Home

User Manual - 作りましょうプロジェクト

Contents

1. Beikaitoru A modern revival of the historic plotter fonts digitalized sic by A V Hershey of the U S Naval Weapons Laboratory during the 1960s and subsequently distributed by the U S Na tional Bureau of Standards and on Usenet This version includes the Japanese character range omitted from most other revivals GPLv3 with font embedding exception FontAnvil A PE script interpreter for editing fonts under program 6l control Overall package GPLv3 some parts also available under BSD style and other licenses Genjimon A set of standalone fonts for the Genjimon symbols which in the main Tsukurimashou fonts occupy the private use code points U F17C1 to U FI7F6 In the standalone fonts they are mapped into the ASCII range for easier access GPLv3 with font embedding exception Icemap A code generator for static maps This allows easier mainte nance of otherwise hardcoded tables of Unicode data built into FontAnvil and IDSgrep GPLv3 IDSgrep A program for searching databases of tree structures stored in extended ideographic description sequence EIDS format This is intended for searching kanji databases according to spa tial structure using a powerful query syntax inspired by regular expressions The need for such queries came up during Tsukuri mashou development but the package may well be useful in other contexts as well Includes code to interface to several different character dictionaries including st
2. kazoemashou 47 The possible values for style will depend on the family Styles generally correspond to sets of preset values for the adjustable parameters of a family s letter drawing code The style values currently allowed per family are as follows Style names marked with an asterisk are not allowed as part of the basic Hamlog name but may be introduced by rewriting as explained below tsukurimashou kaku maru mincho bokukko anbiruteki tenshinokami tsuita soku atama mandeubsida dodum batang sunmoon blackletter lolita cosette kazoemashou seisuu The weight may be extralight light normal demibold bold or extrabold The spacing may be monospace or proportional All combinations of these are allowed for all families except that when the family is kazoemashou the weight and spacing must be nor mal and proportional respectively that constraint is enforced by the allowed_style 4 predicate in select hl The Hamlog name may be subject to rewriting according to rules in the select hl file specifically the do_style_xlat 8 predicate These rules support a few style weight combinations that have been given unique style names of their own both because that s fun and as a legacy of earlier versions in which the weight parameter was not fully implemented This code is also meant to support future per language localization of name components like dem
3. www wenl in com cdl Type Project Adjustable Fonts announced in 2012 Commer cial type foundry offers to provide Japanese language corporate fonts that are adjustable for weight and width It is not clear that they actually have a full set of kanji from the descrip tion on the Web site it appears that they have a ready made set of parameterized kana and then they will create whatever specific kanji the client wants to pay for The parameterization technology is evidently smarter than the purest form of linear interpolation because they have a JavaScript demo highlighting the difference but it still seems to be within the range of what could be accomplished with for instance Adobe Multiple Master http typeproject com projects adjustable TADVyZAX7hH FYyZ Development roadmap This version contains all the kyouiku kanji the ones taught in Japanese elementary schooD through Grade 4 The current plan is to release a minor version after each grade level of kyouiku kanji and one halfway through each grade level which will take us up to version 012 at the end of Grade 6 Version 1 0 will probably be a separate version released shortly after 0 12 with a general clean up and renovation but its possible I might skip directly from 011 to 1 0 There are 1006 kyouiku kanji though the fonts already contain more than that number of kanji glyphs because my general practice is to add other glyphs that are convenient to add wheneve
4. 3 82 or later non GNU versions will not work earlier versions will not work Metapost this comes with most TEX L TEX installations 9 An interpreter of PE scripts FontAnvil is preferred and included with the Tsukurimashou sources FontForge will also work as long as it retains support for native scripting Other things that might also be useful include 52 X3 TEX needed for compiling the documentation 9 Expect makes the build system work better exactly why is explained in more detail in the warning message you will get if you dont have it The KANJIDIC2 database needed for the kanji coverage chart and a planned future fine grained character subset selection fea ture 9 A supported Prolog interpreter not necessary but if one is available then the build system will use it and run more effi ciently 9 A multi CPU computer the build system will by default detect and use all your CPUs since some stages of compilation require a lot of processing it s nice to have several A checksum program preferably sha256sum used for some subtle build system optimizations If you have a current version of TgXLive then you probably have Metapost and X3 TEX already Note that it does have to be fairly up to date Tsukurimashou now bundles its own version of the public domain METATYPEI code so the dependency of previous versions on the mtypel3 distribution of that has been removed from the
5. ENS LES 09 INFASI HH TAVNTPI 1 vY arv Tsukurimashou 09 Parametric Font Family User Manual Matthew Skala mskala ansuz sooke bc ca 201448H17H August 17 2014 This project s English language home page is at http tsukurimashou osdn jp index php en cLOZBZzZhlui B 88O Zh http tsukurimashou osdn jp index php ja 4 User manual for Tsukurimashou Copyright O 2011 2012 2013 2014 2015 Matthew Skala This program is free software you can redistribute it and or modify it under the terms of the GNU General Public License as published by the Free Software Foundation version 3 As a special exception if you create a document which uses this font and embed this font or unaltered portions of this font into the document this font does not by itself cause the resulting document to be covered by the GNU General Public License This exception does not however invalidate any other reasons why the document might be covered by the GNU General Public License If you modify this font you may extend this exception to your version of the font but you are not obligated to do so If you do not wish to do so delete this exception statement from your version This program is distributed in the hope that it will be useful but WITHOUT ANY WARRANTY without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE See the GNU General Public License for more details You should have received a copy of
6. Tsukurimashou scenario Thus make download make When make runs in a completely new checkout it will probably go through a couple of repetitive cycles of re building the Makefiles This repetition should be harmless Icemap is also used by IDSgrep and the IDSgrep configure script unlike FontAnvil s will not speculatively configure the package to use an Icemap binary that does not yet exist That means building IDSgrep in a completely fresh version control checkout on a system where Icemap does not yet exist will not work without additional human intervention IDSgrep will however find and use a built Icemap in a sibling directory even if Icemap has not been installed on the system globally So to build IDSgrep from a version control checkout build Icemap first either by itself or by doing a top level Tsukurimashou build and then run the IDSgrep configure either by itself or by running the top level configure a second time with an appropriate enable parasites setting to allow IDSgrep to build with the freshly compiled Icemap These extra steps are not necessary when building IDSgrep from a distribution tarball because the Icemap generated sources will be in cluded in the distribution automatically U JV OXZ Tool documentation There are a number of scripts included in the tools subdirectory Most of these are intended only to be invoked automatically by the build system but those intended to be invoked by hand are
7. Tsukurimashou s own replace ment for FontForge This is hoped to be of benefit to FontForge development as well because it may make it easier for them to stop supporting PE script should they want to in the future they can en courage PE script users to use FontAnvil instead of leaving such users out in the cold FontForge has very recently made significant improvements to its spline geometry operations remove overlaps find intersections add extrema I have been complaining about those bugs for years they are critical to Tsukurimashou Many thanks are due to Frank Trampe of the FontForge team for his assiduous work on the spline geometry issues The latest changes to these features from FontForge were ported into FontAnvil at the last minute before the Tsukurimashou O9 release along with some quick and dirty last minute hacking on other stability issues infinite loops and segfaults to just get the build working I cannot vouch for Tsukurimashou 09 to build properly with any font compiler except the development FontAnvil bundled with it and I cannot vouch for that version of FontAnvil to do anything except build Tsukurimashou O9 more or less correctly However that specific combination on my specific desktop computer seems to work pretty well I presented Tsukurimashou at TUG 2013 in Tokyo Slides from that talk are at https tug org tug2013 slides skala slides pdf and the as sociated TUGboat article is available to TUG members at
8. also have per glyph non default advance widths Fixed by changing the limit to 22 with a patch I wrote that the FontForge developers accepted in December 2011 FontForge save and then load of an SFD file has the effect of renaming the clig feature to rtla on only one of the three machines where I ve tried it May be related to the x86 64 architecture Not reliably reproducible Reported on mailing list November 2011 mentioned in a developer s commit message December 2011 not observed recently and probably fixed by now FontForge rasterization to BDF via Freelype as opposed to what ever other code FontForge would use sometimes produces cor rupt results which has complicated indirect consequences because the Tsukurimashou infrastructures uses BDF files as input to the auto kerning program As a result when this bug bites the hor izontal metrics go screwy on some fonts notably Mandeubsida 23 Batang PS ends up looking monospaced Reported to FontForge mailing list in November 2011 inspiring Freelype maintainer Werner Lemberg to find and fix an unrelated bug in Freelype it was unable to process BDF files with high code points see http savannah nongnu org bugs 34896 But that bug did not ac tually affect us because FontForge uses its own code for BDFs instead of Freelypes anyway and Lemberg says cant help with the corrupt rasterization Workaround is to compile FontForge without Freelype support from the GUI it is
