f SomeType bvar 25 bvar SomeType amp recl someproc bvar f 3 5 For statement The for statement indicates that a statement is to be repeatedly executed while a progression of values is assigned to a variable called the control variable of the for statement It is an error to have an assigning reference to the control variable within the repeated statement Evaluation of the lower and upper bounds of a loop is performed once before the execution of a loop body in M Pascal Let us consider the following example VAR I J INTEGER J 710 FOR I 1 TO J DO BEGIN 9 9 1 WRITELN I j END This loop will stop after 10 iterations To the contrary in the for statement the for condition for cond and expression are performed every time after stmt is performed CRES for init stmt for cond pt expr
30. eadability of the target C code and more exact source seman tics for each nested class it is possible to create a private wrapper method in parent class engaged only in correct initialization of the object of the nested func tion and call of the execute method the operator in our example class outer typedef short t local type ta b local variables class inner outer amp outer public t operator 37 inner outer amp outer outer outer friend class inner t inner return class inner this public short operator outer short outer return class outer outer t outer inner operator inner body return _outer a _outer b short outer operator outer body 1 b 2 return inner Certainly such a scheme is the fullest in the sense of saving the program logical structure in comparison to the de nest approach and the program func tionality in comparison to the namespaces approach Nevertheless we can list the following rather negative specific properties of this approach e rather difficult class implementation simulating nested structure e losing of simple access to the objects of parent functions this approach requires a full name specification The last issue can be considered as positive because the names of objects of nested functions do not overlap the names of parent functions and
31. end outer Three translation schemes of conversion are possible e De nesting moving the nested functions to the global level e Usage of C namespaces for preserving the nested structure e Simulation of nested functions by nested data structures De nesting functions All nested functions are moved to the global level file scope as required by the C standard All the variables used in nested functions and not declared at the file scope are passed to these functions by references as additional input pa rameters typedef short t t inner t amp a t amp b return b short outer t a 1 b 2 return inner a b This translation scheme apparently is the most obvious conversion resulting in a desirable functionality A little less obvious are the consequences of such a conversion e type declared in the scope of an upper level function requires mov ing it to the global level if it is used at least in one nested function or it is used as a template parameter in some type declarations e g the type of an array element see Arrays This can easily result in a name con flict at the global level if two or more nested types were of the same name and were both brought to the global level Conflict is avoided by the use of automatic unique names for types moved or by manual modi fication of the source Pascal code The latter appears to be more accept able with the
32. erate method for the root node of IR graph prepared by Parser in the figures above the root node is omitted The root node of the IR graph is al ways a structure of PASFE CompilationUnit type When Codegenerator finishes with the sample IR graph the following C code is generated const int a 1 const int b 7 It is possible that the name of an identifier in an M Pascal program coincides with a C keyword To avoid this conflict the user inserts all C keywords reserved identifiers the names of the library functions etc in a special file According to the requirements to the converter all M Pascal identifiers are con verted in the lowcase format When the converter meets the identifier that corre sponds to some C language identifier from the special file it converts the first letter of the identifier in the uppercase That allows us to avoid errors at the C code compilation stage 49 6 2 Operation modes of the converter The converter can operate in the so called batch mode by defining the set of input M Pascal files This set can be specified by wildcards in the command line According to the indicated template the converter searches for the corresponding files in folders specified in the configuration file and organizes the list of found files Then for each file from the list the parser and codegenerator run As M Pascal allows separate compilation of different modules the M Pascal software package consist
33. f a data type up to the boundary of some unit of memory byte word 2 bytes or dou ble word 4 bytes Depending on a processor data layout can be such that e the most significant bit can be either the highest or the lowest bit in a byte e the most significant byte can be either at the highest at the lowest ad dress in a word The implementation of DLP allows ensuring compatibility of modules of composite software hardware systems when some components are being modi fied re engineered for example when some modules written in one program language are translated into another language or one part of hardware is replaced by another one Further in this part we describe the main differences in data layout and align ment between the M Pascal compiler on M68000 platform and the GCC GNU compiler on SPARC Station Unpacked data types M Pascal the boolean type is allocated in a byte GCC the standard bool type is allocated in 4 bytes For DLP of the boolean type typedef unsigned char boolean n is used M Pascal Any field that requires more than a byte integer 2 bytes longint 4 bytes pointer 4 bytes set 32 bytes the enumerated types with the num ber of members more than 256 is always allocated at an even offset in bytes from the start of the record Example 1 1 1 record byte offset byte size a char b longint c char 1 4 1 end S 0 2 6 Sizeof T1
34. fter that at the time of destruction the actual bit object is being updated by the value from the BitRef s copy In this particular case we shall access the field of the type PackedAr ray lt boolean 1 1 3 gt which starts in the first byte at the offset of 3 bits 21 class T7 bit offset bit size char a 3 0 bit do bre c char b 3 3 bit 4 bit public BitRef TO 3 gt B return BitRef TO 3 gt char this 3 sizeof R 1 bytes The method B is used to access the field b of type char as it were of the packed array type 0 With the translation scheme proposed the access to the field b in M Pascal var rec T7 rec b true will be translated into T7 rec reo B access iiL true Note that the access method of the BitRef class is used to extract an object of the type T PackedArray lt boolean 1 1 3 gt in this particular case from the object of the type BitRef lt T W gt Now the class T7 preserves the data layout of the original record of the type T7 This technique may be either automated or performed manually depending on the number of occurrences of these bad types in the source code Record initialization As discussed in chapter 2 the initialization of variables of class types corre sponding to records is carried out through the constructor calls For this purpose two constructors are introduced in a C class e The default constructor for the case without an
35. ialization in the C language but has slightly different semantics Conditional compilation which differs from the conditional compilation in C C in the program constants usage and absence of the preproces sor UNIV type in a formal parameter list This causes type checking to be suppressed that is both the calling and called routines must declare a type for each parameter but the types do not have to be compatible The type STRING n is a special case of PACKED ARRAY 1 n OF char M Pascal contains predefined functions and procedures for string manipulation and a special notation for denoting sub strings Statements or declarations can be copied from a library by using the in clude option Constant expressions e g 5 3 can be used wherever constants are al lowed The strict ordering of definitions in the standard Pascal has been weak ened Array and record types that do not contain files can be compared for equality and inequality Integers are represented by two different predefined types integer and longint Dynamic arrays allow the specification of array bounds at run time The most important requirements to the converter are the following e Data Layout Preservation binary correspondence between M Pascal and C data types e preservation of conditional compilation directives e textual similarity of M Pascal and codes e g textual representa tion of array index expressions the ranges and index t
36. in a call even if the formal parameter has a different declared length The actual parameter is assigned to the for mal parameter so padding or truncation can occur if the actual and formal parameters have different lengths The length of the string in the called rou tine is always the declared length of the formal parameter For simulation of the above indicated properties the class MString is used 30 template int N class MString public Array lt char 1 N gt public Construct MString from literals and other MStrings MString MString const char amp ch MString const char str template int M MString const MString lt M gt amp other MString const MSubString amp other MsubString class will be described below type casting operators DynArray lt char gt amp dynamic array by reference operator MSubString MString parameter by reference substrings MSubString operator long index long sub length assignment template lt int M MString amp operator const MString M amp other any string MString amp operator const MSubString amp other any substring MString amp operator const char str any literal MString amp operator const char ch any char aggregate comparison template lt int M bool operator const MString lt M gt amp other template lt int M bool operator const MString lt M gt amp other Pascal functions fo