9. closeness for the row The sum of closeness for all the rows would be equal to the number of rows in the case of two perfectly vertical lines 230 units apart The kerner program adjusts the margin to margin distance so that the sum of closeness is equal to that It uses a binary search to do that adjustment which is probably not optimal for a fixed exponent there should be an exact analytic solution possible without iterating but has the big advantage of not requiring a redesign should the exponent or even the entire closeness function change The closeness computation is shown schematically in Figure 7 Note that this closeness is calculated on the contours as defined previously rather than the actual shapes of the glyphs it is also done on the centroids of the kerning groups thus generic contours each represent ing many glyphs rather than the contours of any specific individual glyphs The effect of the exponent 3 in the calculation is to give much more weight to points that are close together as suggested by the shading in the figure If were kerning a pair like lt we want to pay more attention to the point of the less than than to the distant ends An 83 exponent of 3 means that points at half the distance count eight times as much toward overall closeness so theres a strong bias toward seeing the points of closest approach If we imagined using a larger exponent this bias would be even stronger in the limit with an
10. database of kanji primarily from a Japanese point of view broken down into strokes and radicals with some curve points and a lot of abstract information about how the strokes correspond to the traditional radical classifica tion so that you can automatically recognize for instance what J and X have in common stroke order and so on This is a valuable resource for dictionaries and handwriting recognition systems Some kind of supervised semi automatic processing could probably turn it into a font but keeping the style consis tent because the database has multiple sources and adding the serifs and other visual information needed for styles like Mincho would require some significant work http kanjivg tagaini net See it in action in Ben Bullocks handwritten kanji recognizer at http kanji sljfaq org draw html Character Description Language from Wenlin Institute Inc Cur rent in 2011 This is a commercial product It is apparently though the Web page could be clearer on the exact nature of what they re selling a database of character descriptions similar to KanjiVG though with wider coverage plus a binary only ren 16 dering library the combination available for license at some un specified price It says it s capable of generating MetaPost as one of its several output formats Not clear to what extent there is parameterization but presumably that would be in the converter rather than in the resulting MetaPost http
11. described below 65 e autodep automatically maintain the header inclusion lines in METAFONT files so that each file will include the files that define macros it uses and only those Files intended for use with autodep should include a comment line consisting of the case sensitive text 4 AUTODEPS and any automatically maintained input lines following that terminating with a blank line When the tool is invoked with the command line tools autodep mp mp mp from the root of the source tree it will scan all the METAFONT language sources find definitions of macros and where necessary rewrite the blocks of includes tagged with the special comment line in order to ensure that every file includes the definitions of all needed macros There are a number of exceptions and special cases included to make it do the right thing with macros defined in multiple places as a result of style overrides and so on build plot generate a plot using GNUplot which is required of the current progress of a full build This was hacked in at the last minute to fill a specific need during the rather hurried packaging of version 09 and it may or may not be useful in any other contexts Run it from the root build directory while a full build is in progress It will create a file named buildplot png which shows the current state of build progress Computing the overall progress status of a Tsukurimashou build is complicated because even knowing which
12. have any code point mappings at all in the finished fonts On the other hand removing their code point mappings means it is no longer possible to access the glyphs directly by code point then short of using third party software to bypass the mapping tables and access glyphs by raw glyph numbers users must use the glyph substitution features instead which means both that the glyphs are only available in the contexts where the substitution features are designed to provide them and that the users must have Openlype rendering software that at least sort of works which is not necessarily common in the field The xpua feature allows a configuration choice between these two situations With it turned on which is the default most of Tsukuri 4 mashou s PUA glyphs will have their code points stripped in a late stage of font packaging It is not universal some glyphs that do not correspond directly to any Unicode character remain accessible at PUA code points because that s relatively harmless and they may be difficult to invoke in other ways But glyphs that exist only to support OpenType features such as the ones needed for vertical fractions and Korean syllable layout end up as unencoded glyphs in the final fonts As of version 0 7 the Tsukurimashou PUA encoded characters that keep their encodings in the final fonts notwithstanding xpua are exactly the glyphs in the U FT xx range Turning off xpua causes every glyph to
13. it has no meaning such as lead jamo shaping applied to English text will simply have no effect Specific languages are nonetheless 32 mentioned in the font tables for the benefit of FontForge which doesnt seem to handle defaults very well and any other software that behaves like FontForge Some of the features mentioned in this section are not actually Open Type features in the technical sense they are instead options that con figure the Tsukurimashou build system to control things that should or shouldn t appear in the OpenType font file output They are given customized four letter tags and configured through the same interface as true OpenType features because they are somewhat related and it s more convenient than having to have some other interface just for them ADDTT Alternate Fractions afrc Tsukurimashou does not contain any special support for diagonal frac tions but it does support vertical or nut fractions using the Open Type feature afrc With the afrc feature turned on any sequence of up to four digits a slash and up to four more digits becomes a vertical fraction 1 2 gt z 34 56 gt amp 189 123 gt 4567 8901 ss This feature works with both the narrow digits and slash ASCII code 47 57 Unicode U O02F U 0039 and the wide ones Unicode U FFOF U FF19 If the input digits are narrow and the fraction is one digit over one digit then the resulting glyph will be narrow the same
14. list In preparing METATYPE 1 code for bundling I discovered a previously unknown dependency of previous versions on the tlasm package from tlutils that has now been bundled as well Tsukurimashou compilation depends heavily on scripts written in the FontForge native scripting language which I call PE script for neu trality Although as of July 2014 mainline FontForge continues to be able to interpret this language the FontForge development team treats PE script as a legacy feature and encourages users to switch to Python instead It is highly unlikely that the Tsukurimashou project will ever switch to Python To ensure continued support for Tsukurimashou I now maintain a standalone PE script interpreter called FontAnvil this should be available wherever you got the Tsukurimashou source code 53 and is the preferred PE script interpreter for building Tsukurimashou As of version 09 both FontAnvil and FontForge will work more or less equally well Continued FontForge compatibility in the future is not guaranteed EJLROZATA Build system The build system is based on GNU Autotools and should be reasonably familiar to anyone who has compiled popular free software before Run configure help for a description of available configuration options run configure possibly with other options as described in the help message to set up the system the way you want it then once it is configured to your liking run make to bu
15. long running programs At some time before version O 641 the solver mistakes were fixed by the limited serial number scheme but serial num bers would run out and cause a fatal error when they reached 225 In version 1 501 the limit was increased to 23 Any suffi ciently long running Metapost program will eventually die as the limit is exceeded Some experimental versions of the Blackletter Lolita curve fitter would exceed the 2 limit the current exper imental version is less computation intensive probably wouldnt exceed qee and certainly fits comfortably in gal but it seems like the limit should not exist at all Debugging is hindered by some currently in progress redesign work on Metapost s data structures such that as of October 2011 the cutting edge de velopment version leaks memory fast and crashes for that reason long before the serial numbers can run out anyway Memory leaks acknowledged and planned to be fixed by Metapost main tainer Taco Hoekwater but it may take a while I ve posted a link target to track this issue at http ansuz sooke bc ca entry 213 METATYPET sometimes runs glyph names through the METAFONT 21 tokenizer At this point I dont know how essential that is to the operation of METATYPE1 or whether it can be changed It has subtle effects that can cause problems One issue shows up in glyph names that contain a decimal digit followed by a dot as in unillOO bug then what gets written into the Posts
16. look good without ligatures even in proportional fonts The Mandeubsida fonts however contain an extensive ligature table for hangul and use of this feature is required to get full support for hangul script Without it precomposed syllable glyphs will not be used and every syllable will be rendered in one of the six generic layouts See the section on Korean language support 35 FINA Mark to Base Positioning mark This feature only exists in the proportionally spaced fonts a similar effect is achieved in monospace fonts via the calt feature and zero width glyphs It should be turned on by default The feature stores data for Openlype renderers to attach accents to letters allowing for combinations of letter and accent that do not have precomposed glyphs of their own For instance U 0071 U 0306 will generate d a lowercase q with a breve which does not correspond to any Unicode code point and would not otherwise be available As of version 0 6 support for this feature is limited Not every base letter you might want will necessarily be available all the most popular combining diacritical marks exist but it s easy to imagine others you might want that are not included in many cases especially when capital letters are involved the mark to base version of a combined character ends up looking significantly different from the precomposed version some of the combinations just dont look very good and intended extensions of the