37. initializing list It initial izes all data by zero just as in M Pascal e The initializing constructor with a variable number of parameters It re ceives the number of parameters known for each particular record and related to the fixed part fields in the record After that it receives all the other parameters through ellipsis to initialize the fields of vari ant parts of the record 22 Records without variant part M Pascal allows initialization of a part of a record an initial list can contain a number of values less than it is necessary to initialize all the elements in the record For example in the following initialization of a variable of the type TI only the fields a and b can be initialized VAR var rec 1 10 To properly handle initialization in the target C code it is necessary to ini tialize the rest fields by their default values in the constructor definition class Tl unsigned char a int b unsigned char c Sue access methods T1 memset this 0 sizeof T1 T1 unsigned char a 0 int b 0 unsigned char c 0 _ b b c c_ Records with variant part As the amount of fields and their types in variant parts can differ the initial izing constructor for variant part would receive the variable number of parame ters Let us consider the init constructor for the type T10 T10 boolean variant a va_list Marker va_start Marker a
38. l interface is provided i e the assignment operators for all types arithmetic operations for sets and so on e Special constructors are defined to handle the initialization This approach to initialization gives us a guarantee that local variables not initialized explicitly would be implicitly initialized by zeroes by a call of a default constructor which is a subject to be defined in every such class 3 TRANSLATION SCHEMES 3 1 Arrays Arrays of M Pascal are not different from those of the standard Pascal nei ther syntax nor semantics is changed It was decided to translate M Pascal arrays to C classes on account of the following see Initialization chapter as well 11 e there is no Packed Arrays in C e there is a requirement to preserve the data layout in arrays see DLP chapter e as distinct from there is a bound checking mechanism for the M Pascal arrays e target C code would preserve the text notation of array index ex pressions e itis necessary to preserve the assignment operator and the comparison operation of M Pascal arrays The template C class which models the M Pascal arrays is defined as fol lows template lt class T long L long bool is word aligned class Array uchar storage H L 1 is word aligned amp amp sizeof T 2 sizeof T 41 sizeof T public T amp operator long index const T amp operator long index const bool operator co
39. n it at the time of analysis of a CCC comment an additional parser run This parser creates an internal form of an expression contained in the CCC The resulting parse tree of the expression is then being attached to the definite lexeme and after that to the object of an Intermediate Representation IR of a source pro gram Later at the code generation stage for every object in the IR the code generator checks if there were any comments or CCC comments attached to the object and output them properly into the target code The other task arising during construction of the IR of the source program is to set up a correct correspondence between named objects in the IR variable constants etc Let us consider an example Sif exprl var A integer 1 tend Sif expr2 var A longint 2 tend proc A any routine call During construction of IR for this source program the following should be emphasized e Two named objects variables with the same name A are being cre ated Having the same name they differ in their types as a matter of fact Each of the two objects is represented by a node of IR Variable A declaration node 44 To preserve the semantics of the routine s input parameters only one out of two existing A s should be considered For example in the call of proc it is necessary to know the exact type of A to generate appro priate typecasting operator in the target C code The f
40. nst Array amp other const bool operator const Array amp other const Array amp operator const Array amp other Here the template parameters are e class is array s element type e are lower and upper array bounds correspondingly e is word aligned is set to TRUE if the array element is word aligned and the element type is of odd size see Example 1 2 Packed arrays An array with the element size more than 8 bits 1s not being packed If the element size is smaller than one byte then it occupies the minimal amount of bits divisible by exponent of 2 required to contain values of the array element For example the 3 bit size element occupies 4 bits the 5 bit size element 8 bits The signature of the C class for packed arrays is the following template class T unsigned long B long L long H unsigned long W 1 1 2 2 B lt 4 4 8 class PackedArray uchar storage H L 1 W 8 1 H L 1 W 8 H L 1 W 8 J 12 public BitRef T B operator long index bool operator const PackedArray amp other const bool operator const PackedArray amp other const Because of impossibility to address a bit within a byte in a packed array an auxiliary BitRef class has been introduced template class T unsigned long W gt class BitRef protected unsigned char byte a pointer to the byte unsigned long bit bit position within the byte public