17. of character decompositions make fea perform text substitutions on feature source code fsc files to generate Adobe format feature fea files expressing code like font metadata make flog collect a bunch of debugging information and generate a font log file see add flog above make gpos compute information for GPOS features mark and mkmk by scanning the proof files to determine anchor loca 68 tions and compatibility and some generated FontAnvil scripts to determine bearing adjustments made during kerning make hglpages generate METAFONT source code for the hangul precomposed syllable subfonts which are all identical except for their starting indices and generated from a template make kchart generate TEX code for the kanji coverage chart 9 make kddata extract KANJIDIC2 data in Hamlog form make name generate a fonts NAME metadata table 9 make proof generate TEX code for proofs and pretty printed source code Ss mp2pf stripped down translation from Awk to Perl of a similar program in METATYPET this invokes Metapost in the appro priate way to generate something that can be processed into a Postscript font and then does process the result into a Postscript font Note that hinting support has been removed from this code and the resulting Postscript fonts may not really be good or en tirely format valid they are only intended as a way of getting the outline data into FontAnvil
18. scheme to for instance allow adding dakuten to Japanese kana dont exist yet at all But many common cases should be covered RINDA Mark to Mark Positioning mkmk This feature works in combination with the mark feature above and it only makes sense to use when mark is turned on like mark it is only available in proportional fonts It stores data used by OpenType renderers to add more marks to glyphs that already have some marks in particular this should allow stacking up more than one accent on the same letter to create many glyphs that would be impossible by other means In version 0 6 this feature has been expanded to cover a few more of the Latin accents than before allowing the construction of such characters as X It is still somewhat experimental however AST 5 Metadata Table name This is not the usual kind of OpenType feature but it can be turned on and off through the same build system interface By default the font 36 will contain a name table listing metadata such as creatorship and licensing in both English and as appropriate one of either Japanese or Korean Some of this information is obligatory so if you turn off the name feature you actually still get a name table but it will be less completely populated EMER Ornaments ornm Tsukurimashou provides eight tomoe ornaments that look like this e SS 5099089 F These glyphs are encoded to the Unicode private use code po
19. show movement of 43 lines 1 through 6 of hexagrams or 1 through 3 of trigrams encoded in that order The idea is that you can typeset something like this using a sequence of codes like this U 4DDE U F1702 U F1 04 U 2192 U 4DEF BTE OX Ideographic Description Characters Unicode defines a set of special code points U 2FFO to U 2FFB and a syntax for using them to describe what Han script characters look like For instance Bas might be described by the string cub HILUX which expresses that consists of E written above a composite sub glyph that itself consists of L wrapped around K Unicodes scheme for character description appears to have been inherited more or less unmodified from a similar scheme in the Chinese GBK standard Tsukurimashou includes glyphs for the special characters There is also support in the build system for a make eids target which generates a file called tsukurimashou eids in the txt directory That file is a collection of Extended Ideographic Description Sequences EIDSes describing the construction of the kanji in the font This is the inter face to IDSgrep a program for searching kanji according to powerful criteria on their visual structure IDSgrep is available from the same OSDN Japan project that hosts Tsukurimashou see that package and its documentation for more information on EIDS format and how to use it With or without IDSgrep Tsukurimashou may be used to typeset s for the de
20. the GNU General Public License along with this program If not see http www gnu org licenses 2 E R Contents 4 RO Introduction OMA Whats new in QUT us or Red Pec EA dC AoZ7mD7z2h Other similar projects 6 66 v oon T ADyZ hH FYs7 Development roadmap RHOJIRDIXT Relevant bugs in other software METAFONTAZEC METAFONT ete o o o RR averse PontFofg amp ecosistema AIPA ASIEN oso hok eR ORES RR x GG X ERED owed E X SE Ra MOOV 2 Peer Other software ewe Geen ewe CASS EYELID Using Tsukurimashou OpenlypeM74 F Openlype Features 4405345 Alternate Fractions afr yT PREILI IZT Capitals to Small Caps c2s6 lt ss co XAK Contextual Alternates calO 7J70ft amp t4f amp Glyph DeComposition ccmp AF Ligatures liga 2e XFU GA Mark to Base Positioning mark ALUNGA Mark to Mark Positioning mkmk AYF H Metadata Table name 2 o Eont Ornaments ornm ceo ee ee go yos FontAnvil7203 00344 77 5 FontAnvil specific Metadata pfed AC HhreE Small Caps mep se sea 8 ee o o en E 25 V72A427h Heavy Metal Umlaut ssOD FO SF Enclosed Letters and Numerals ssO2 ALFATTS tiie amp ie tT Limit PUA Code Points xpua MANIC Extra glyph coverage TRA Genel s ea eae ane saioei i caia ea Peet Ce au ee hake OS ARA fk OXF Ideographic Description Characters ma
21. the Japanese language typeface enough work 20 to 25 man years that in practice it s rare for such a project to be completed by just one person working alone at all There is also the minor detail that I said a typeface family one of those typically consists of five or six individual faces so the time estimate increases to between 100 and 150 years However I ve already decided to spend the time on the language learning project And I m sure that if I design a glyph for a character I m going to remember that character a lot better than if I just see it a few times on a flash card And as a highly proficient user of computer automation I have access to a number of efficiency increasing techniques that most font designers dont So the deal is I think if I study the language and also design a typeface family for it I might be able to finish both big projects with less or not much more effort than just completing the one big project of learning the language alone And then I d also end up with a custom made Japanese language typeface family which would be a neat thing to have And so here I am building one Because this is a parametrized design built with Metafont users can generate a potentially unlimited number of typeface designs from the source code but I ve provided parameters for several ready made faces each of which can be drawn in monospace or proportional forms The mainline Tsukurimashou fonts are intended for Japa
22. this will be beneficial to both Tsukurimashou and FontForge by providing a stable platform for Tsukurimashou s scripts and also freeing the FontForge team from the ongoing need to support features that do not fit into their marketing strategy FontForge spline geometry operations such as overlap removal and simplification these have historically tended to be very numerically unstable and subject to some combination of infinite loops segmentation faults unexplained floating point exceptions bizarre error messages incorrect output and so on The most recent development versions seem to be relatively good best advice is to switch to one of those turn debugging on and hope It is also sometimes possible to get past individual problems by switching between the different floating point formats offered by FontForges compile time configuration float double and long double But which of those is the best choice isnt always predictable and as of October 2012 some FontForge developers are attempting to remove the option to choose between them which should improve stability in the long term by improving maintainability but may harm stability in the short term Some numerical tweaks in actual glyph outlines also exist to try to work around these issues FontForge usually fails to automatically insert appropriate sub table breaks when it reads a feature file describing a large Open Type table Workaround our Perl code inserts explicit brea
23. tug org lUGboat tb05 2 tb10hobby pdf John Hobby has posted the source code described in the paper on his Web site at http ect bell labs com who hobby hobbygu tar gz but as he says in the enclosing Web page the files are written in the now obsolete METAFONT79 language and they are of limited use because they are not compatible with todays METAFONT pIEX ASCII Corporation 1987 onward not a font project but a TEX engine modified to handle 16 bit character codes and using existing fonts from other systems Very popular in Japan to some extent it still is though other projects of similar nature mostly not listed here have gained a lot of market share in recent years The Quixote Oriental Fonts Project spearheaded by Dan Hosek announced in a paper at the TUG 1989 Conference Intended to be a parameterized METAFONT native family for Chinese Japanese and Korean Hosek apparently had some source code in hardcopy form displayed at the conference but I ve not been able to find the code nor any subsequent discussion of the project http www tug org lUGboat tb10 4 tb26hosek pdf Poor Man s Chinese and Poor Man s Japanese 1990 Tom Ridgeway technology for displaying 24x24 bitmap fonts through METAFONT This was not curve tracing nor smooth scaling but a way to 14 actually display the dot matrix jaggies and all Still available in CTAN package poorman but considered obsolete GS JemIEX 1991 Fran ois Jalbert in
24. un affected by this option and trying to disable it would be a bad idea Autotools encourages the use of a separate build directory with the sources remaining inviolate elsewhere but that is not really recom mended for Tsukurimashou I try to make it pass make dist check right before each release which implies making separate build directo ries work but if you are building Tsukurimashou from a checked out SVN version it may or may not work It s safer to build right in the main directory Even if VPATH builds work they are only intended for the case of having an untouched set of sources in one directory and a build in another If you try to do the overlay thing with modified versions of some source files in your build directory it is unlikely to work because of the large amount of bolted on filename and path logic that doesn t go through GNU Make for name resolution If you look in the source of the build system specifically files like configure ac you ll see that I did a whole lot of work ripping out sections of Autotools that were designed for installations of executable software GNU standards require the definition of a ridiculous number of different installation directories almost none of which are applicable to a package of this type and I took out most of the support for those to reduce the cognitive load for users who would otherwise have to think about their inapplicability This package doesn t install any executables l
25. using the enable parasites option to the top level configure script but each one in its own subdirectory is also a potentially standalone Autotools package with its own configure script and capable of building a standalone distribution of itself with the make dist target Parasites will gen erally look for a Tsukrimashou Project build in the parent directory and connect to it if one is found The current parasites are Beikaitoru FontAnvil Genjimon Icemap IDSgrep Kleknev and OCR Each of these has its own detailed docu mentation which see the notes here are just a summary to give some idea of what the packages are all about Most of these existed in some form prior to the introduction of the parasite build mechanism and in some cases vestiges still exist of their old less systematic ways of connecting to Tsukurimashou In the future the connections will be more streamlined and normalized Note that parasite packages generally have their own version numbers which are not necessarily synchronized to specific Tsukurimashou re leases in some cases a sukurimashou release will contain a pre release or even a broken version of a parasite But it should usually be rea sonably safe to mix and match parasites and Tsukurimashou versions of reasonably similar vintage Parasite packages may also have their own license terms but must always have license terms that allow them to be distributed within the GPL3 top level project distribution
26. work and it will also probably encourage some people to use the fonts who wouldn t otherwise One of the bizarre aspects of human behaviour is that some people will buy a product they would not accept for free Okay whatever in such a case I m happy to take the money for it Having a pay option will also give anybody who wants to support my efforts an easy way to do that For now though I am distributing Tsukurimashou only as this source package with precompiled versions included for the Kaku and Mincho styles The Mandeubsida fonts which dont need the in progress kanji glyphs may also be available in precompiled form as a separate package If you want other styles you ll have to compile them yourself or get them from someone who has done so This limitation is deliberate with the fonts in their current partial form I d rather limit their circulation to hobbyists I maintain several other free font projects which as of version 0 7 have largely been merged into the Tsukurimashou packaging and build system because they share code These parasite packages appear as subdirectories of the Tsukurimashou source distribution and may also be available as separate packages See the discussion in Subsection for more information about parasite packages This documentation file gives some notes on the build system and on how to use the Openlype features built into the fonts Other documentation files included in the package demonstrate what the f
27. A Whats new in O9 Version 09 has been a long time coming There are not all that many new kanji The current count is 1754 kanji up from 1502 in version 0 8 roughly a year ago This includes all the elementary school kanji through Grade Four and 100 of the 185 kanji in Grade Five A lot of infrastructure work has also been done This version was prepared in a hurry for reasons to be described below This version of Tsukurimashou introduces FontAnvil a standalone in terpreter for FontForges native scripting language PE script The decision to bundle a more or less complete font editor drastically increasing the amount of code in the project and thus increasing the maintenance costs was not made lightly The issue is that within the last two years or so I have been involved in FontForge development and it is clear that the strategic direction of that project will present problems for the long term use of FontForge within Tsukurimashou FontForges development team is primarily interested in end user GUI features even going so far as social networking and encourages Python for all new code they have even seriously discussed though not yet committed to removing PE script support entirely which 10 would be fatal to Tsukurimashou if Tsukurimashou continued to rely on FontForge Especially given that Tsukurimashou already bundles its own versions of several other critical tools such as Metatypel and tlasm it seems reasonable to bundle
28. BuDDy binary decision diagrams library 9 lex preferably flex Q9 several files of Unicode character database information and Q9 if using the built in support for automatically downloading the Unicode database files then wget and unzip Check out the project by anonymous Subversion from OSDN Japan svn checkout https svn osdn jp svnroot tsukurimashou trunk tsukurimashou OSDN is the preferred repository but you may instead obtain it from Github git clone https github com mskala Tsukurimashou git Either of these commands downloads the sources from the reposi tory and creates a subdirectory called tsukurimashou in the current directory Use autoreconf to build the build system itself 63 cd tsukurimashou autoreconf 1 The i option tells autoreconf to create missing files This command is likely to produce many error and warning messages but should work nonetheless if the Autotools versions are suitable In particular note that when autoreconf runs it may generate repeated underquoted definition warnings in relation to files in usr share aclocal or similar This is the fault of some other package on your system it is not related to Tsukurimashou and should be harmless to the Tsukurimashou build The issue is that many packages IMLIB is one common culprit install M4 files globally to provide customized Autoconf tests and then if those files are buggy autoreconf complains whenever it is
29. ETATYPE1 bundled with Tsukruimashou may be useful for other fonts that build from METAFONT language source files to produce modern vector file formats It would be nice to be able to manage these kinds of associated packages in a way that maintains their link to Tsukurimashou so that for instance updates in the shared code will propagate to all packages that use it but nonetheless allows the other packages to be distributed as separate entities to users who don t want a copy of the entire Tsukurimashou system There are relevant side issues too such as a hope for attention directed at one package to attract interest to other packages One possible approach might be to attempt to factor out code that will be shared into some kind of Tsukurimashou library package that could then be distributed separately and made a dependency of all the other packages that use it A problem with doing that is that which code is or isn t shared will vary a lot due to the wide range of different packages that might want to share some part of the system Instead the approach taken as of Tsukurimashou 0 7 is to introduce what I call parasite packages These are sub packages for Autotools 60 purposes Each parasite appears in its own subdirectory of the full Tsukuri mashou Project distribution tree or SVN checkout By default they are distributed in distribution packages of the full project but not built by a top level build They can be activated
30. I O so how can Hamlog programs communicate with the outside world The interpreter after checking for the version and debug options which do fairly obvious things interprets its first two command line arguments as a template and a query The template ought to be a valid Prolog compound term for compatibility with other interpreters but Hamlog actually treats it as a string Then it backtracks through all solutions to the query attempting to instantiate all variables and writes newline separated to standard out all the distinct values assumed by the template This is basically the same operation as Prolog findall 3 followed by sort 2 which is how the Prolog shell for Hamlog implements it Any remaining command line arguments and standard input if there are none of those will be read in the usual lt gt way to fill the Hamlog program database Hamlog code is conventionally stored in hl files In the build system the string of comma separated tags for things like characters to be selected gets converted by the make cfghl Perl script into a few clauses of Hamlog and written into the file con fig hl Also written there is a list of page_exists 1 facts naming the 256 code point blocks for source files that exist in the mp directory Then elsewhere in the build system it invokes Hamlog with appropri ate queries against config hl and select hl to get lists of characters 77 OpenType features and other thing
31. The main interesting content is the command line that was given to configure and a dump of most of the variables known to Make There s also a 37 copy of the copyright notice and URLs for the project home pages giving in more verbose detail some of the same information included in the name table I recommend activating this feature especially if you will be building fonts for distribution to others It may make bug reporting easier because it means that anyone who gets a copy of the font also gets some information on where that font came from if someone builds a font and then has trouble with it it ll be easier to help if the font contains this debugging information It also improves the chances that should a font file get separated from its context someone stumbling upon it will be able to figure out what to do with it The Net is full of inaccurately labelled fonts with unknown authorship often being sold by shady commercial enterprises that have no legal right to do so and we all ought to do what we can to stamp that out Detailed human readable metadata in general is a Good Thing But the font log contains information like software version numbers user account names and installation directory names Some people have funny ideas about the sensitivity of such information in relation to system security they may think that revealing it creates a real risk not otherwise present of people breaking into their computers They
32. be encoded at its internal code point That was the state of affairs in versions of Tsukurimashou prior to 0 7 which didn t offer a choice 42 MX Extra glyph coverage FACEA Genjimon Glyphs for the 54 Genjimon are encoded to private use code points U FT7C1 to U FT7F6 in the order of the corresponding Tale of Genji chapters Kiritsubo to Yume no Ukihashi The style of these glyphs varies a fair bit between the different font styles Here are samples S Kaku MINM mill Maru Mill MM mil Anbiruteki Il MM ml Tenshi no Kami MINM mil Bokukko HM il Mincho A set of Genjimon only fonts derived from the Tsukurimashou code exists as a spin off project in the genjimon subdirectory of the Tsukuri mashou distribution The documents there contain some more infor mation about the history of these symbols and what they signify ASSEN I Ching All the I Ching related characters defined by Unicode in the Basic Multilingual Plane are supported trigrams U 2630 to U 2637 mono grams U 268A and U 268B digrams U 268C to U 268F and hexagrams U 4DCO to U 4DFF These characters are sized to fit in the same box as kanji and so in most cases you will probably want to scale them to a significantly larger point size than that of nearby text The style does change somewhat from one font to the next There are also some I Ching dot combining characters in the private use code points U F1701 to U F1709 intended to