41. nt differs in M Pascal and C for example field b in type T1 e additional empty fields char n n lt 8 for byte alignment and char 8 for word alignment see types T6 and T9 itis possible to preserve data layout for almost all records except those considered in item 4 of chapter 1 see examples 1 7 and 1 8 DLP for the records with fields of non integral types The translation scheme described does not preserve data layout for packed records similar to the type T7 i e for records containing a field of a packed re cord or array of size less than 8 bits allocated in the same byte with other fields see example 1 7 To translate such records we shall treat the non integral types of fields as in tegral types For example the bit field b in the type T7 is defined as follows class T7 bit offset bit size char a 3 0 bit 3 bit char b 4 3 bit 4 bit sizeof R 1 bytes we have to do now is to implement some interface inside the class T7 which will allow us to access the field b as it were of packed array type For this purpose we shall use an auxiliary class BitRef the same one we used to access an element of a packed array class see Arrays An object of the type BitRef is used to extract the value of the bit object being referenced at the time of its Bi tRef construction All the modifications are carried out over the temporary Bi tRef object which keeps a copy of the actual bit object A
42. of a string the standard Pascal string which is of the type PACKED ARRAY 1 n OF char and the M Pascal string which is of the type STRING n The correspondence of a character string with a string type 1s set by relating the individual characters of the character string to the components of the string type A variable of the type STRING n has all the properties of a variable of the string type PACKED ARRAY 1 n OF char In addition it can be assigned to other variables of the predefined types STRING char or PACKED ARRAY 1 m OF char even if the declared lengths differ m lt gt it can be compared with other variables of the predefined types STRING char or PACKED ARRAY 1 m OF char even if the declared lengths differ m lt gt n sub strings can be extracted from it strings can be modified by assignment into sub strings Example VAR a b STRING 10 5 hello assigns string from char b a assigns string from string b 1 3 Hello extracts 3 symbols starting from the first one in b assigns into sub string of b writeln b would output Hello it can be passed as a variable parameter to a procedure or function even if the formal parameter has a different declared length In this case the size of the formal parameter is ignored instead the actual length of the string being passed is used within the called routine it can be passed as a value parameter
43. ollowing technical solutions are proposed to achieve the proper con struction of IR and successful code generation l A so called global CCC stack 1s introduced A pointer to the parse tree of a conditional expression 1s pushed onto the stack when encountering an if directive The pointer is being popped out as soon as an end directive is encountered During construction of a semantic named object corresponding to some named object of the source program variable constant etc not IF node the current state of CCC stack 1s being copied to a special stor age inside this named object Thus for every named object we obtain a set of conditions CCC Set under which this object was declared In the above example two objects of the variable A are created The CCC Set of the first object contains expr and of the second one expr2 All the objects with the same name A in our example declared on the same scope are collected into one list A scope descriptor contains a pointer to the head of this list Now to set a correspondence between a name and an existing object in the current scope the converter searches for the pointer to the list of objects with the name specified and then if the list is found searches for the object with CCC Set corresponding to the set of conditions in the global CCC stack its current state It is theoretically possible that an object has several declarations under differ ent CCC directives but is use
44. on by using a temporary object for initialization This object simulates an input stream for a variable number of initial values of the array elements Once the temporary object is filled with the initializers it is being passed to a constructor of Array For these purposes a supplementary interface 1s added to the class Array class Array typedef Init lt T 0 L H is word aligned Init public Init operator const T amp elem Array const Init amp ar In this case the array initialization in C looks like A V AQ lt lt 128 lt lt 256 Both approaches do not conflict with each other and are implemented within one template class Array 3 2 Sets A set type is by no means a new type in M Pascal but it is still worthy to de scribe the translation scheme for this type because of the global requirements Data Layout Preservation see DLP chapter and initialization handling TYPE T SET OF char A set is a well known data container therefore there are many different im plementations of it Probably the most famous of them is the template class std bitset from Standard Template Library STL As there is no proper mecha nism to handle Pascal like initialization and STL library is highly system dependent some bitset implementations restrict the size of a set to be a multiple of four bytes a new class is proposed to simulate M Pascal sets 14 The base class for all set implementations is the