33. ble to write the full Japanese language as most reasonably competent native readers know it Many more kanji exist I dont know how far I l want to take this project toward covering them all For reference looking at some other fonts I have handy IPA Mincho contains 6682 kanji probably aiming to cover the JIS 208 standard and Sazanami Gothic contains 12202 probably aiming to cover JIS 212 Those might be reasonable milestones for 4 0 and 5 0 Figure 1 is a chart of the progress to date Note the horizontal axis is labelled by version but scaled by time Even numbered versions tend to take more time because I tend to do the easier characters in each grade level first All the glyph counts in these charts are for the Tsukurimashou Japanese fonts alone The Mandeubsida Korean fonts contain the 11172 glyph block of precomposed syllables which because they are algorithmically generated cannot be well compared to the more manually created kanji and other glyphs The Mandeubsida fonts also contain a few hundred non precomposed glyphs beyond the core they share with Tsukurimashou Figure 2 gives a different view of development progress the number of lines of code total lines in mp mp including comments and blanks 18 3000 total e did ees 2500 A gt E C 20007 9 c PR IR a Pod T PUE oci 2 1500 RI kanji _ VR oo PO 1000 yee AR soo e O gE gt 0 1 0 20 3 04 0 5 0 6 0 7 0 8 09 ve
34. calt features substitutions if those are enabled Similarly this feature will substitute precomposed voiced kana glyphs for combinations of unvoiced kana plus combining dakuten or han dakuten In the Mandeubsida fonts this feature does extensive processing of precomposed syllables conjoining jamo and jamo clusters to render each syllable into one glyph for the lead even if it contains more than one jamo one glyph for the vowel and zero or one glyphs for the tail with the lead and vowel glyphs selected from a choice of alternate forms to take into account the layout of the syllable which is driven by the vowel and the presence or absence of a tail In the case of syllables that do not have standard whole syllable code points i e those containing archaic or uncommon jamo or jamo combinations the output of ccmp is the best we can do Main line standard syllables which have their own code points and corresponding precomposed glyphs will get those glyphs substituted in when the liga feature runs af Ligatures liga The Tsukurimashou fonts do not contain ligatures for the Latin script at present Their basic letter forms were designed with monospace setting in mind where ligatures don t make sense and given the importance of keeping everything legible to readers whose native language is Japanese and who may not be familiar with some of the advanced aspects of the Latin script it was a design decision to make the letters
35. ch glyphs For instance in Mandeubsida there is a glyph named unill24 for code point U I12A but there is also a glyph named unill2A bug which looks identical but is referenced in the substitution table in order to work around a bug in the ICU library used by X3T X The unill2A bug glyph as its name implies is used to typeset the character U 112A but it is referred to internally using the code point U FF 2A That is a code point in one of the Unicode Private Use Areas PUAs which are special ranges of code points reserved by Unicode for purposes like this The U FF72A code point is the one that could for instance be used to enable or disable the unill2A bug glyph in the enable chars option to the configure script The Tsukurimashou package uses hundreds of these PUA code points They are necessary for keeping track of things during the compilation and assembly of the fonts However most of the glyphs with PUA code points are not intended to be directly accessed through their code points only indirectly through glyph substitution Having an unnecessary PUA mapping in the font for any given glyph may cause some buggy software to ignore the glyph s name resulting for instance in incorrect text when cutting and pasting from a PDF file Encoding glyphs to PUA code points could also possibly conflict with other uses of the same PUA code points by other software So all in all it is usually preferable that these glyphs should not
36. cluded a program called jis2mf which would auto trace 24x24 bitmaps to produce non parameterized METAFONT code Many sources from the 1990s for instance a regular Usenet posting aiming to list all then available mf format fonts describe the availability of Metafont for 61 Japanese fonts which is the output from this program The CJK package described in TUGboat at least as early as 1997 still available though no longer popular Werner Lemberg Not a font project but a system for typesetting CJK text in TEX under the standard 8 bit engines getting around the encoding issues by splitting each font into many smaller virtual fonts similar to Tsukurimashou s page system Fonts for this at least at the outset were usually bitmap fonts imported from other systems one popular one was 48x48 later as free vector fonts became available those started to be used some of them via auto conversion from formats like TrueType to non parameterized METAFONT HanGlyph 1997 and 2003 a language for describing Chinese characters and support for rendering them in MetaPost and TEX This is intended to address the Chinese equivalent of the gaiji problem how to typeset rare characters that are not in cluded in standard fonts or encodings The user can describe the missing character and a small font containing just that character will be automatically created and used In principle HanGlyph s technology could be used to create a full co
37. continue to work around the issue FontForge Metrics window crashes when OpenType feature choice is changed in the presence of a complicated multi lookup 26 GSUB feature Reported as FontForge Github issue number 1138 https github com fontforge fontforge issues 1138 January 2014 9 FontForge crashes when generating BDF if the copyright string is a null pointer because of using instead of when at tempting to test for that case Pull request to fix the typo created and merged https github com fontforge fontforge pul l 1143 January 2014 X3TEX XJlEX 9 X3IEX fails to advance the glyph pointer after a successful match in a GSUB table which has complicated consequences for chaining substitutions most notably that ignore sub rules have no ef fect This is actually a bug in the third party ICU library which X3IEX uses Reported to X3lEX mailing list in November 2011 af ter discussion there Chttp tug org pipermail xetex 2011 November 022298 html I found an existing issue in ICU s bug tracker dat ing from June 2010 Chttp bugs icu project org trac ticket 53 Poor workaround is to carefully write all substitution features to work regardless of whether the pointer advances one useful technique is to make what would be an ignore sub rule instead substitute to a series of identical looking glyphs that are never matched on input to subsequent rules As of mid 2012 there was interest expressed by X
38. cript output is unillOObug because that is equivalent but more canonical in METAFONT syntax A more serious issue shows up with the glyph name unil100 11 from Tsukurimashou 0 5 in the new METATYPE1 version 0 55 that gets tokenized in a context where the token I is a spark and so the whole compilation fails The workaround for that was to change Il to lj for lead jamo A further workaround introduced in Tsukurimashou 0 6 was to modify the bundled METATYPE1 code to save the glyph name as a string and put that in the Postscript file instead of the tokenized version It still also parses the names as tokens however so there is also code to insert a bunch of underscores in the internally used name before doing that to reduce the chance of name collisions from this parsing METATYPE1 pen_stroke_edge macro as of METATYPE version 044 for reasons unknown to me if left to its own devices it will sometimes attempt to evaluate the turning angle of a zero vector and then blow up This seems to happen most often when stroking a vector in a direction of approximately 290 degrees As of Tsukurimashou 0 6 we are bundling a macro derived from METATYPE1 version 0 55 which seems to have fixed this bug the workaround in previous versions of Tsukurimashou has been removed Not really a bug because they warn about it in the documen tation and it is a reasonable consequence of design decisions made for good reason
39. d Latin l characters The division into pages provides some advantages for the build system when something is changed on one page it may be possible to avoid re compiling the others and if there are many pages to be compiled then they can be compiled simultaneously on a multi core computer So we would probably choose to keep this design feature even if some future version of the system were no longer bound by the Postscript 8 bit limit that originally made it necessary The most authoritative form of the name of a font is the Hamlog name which is an ordered quadruple of Hamlog atoms The Hamlog language is described elsewhere in this document Hamlog names are manipulated by for instance the code in select hl The items in the quadruple are called the family the style the weight and the spacing An example Hamlog name might be tsukurimashou bokukko demibold monospace The family represents the broad category of the font Generally all fonts sharing the same family value will have basically the same glyph set and the same overall shapes for all glyphs though the visual style of the strokes will vary If there are significant differences in shapes or glyph sets they will be put in a separate family At present the supported values for the family are tsukurimashou tsuita and mandeubsida though there is also some experimental less than alpha status code for families to be called blackletter lolita and
40. defined to be the larger of 100 glyphs and three times the total number of glyphs divided by the total number of classes The purpose of this rule is to counteract a tendency seen in some experiments for the classifier to create a few huge classes for instance a single class containing a large fraction of the 11172 precomposed Korean syllables that cause font subtable size problems It is not clear just where the limits are on how big a class may be but this limit appears to work at the moment 8 Glyphs are examined in this way until no more such moves are possible The idea is that at the end of it the glyphs will all be in classes that are as tightly clustered as possible It s not guaranteed to be a global optimum in other words it s possible that some other assignment of glyphs to classes might be better really optimizing this problem is difficult but it s guaranteed to be a local optimum in the sense that it cant be improved by changing the assignment of just one glyph and its expected to be pretty good overall Note that the initial assignment was deterministic where random would be more usual for this kind of algorithm because it seems undesirable for the kerning distances overall to be non deterministic my copy of the font shouldn t have different metrics from yours if they were compiled from the same sources with the same options After classification weve got 200 classes of left contours and 200 classes of right contour
41. determine accent anchor locations But since those must be adjusted when the bearings are set it also needs before bearing information extracted by getbearings and after obtained by reading the bearings script from kerner The feature files for kerning mark composition GPOS and family features are all combined and applied by the kern font script which runs the bearings script as a subroutine finally producing a complete OTF font 72 I think users will object to that the existing dependencies of this package are already hard enough to sell I also didn t want to bundle a Prolog interpreter even though good ones are available on permissive licensing terms because of the file size and build system complexity consequences of bringing a bunch of compiled language software into the Tsukurimashou distribution The solution Tsukurimashos build system will look for Prolog and use it if possible But the package also ships with something called Hamlog which is a toy Prolog like logic programming system written in Perl A ham is like a pro but less so If the build system cant find a Prolog interpreter it will use Hamlog instead Hamlog is slow and internally horrifying but it works in this particular application It is not particularly recommended for any other applications The configure script looks for SWI Prolog ECL PS CLP and GNU Prolog If one of these can be safely detected it will be used ECL PS CLP has the iss
42. e left and right sides independent of the width of the glyph All the left contour vectors and independently all the right contour vectors are subjected to a modified k means classification Initially the contours are put in 200 classes according to the values of a simple hash function applied to the glyph names This is a change from earlier releases in which a round robin was used the advantage of the hash function is that although it remains deterministic it helps break up a phenomenon that tended to happen with the Genjimon glyphs where because of excessive symmetry in the initial arrangement the classifier could never put the similar glyphs together 80 Figure 6 Glyph contour computation Then for each contour the program asks the question How far is this contour from the centroid of its class and if I moved it to a different class how far after accounting for the fact that the centroid changes when I add the glyph would it be from the centroid of the new class If moving the glyph to some other class would make it closer to the new centroid then the glyph gets moved to the other class where its distance to the centroid will be minimized Note that a glyph in a class by itself will never want to move out of that class because its distance to the centroid is already zero There is an extra rule that the classification will never move a glyph into a class in such a way as to make a class larger than the class limit
43. e parts This is a sufficiently arcane scenario that the build system will not check for it 57 The make dist target defaults to building a ZIP file only instead of GNU s recommended tar gzip This decision was made in order to be friendlier to Windows users who tend to have seizures when confronted with other archive formats and TEX users who are accustomed to ZIP as well However the makefiles should also support many other formats via make dist gzip and similar I am not confident that make dist will really include everything it should if you have disabled optional features with the above options to configure Please make me one with everything as the Bud dhist master said at the hamburger stand before trying to build a distribution The make clean target and its variations should make things pretty clean now that were attempting to make the system pass make distcheck before each release Use them with caution however There is a make check target which will run FontAnvils fontlint program on all the OTF files This is a very stringent test and its quite likely that fonts will fail despite really being in pretty good shape fontlint flags a lot of things that I am not convinced are actually errors at all so if you run it and get nothing but complaints don t panic At the very least you will need a PE script interpreter with my all extrema patch it is included in recent versions of FontForge and all
44. e to throw away the kerning table at this point and just use the bearings then in some sense these bearings would give the best possible approximation of the discarded kerning table Then for each left class which is actually a class of right contours it is a class of glyphs that can appear on the left in a kerning pair the program finds the maximum that is farthest apart amount of kerning between that left class and any right class Two thirds of that kerning amount are added to the right bearing of the left class The concept here is that we generally want kerning to be pushing things together not pulling them apart so the default amount of kerning indicated by 84 the bearing should be near the maximum distance apart from which individual entries can then push things closer Also we generally want most two thirds of this adjustment to happen to the right bearing of the left hand glyph in the pair Then to clean up the rest the program examines each right class which is a class of left contours and similarly finds the furthest apart kern pair and adds that to the left bearing of the right class adjusting all kern pairs appropriately At this stage it s guaranteed that all the kern table entries will be zero or negative kerning only pushes glyphs together from where they would otherwise be it never pulls them apart Kern table entries are dropped if they are less than ten font units that cuts the size of the table con
45. emove overlaps RMO and end up in pfb The make ass script also scans the Metapost files to get a list of defined code points and some other metadata its main output is a FontAnvil script that assembles the page fonts into full coverage OpenType or FontAnvil native fonts Non proportional fonts become OpenType and are finished here proportional fonts become FontAnvil native and are processed further The assemble font script also makes use of per family feature files from make fea defining things like glyph composition and per font feature files from make name defining metadata table contents The proof files in prf are generated as a side effect of compiling the PostScript page fonts and pass information about glyph construction to both the automatic documentation generator and the GPOS table generator 71 std sfd er euS pe bdf pe pe getbearings pe bdf bdf bearings kerner c prf prf Tu zc kerner make gpos TA pe bearing pe fea ktmp fea fea fea SS oe fea kern fea lr gt otf otf fea gpos fea Figure 5 Additional processing for proportional fonts The SFD file containing the font outlines is converted to a BDF bitmap font which feeds the kerner program That generates a ktmp feature file with the actual kerning data and a FontAnvil script to apply the chosen bearings At right the make gpos script reads the proof files to
46. ers other than the 26 ASCII ones and I may start using unencoded glyphs rather than mapping them into the PUA In monospace fonts these small caps are about the same height as most lowercase letters in order to allow their use as substitutes when combining accents are applied to uppercase letters In proportionally spaced fonts small caps are a little taller than lowercase letters but shorter than ordinary caps AE AFIV ILI Tb Heavy Metal Umlaut ss01 With Stylistic Set 1 OpenType feature ssO1 turned on umlaut or replaced by a heavy metal umlaut intended for spelling musical names like Mot rhead Spi al Tap and Mormon Tab rnacle Choir The heavy metal umlaut differs from the regular umlaut in that the dots are larger and pointier Glyphs to support this feature including a spacing heavy metal umlaut character are encoded into the private use area at U F1 40 through U FT74E ADE Enclosed Letters and Numerals ssO2 With Stylistic Set 2 OpenType feature ss02 turned on the enclosed characters described by Unicode become available as contextual sub stitutions for sequences of in most cases ASCID characters 0 gt through 50 gt 39 A gt 0 through Z gt 0 a gt through 2 gt 7 gt through C2 gt 1 gt through 10 gt 0 gt O through 20 gt O A gt O through Z gt 6 A gt A through Z gt e lt A gt g
47. erscore marks it as requiring special handling in the Perl wrapper and are removed often leaving fewer than four tokens in the result Short names are always derived from the Hamlog name before rewriting Subfonts also have short names of their own formed by adding an extra token at the end which is the two or three digit hexadecimal representation of the page number For instance tsuku mi ps OO would be the short name of the subfont covering U O000 to U OOFF of tsuku mi ps These names are often used in constructing filenames for intermediate files during build Anyone contemplating modification of this system should be aware that short name tokens other than page numbers must not be valid hexadecimal digit sequences that could be mistaken for page numbers in other words each non page number token should contain at least one letter in the range g to z The build system needs to be able to distinguish subfont and full font short names from each other by testing for a hexadecimal number as the last token To prevent terms like extra bold and demibold from abbreviating to valid hexadecimal numbers terms derived from the word bold are abbreviated as if it 47 were spelled qold or bqld Fragments of short names are also used to name source code files Files that follow this scheme not all source files do have names consisting of one family token a hyphen and one other token followed by the appropriate e
48. evel and some support for that exists already but it probably doesn t work yet The enable ot features option works the same way for OpenType features note that Openlype features may also be automatically and silently disabled in whole or in part overriding this setting if you have used the previous setting to disable characters necessary to implement the features For instance you cant have Openlype contextual substi tution support of fractions or enclosed numerals if you have disabled 56 the numeral characters though you can build a font with enclosed and not regular numeric glyphs because glyphs are mostly indepen dent of each other This automatic disabling may not be perfectly clean either in some cases disabling a character might not disable the feature code that mentions it and in that case FontAnvil may create an empty glyph for the character even though you disabled it That shouldn t happen unless you attempt something very unusual and non standard however Currently supported tags are all none and a four character tag corresponding to the Openlype name of each feature that exists and a few that don t See Section for a detailed list The enable styles option allows similar selection of type styles For instance tsuku amp kg el would enable all Tsukurimashou Kaku fonts in all weights and both monospace and proportional except no Extra Light fonts The atomic tokens for this selection incl