45. ontained in the second third byte GCC Bit fields of some types e g short can be allocated in two bytes in the same word Example 1 6 struct T6 short a 3 short b 3 short o 3 7 short d 3 short e 3 15 bits sizeof T6 2 bytes Two bits ofc are allocated in the first byte the third bit in the second one M Pascal A packed record or array of the size less than 8 bits can be allocated with other fields in the same byte Example 1 7 TO packed array 0 3 of boolean 4 bits T7 packed record bit offset bit size ac i0 sh 0 3 bs 5507 3 4 end sizeof T7 1 byte GCC C always allocates the class array in a separate byte starting in no other bit than 0 it doesn t allow specifying any bit field for a field of type class array or their part Example 1 8 boolean 01 3 struct T8 bit offset bit size short a 3 pe 00 3 k T01 b 72 8 4 sizeof T8 2 bytes Translation schemes have been developed taking into account these data lay out differences 2 INITIALIZATION M Pascal In contrast to the standard Pascal variables in M Pascal can be initialized at the time of their declaration that is an initial value can be specified explicitly for variables of any type at any scope of declaration global scope or proce dure function scope For example VAR built in data type A integer 1 record with variant part B RECORD ID one two three CAS
46. property and only them are valid for translation Example 5 sif boolean expr IF expression THEN BEGIN END statement sequence sif cond_expr END SEND The CCC is simple therefore it meets condition Example 5a The CCCs with alternative CCSsA sif 1 IF expressionl THEN END sif cond_expr2 IF expression2 THEN END statement sequence ENDIF don t meet property by its definition and because they contain program fragments derived from non iterative grammatical sequences Example 6 The CCCs with alternative Sif DEBUG mls key Send Sif not DEBUG mls_key Send meet properties because they contain program fragments derived from itera tive non terminal initial value 43 Example 6a The CCCs with alternative a integer IF DEBUG 1 SEND SIF not DEBUG 2 END do not meet property by its definition and because they contain program fragments derived from non iterative non terminal initial value list 5 3 CCC implementation In the approach proposed the initial M Pascal grammar is not being extended to provide the analysis of the source M Pascal code with arbitrary insertion of CCCs Instead to preserve the layout of the source code conditional compilation directives are treated as comments and attached to definite lexemes of the source program While an ordinary comment does not need any parsing to be performed o
47. r arrays const size t ArraySize const long LowerIndex const const long UpperIndex const const size t StrSize const MString StrScan long StrScan const MSubString amp pattern The given class 1s inherited from the base type Array see Chapter Array which allows applying to the object of the given type all of the array operations element access array size and array comparison Besides the given template class allows us to preserve the semantics of the source program and its external representation which would be impossible with the use of the standard C strings array of char The MString class provides a rich interface to handle operations with strings 31 As it was emphasized at the beginning of the chapter there is an additional M Pascal mechanism of extracting a sub string from a string A sub string of a string object can be considered as a separate object of the string type inhered in the parent string but limited by the specified range It can be clearly demon strated by the example of passing a string to a procedure function by reference as a VAR parameter For such string parameters the size of the string and the address of the string are passed So the string size of the formal parameter is ig nored and the actual length of the string is considered inside the routine To preserve this string type semantics a new class MSubString has been in troduced class MSubString public DynArray lt char gt
48. ring also denotes the type of a constant of a string type since its size is only known at run time when concatenation is performed in M Pascal concatenation is performed at compile time The cast operator DynAr ray lt char gt in the class MString is used to cast a parameter when passing an actual parameter of the STRING N type and accepting a formal parameter of the dynamic array type The StrSize method is properly defined to return a run time length of the string MSubString contains a pointer to the string storage instead of the storage Itself Therefore the size of the object of the type MSubString is equal to four bytes which preserves the data layout in parameter passing For that purposes the base class DynArray has the only member bounds of the pointer type to the following structure template class T struct Dynvar T data In the class MSubString it points to the beginning of the string int lower bound upper bound The bounds of the dynamic array bool preserve data serves to indicate that this dynarray is casted from the static array template class gt class DynArray protected Dynvar lt T gt bounds class DynArray implementation Such an approach fully simulates the M Pascal string type parameter passing and keeps the semantics of working with sub strings in M Pascal Let us consider the example of MSubString usage type string string 32767 procedure procla VAR s