49. f It passes configuration information mostly in command line arguments to pretty much all parts of the system NAT Hamlog Evaluating the consequences of build options like enable chars re quires doing a certain amount of logical inference for which shell scripts are not well suited It might be possible to get GNU Make to do the necessary computations inasmuch as its quite programmable already required to build Tsukurimashou and fundamentally a logical inference engine at heart But that would probably involve creating many tiny files corresponding to logical propositions which would waste space and cause other problems on some filesystems A more elegant approach would be to use a real logic programming system ie Prolog which happens to be one of my research interests But I didnt want to create a dependency on a Prolog interpreter because 70 fea fsc sum sum mp mp txt jnames txt chars lst BW make ass pfbtmp pfb prf prf hamlog jnames hl make fea pe rmo pe make name tsukurimashou nam fea fea pfb pfb fea name fea sfd flogged sfd Oe 2 pe assemble font pe sfd sfd otf otf Figure 4 Building the fonts themselves Checksums in sum provide fine grained change tracking on the Metapost files in mp which define Postscript page fonts each covering up to 256 code points Those are processed to r
50. fault j multi CPU option In version 0 6 Expect was rendered less necessary by the replacement of mtypel3 with a homegrown and less chatty Perl script but the support remains in place to handle TEX s lesser jubilations You can also add KEEPTMP T to prevent deletion of the temporary directories created while running Metapost This feature is primarily useful for debugging the pfb generating scripts themselves Several options to configure share a common argument syntax under which you can specify a comma separated list of commands to include and exclude subsets of things This syntax is actually translated into Hamlog code and evaluated by the logic programming system during build It may be easier to understand by example than to explain rigorously but here is an attempt at a precise explanation A specification consists of a sequence of commands separated by commas or spaces Each command either adds or subtracts some 55 subset of the things being configured to or from the currently selected subset which is initially empty A command consists of one or more alphanumeric tags representing subsets separated if more than one by ampersands or colons which mean AND and optionally prefixed 66 99 by plus minus or exclamation point If the prefix is or then the AND of all the tags will be subtracted from the current selection otherwise it will be added The allowed values of the tags depend on which option we are con
51. files need to be updated requires thousands of lines of code spread across several languages and systems so this is an approximation It reads the makefile vars inc file and attempts to identify all the Postscript SFD and OTF font files to be built in the current configuration Then it checks the time stamps of all those files Gif they exist at all and computes what fraction of them have been updated since the current build began It knows when the current build began by checking a tag file that is updated in the kill empty step already wedged into an Autotools hook to execute at the start of a full build thus this will only work when the current build actually is a full build initiated by just typing make rather than asking to build individual files The completion estimate will also only be correct when all those files actually need to be built that is building 66 from a fresh copy of the distribution or after make clean It will probably produce incorrect results if a build is not actually in progress for instance after a build has completed or been aborted And the ETA estimate will usually be wrong Despite all these limitations it s a fun thing to run in a loop in a window to get some idea of where the build stands beyond the endless columns of MT1 PERL RMO messages put out by make udcpyright check that all years in which each file was modified are included in its copyright notice This assumes one has an SVN checkout i
52. for later stages of processing 9 mpdep generates a list of the dependencies of a METAFONT language file or a set of them for automated Makefile gen eration Compare with autodep which actually edits the files to include the other files they should based on the macros they use mpdep only looks at what other files each file does include and reports that 69 progress invoked during build to give progress reports and completion time estimates CJLFOZATAOBd m Build system diagrams 69 foo bar foo bar comes with the distribution naninani dat naninani dat is generated during build A B A is input for B C gt D C creates D Esprit gt F E sometimes creates F Figure 3 Legend for build system diagrams This section is still under construction but it is planned to be a graphi cal summary of what is going on in Tsukurimashou s rather complicated Makefile Figure 3 describes the symbols used and Figure 4 shows the process by which fonts are built Proportional fonts go through additional processing to generate the kern tables and so on as shown in Figure 5 Be aware that these diagrams are meant to clarify not to formally describe the code this is not a formal modelling language like UML and the semantics of the symbols are not necessarily entirely consis tent Also left out of the diagrams are language interpreters like Perl and a great deal of information flow to and from the main Makefile itsel
53. he default value is all Be aware that just because you selected a feature during config does not mean you can actually use it Features are only included in fonts if they make sense for the particular font being built thus monospace fonts will not contain proportional only features and if all necessary character codes are available thus fonts with Japanese glyph coverage will not contain shaping features specific to Korean script Also build time configuration only controls what will be included in the font Given that the feature is in the font your renderer must then support the feature you want to use and you may need to adjust the renderer configuration to tell it to use the feature Some renderers make that easier or harder to do than others and some renderers do not turn on by default certain features that I recommend and the standards say should be turned on by default Note that although Openlype would permit such a thing features in any single Tsukurimashou font are NOT language or script specific for instance the set of features available for Latin script is the same as the set of features available for Japanese script in the same font and the behaviour of each feature does not change depending on whether the text is tagged as English or Japanese This even extends as far as for instance the fact that you can really kern kanji in Tsukurimashou proportional fonts A feature applied to a code point sequence for which
54. https www tug org members TUGboat tb341 3 tbl08skala pdf it will eventually be come open access Most of the source code for the article maybe not including the absolute latest edits made by the journal editors instead of by me is in the tug2013 subdirectory of Tsukurimashou s version control system In this version the Korean language fonts have had their names changed to Mandeubsida from talking to some people at the TUG meeting it appears that this is a better translation of the name Tsukurimashou into Korean The GSUB features to support Korean glyph shaping have 11 been redesigned to better work with some shapers in particular recent versions of HarfBuzz Also in this version build system support for an experimental font family called Kazoemashou lets count has been added this is intended to become a Unicode Math font based on Tsukurimashou but as of this writing it is very fragmentary and not usable An academic article about IDSgrep is currently in submission to ACM Transactions on Asian Language Information Processing You can read a preprint version at http arxiv org abs 1404 5585 Version 09 was prepared in a hurry and there are some ongoing questions about the future of Tsukurimashou as a result of my life and employment situations This release is a fair bit less polished than other releases have been and some parts of the documentation especially in the parasite packages may not have been pr
55. iables start with an uppercase letter or an underscore and may contain uppercase alphanumerics and underscores Atoms start with a lowercase letter or numeric digit and may contain lowercase alphanumerics and underscores For Prolog compatibility atoms starting with digits should not contain anything other than digits and the only atom starting with zero that should be used is zero itself but Hamlog doesnt care about those points Things containing a mixture of upper and lowercase alphabetic characters should not be used The special tokens and are technically treated as atoms too but you should only use them in their typical meanings of cut and unification and should only be used with the 66 general prefix syntax applicable to all functors not as an infix operator see above Variable names starting with and especially the unique variable name consisting entirely of an underscore are not special in Hamlog Beware that means fooC D contains only one variable occurring twice when interpreted as Hamlog not two distinct variables as in Prolog and it violates the only one appearance of each variable in a head guideline The unique variable name consisting of one underscore is probably best avoided entirely But it may be desirable to use variable names starting with underscores anyway in some cases because of their specialness to Prolog interpreters I was really tempted to allow and use arbitrary UTF 8 in partic
56. ibold in particular so that Korean fonts wont end up with Japanese in their names though that aspect is not yet fully implemented Here are the rewriting rules tsukurimashou maru extralight X is rewritten to tsukuri mashou tenshinokami normal X tsukurimashou maru extrabold X is rewritten to tsukuri mashou anbiruteki normal X 48 kazoemashou X normal proportional is rewritten to kazoe mashou X normal monospace Bearing in mind that monospace translates to the empty string in the long name the point here is to avoid a redundant PS in the long name see below for the Kazoemashou family where there is only a proportional version The Hamlog name is translated into a few other kinds of names by a combination of Hamlog code in select hl and Perl code embedded in Makefile am One important translated name is the short name which is a sequence of two three or four short ASCII tokens which in different parts of the code may be separated by either hyphens or underscores An example short name is tsuku mi ps corresponding to the Hamlog name tsukurimashou mincho normal proportional Each element in the Hamlog name is translated into a token for the short name according to the table given by the short name 2 predicate in select hl but some Hamlog atoms namely normal and monospace translate to the special value da for lambda meaning empty string triple und