49. ructure To illustrate the process of conversion we consider application of the converter to a small piece of M Pascal code const 1 7 parsing this code the parser builds the following parse tree eorr constant list constant definition Geentition EQUAL Constan Simple N Cxpression EQUAL Constan L expression expression C term D nsigned 3 Constant primary factor primar 1 prima VU facto U y nsigned constant AIA V identifier H 48 Simultaneously the parser generates IR graph from PASFEO and PASFE ob jects The fragment of IR graph corresponding to the parse tree is the following PASFE ConstDeclaration PASFE ConstDeclaration data data PASFEO Const PASFEO Const name a name b type null type null value value PASFE IntConst PASFE Add int const 1 so left_term is_hexadecimal false P d right term PASFE_AccessData PASFE_IntConst it const is hexadecimal false This IR graph is passed to Codegenerator The code generation process this part of the converter is also called Back End calls the It
50. s of a large amount of so called header files and com mon Pascal files Header files contain declarations of global types global vari ables and constants and global procedures and functions Common Pascal files contain the bodies of procedures declared in the header files Besides they may include many other header files by using include directives similar to C ones Since the sets of used header files by different Pascal files intersect the converter needs to regenerate the target C code many times for the most of the header files in the batch mode Therefore several other operation modes have been implemented to reduce the translation time on a large amount of input source files e Common mode the converter generates the C code for each header and Pascal file at the parser codegenerator stage of the converter e Only Pas mode the converter parses all the header files found builds the IR graph in memory for them but does not generate the C file if it already exists Therefore the codegenerator stage is switched off for header files e Quick mode the converter neither generates C file as in the previ ous mode nor rebuilds the IR graph for a header file The converter preserves all the pointers to the roots of trees built in a special table As required the converter searches for the necessary tree and attaches it to the current tree The last mode may be considered as a precompiled headers mode imple
51. template class Set whose pa rameters are lower and upper set bounds and its size is determined by M Pascal data layout rules see below about packed and unpacked sets The set itself is represented by a byte array storage data of the corresponding size with bit ad dressing organized Besides all set operations union intersection difference set comparison and element membership checking are implemented within the Set class define MAX SET SIZE 255 define BITS PER WORDT 8 sizeof char define BITSET WORDS n n 121 n BITS PER WORDT 1 BITS PER WORDT template size t LOW size t HIGH size t SIZE MAX SET SIZE gt class Set private unsigned char storage BITSET WORDS SIZE J public Interface Unpacked sets The size of any unpacked M Pascal set is equal to 32 bytes as it contains 256 elements The class BitSet represents all unpacked sets and is inherited from the base class Set with the third template parameter instantiated by 255 0 is included in the set template size t LOW size t HIGH class BitSet public Set lt LOW HIGH MAX SET SIZE public typedef Set LOW HIGH MAX SET SIZE Base BitSet In the general case only part of memory reserved for the BitSet object 32 bytes is really used The low and high bound checking is carried out dynami cally using the first two template parameters LOW and HIGH Packed sets Unlike the case of unpacked se
52. tion Such a situation occurs for example when the DOWNTO loop variable is of enumeration type and the loop lower bound is LOWER enumeration type The condition loop variable gt 0 will never become false This leads to the following translation scheme in C T last var var init val for last var final val vartt stmt if var last var break 3 6 UNIV formal parameters The UNIV keyword is an utterly specific feature of M Pascal The following is a citation from the language guide The prefix UNIV in a parameter group suppresses compatibility checking of the actual and formal parameters UNIV cannot be used when either the actual parameter or formal parameter but not both is declared as having the STRING type The compiler issues a warning mes 2 sage whenever the size of the actual UNIV parameter is less than the size of the formal UNIV parameter The following example will be considered below VAR a AT PROCEDURE p UNIV a FT BEGIN END It is obvious that an explicit casting C mechanism should be used to repro duce the behavior of the M Pascal compiler since C itself does not provide this kind of suppressing being an extremely type safe language There is a set of casting operators available static cast dynamic cast const cast reinterpret cast and C like type casting The above example could have been translated into AT a void FT p reinterpret
53. ts the memory size necessary to keep all ele ments of a packed set depends on the real set bounds or to be precise on its upper bound HIGH The M Pascal compiler does not use information about the 15 lower bound when determining the size of a set That is in the general case simi lar to unpacked sets memory is used partially The size of an unpacked set varies from 1 to 32 bytes according to its upper bound The class PackedBitSet inher ited from the class Set with the third template parameter equal to the real packed set size preserves data layout for this type template size t LOW size t HIGH class PackedBitSet public Set LOW HIGH HIGH 1 gt public typedef Set LOW HIGH HIGH 1 gt Base BitSet Bound checking is carried out just as in the case of unpacked sets Since the packed and unpacked sets are inherited from the same base class Set they can be mixed in any set operations assignment relations difference union intersection Initialization is handled similarly in both packed and un packed sets Let us consider initialization of unpacked sets for simplicity Set initialization Any M Pascal set can be initialized by a set of values or by an interval range of values Correspondingly two auxiliary initializing classes are intro duced define ENDSET 1 class SetValues public Set 0 MAX SET SIZE MAX SET SIZE public SetValues int first class SetRange public Set 0
54. turally translated into C classes using the UNION construction for variant parts Example 3 3 1 see Examples 1 5 The M Pascal record type T1 1 record a char b longint c char end can be converted into the C dass class Tl unsigned char a int by unsigned char c access methods 17 Example 3 3 2 The type T9 packed record bit offset bit size a boolean 0 d CASE b boolean OF 1 1 TRUE cs 0ft 9T5 2 6 tty 8 7 FALSE e A z 8 7 end sizeof T9 2 bytes can be converted into the C class class T9 bit offset bit size boolean a 1 0 lo 7 boolean b 1 1 L ox union struct unsigned char c 6 8 6 c unsigned char d 7 16 7 unsigned char e 7 8 p E access methods sizeof T9 3 bytes The necessity to preserve data layout and initialization features essentially in fluences the natural translation scheme shown above Data Layout Preservation for records without variant part The size of M Pascal type is equal to 7 bytes the size of T1 in C 12 bytes With the use of GCC compiler extension attribute packed at tached to an enum struct or union type definition it specifies that the minimum required memory will be used to represent the type it is possible to get the following T1 representation class 1 byte offset byte size unsigned char a 0 T int b 1
55. ve readability and maintainability of the target program 36 Such an approach appears to be invalid as soon as multithreading is considered In this case two or more threads can concurrently use the same data objects from the same namespace as a matter of fact destroying the data of competing threads Since it is not known beforehand that only one control thread exists the translation scheme using namespacing can be expanded to the scheme that uses classes instead of namespaces It becomes possible to create the own set of ob jects of such classes for each control thread Simulation of nested function structure by nested data structure Every function is represented by a class containing e nested type declarations e private members corresponding to the function parameters they are ini tialized in a constructor e private members corresponding to the function local data variables and constants the constants can be declared statically with the purpose of optimization e areference to the parent function class only in nested function classes initialized in a constructor e nested class declarations corresponding to the nested functions in the source code and also e a public constructor accepting the list of function arguments and for nested functions only a reference to the parent function class e the single public method e g an overloaded operator containing implementation of the function body For the best r
56. ypes in M Pascal differ from those in C and preservation of copy and comparison op erations of structured types Moreover the following is desirable e encapsulation capability of hiding some aspects of data representa tion For example the M Pascal record fields become private members in C classes and one has access to them through methods In chapter 1 a conception of Data Layout Preservation DLP is presented Chapter 2 describes the general approaches to initialization Chapter 3 is over view of the translation schemes of the most interesting aggregate M Pascal types records without and with variant part arrays and sets In chapter 4 the transla tion schemes of nested procedures in M Pascal are given In chapter 5 the trans lation scheme of conditional compilation directives is described the sets of trans lated and non translated Conditional Compilation Clauses CCC are defined All the examples are concerned to the M Pascal compiler on M68000 and the GCC GNU compiler on SPARC station 1 DATA LAYOUT PRESERVATION One of the main requirements to the converter is data layout preservation DLP This term means the following e preservation of the size of each data item including every component of its decomposition e preservation of data items locations up to bits in memory this in cludes alignment issues and overlapping in variant records Alignment is an extension of the memory allocated for all values o
Download Pdf Manuals
Related Search