57. ibraries C header files or similar at all CAlthough some of the parasites may Cross compilation and executable name munging were removed for the same reason some C programs are compiled for kerning and Type 1 opcode assembly but these are only meant to run on the host system during build with all the installable files being architecture independent The hacking I did on Autotools 59 means that if you modify the build system such that you would be re running Autotools it s likely to break unless your version of Autotools is close to the 2 65 version 1 used The configure script will try to detect such a situation and warn you The design of this build system was influenced by Peter Miller s in teresting article Recursive Make Considered Harmful available at http miller emu id au pmiller books rmch The fonts as such are all built from a single large Makefile However in the current version there is also some use of recursive Make and recursive Autotools to support the parasite packages next section should those be enabled Some notes on the design of the build prior to the introduction of par asites are in my Web log article at http ansuz sooke bc ca entry 226 I will probably write a similar article about the parasite mechanism at some point in the future BENI Parasite packages Some of the technology developed to support Tsukurimashou has more general applicability For instance the stripped down version of M
58. ild it The completed font files end up in a directory called otf in the distribution tree A few ready made ones should be there already when you unpack the distribution for the benefit of the probable majority of users who cant do their own compilation If you run make install it will attempt to install the fonts and documentation files in sensible locations but it s not really customary for fonts to be installed like other software and you may be better off simply taking the compiled files from the otf directory and doing with them whatever you would normally do with fonts instead of using the automated install At this time there is no TpX specific font installation support though that might be a nice feature for me to build in the future The install target uses the prefix and similar options to configure in a reasonably obvious and standard way If you have KANJIDIC2 place it in the gzipped XML form in which it is distributed in the build directory the doc subdirectory off of the build directory your system s dict or share dict directories possibly under usr or usr local or your configured installation pre fix or your home directory or dict or put it somewhere else and tell configure where with the with kanjidic2 option Although what s used by Tsukurimashou is factual information not subject to copyright and KANJIDIC2 which also includes creative content that might be subject to copyright is d
59. infinite exponent the kerning would be determined solely by setting the closest approach to the optimum without reference to any other points That is how most auto kerning software works but the results tend not to be good because in a serif font with a pair like AV inserting the ideal vertical to vertical distance between the serifs is going to place the stems which are much more visually important too far apart Using an exponent somewhat less than infinity causes the stems to still have some significant weight The value 3 was chosen by trial and error and may be subject to further adjustment Once all the class to class distances have been chosen it remains to choose the bearings for the characters Recall that kerning was computed from margin to margin that is the amount of space to insert between the strict bounding boxes of the glyphs Adding a certain amount of extra space to the glyphs themselves and subtracting it from the kern distances may result in a better more concise kern table as well as better results when glyphs from this font are set next to spaces things from other fonts and so on The first step is that the kerner program finds the average of all kern table values and puts half that much bearing space on either side of every glyph adjusting the kern values accordingly This has the effect of giving every glyph an average amount of space and changing the overall average kern adjustment to zero If we wer
60. ints U F1731 to U F1738 and always available that way If the ornm feature is turned on then they will also appear as substitutions not alternates for the ASCII capital letters A through H so that the text A BIG TEST comes out as 9 OIS TEST It is likely that the way this OpenType feature works will change in the future since it seems not to be current best practice to implement ornm by substitution but I m not quite sure yet what the best practice actually is These ornaments look the same in all the fonts they do not change from one style to the next FontAnvilZ lt 00 9 T7 9 FontAnvil specific Metadata pfed This is not the usual kind of OpenType feature but it can be turned on and off through the same build system interface When selected it causes the build system to add a pfed table with a flog sub table in the generated OpenType fonts containing verbose information about the installation on which the fonts were built This table is a FontForge FontAnvil specific extension of OpenType format the name pfed refers to PfaEdit an earlier name of the software that later became FontForge Other software will ignore this table but users with FontForge can examine the verbose metadata under the FONT LOG heading in the Font Info dialog The data is also visible as a chunk of plain text near the end of the file when the OTF file is examined with a general purpose file viewer such as less
61. invoked even if as here the buggy files are not actually being used Building Tsukurimashou requires FontAnvil or FontForge if you do not already have that installed then the Tsukurimashou configure script will automatically build FontAnvil However some C source files for FontAnvil are not actually written under human control but by the Icemap utility 1f you were building from a distribution tarball those files would be included ready made in a version control checkout they must be built using Icemap and that means Icemap in turn must be built So the configure script will automatically build Icemap too if you have neither FontAnvil nor Icemap already Running configure with no options should be able to generate a usable configuration if the necessary prerequisites are available but it is likely to give some warnings suggesting better options configure Read any warnings produced run configure help for more infor mation or check other sections of this document and if you wish run the configure script again with your choice of other options For instance configure enable ot features al l 64 The C sources built by Icemap for FontAnvil are based on Unicode character database information which must be downloaded from the Unicode Consortium so there is an extra make download step or manual equivalent necessary before the main build when building from a source control checkout in the Icemap FontAnvil
62. ir values it is dangerous to use the same variable more than once in the same head so constructions like fooCXY XO should not be used If you want to do that you should instead write foo XY Z X Z Note the non sugary use of as a functor since the more common infix notation isn t supported Note also that there should be a space between and Hamlog doesnt require that but it may reduce the likelihood of parsing problems should the same code be run on interpreters other than Hamlog Variables in clause bodies are renamed once using the clause serial number when the clauses are loaded as a result if the same clause body gets expanded a second time while variables from its earlier 74 expansion are still unassigned there could be trouble This is not a very likely scenario but it s worth mentioning Clauses in the database have serial numbers and when a choice point goes on the stack the serial number of the clause at which it matched is part of the record on the stack Then if the interpreter backtracks to re satisfy a clause it writes the regular expression in such a way that it can match all and only serial numbers greater than the last place it matched Creating an integer greater than N regular expression was surprisingly difficult its a simple enough concept but there are several cases that all must be handled properly or weird bugs turn up Syntax is simplified from Prolog Var
63. istributed freely it is not dis tributed under a GPl compatible license and so for greater certainty I m avoiding including KANJIDIC2 or any data extracted from it in 54 the Tsukurimashou distribution What you get if you have KANJIDIC2 is any features that depend on knowing the grade levels and similar properties of kanji characters such as the chart showing how much of each grade has been covered which is important for my own de velopment efforts in planning what to design next and knowing when to release The build system for Tsukurimashou is fairly elaborate it may seem like overkill but given that I m running it on a daily basis over the course of years its important that it should be pleasant for me and efficient in time consumption Thus it defaults to silent build rules uses pretty ANSI colours and does a bunch of complicated checks on whether file contents not just modification times have changed in order to avoid expensive dependency rebuilds unless those are really needed while triggering them automatically when I add new source files One important tip is that if things are failing you can add V 1 to the make command line as in make V 1 pfbtmp tsuku kg OO pf b to temporarily disable the silent build rules and see what s going on Note no hyphen because this is a variable setting instead of an option and you will probably also want to specify a target filename which implicitly overrides the de
64. ith the RawFeature option It is possible that more careful language tagging specifically of Korean text in the TpX input might reduce or eliminate this issue Less relevant now January 2014 that Mandeubsida no longer uses limo and jmo 9 Version 24 of the fontspec package which I first encountered in TpXLive 2014 changes the syntax of the fontspec macro to place its optional argument after not before the required argu ment It claims to be backward compatible but this is a problem when the macro is followed immediately by text intended to be written in literal square brackets as a result Tsukurimashou user manuals prior to version O9 cannot be typeset with new fontspec Switching documents to the new syntax would break compatibility with earlier versions of fontspec Instead recent Tsukurimashou documentation inserts relax commands to pre vent square bracketed text from being interpreted as an optional argument 077h 7rF Other software ATEX tocloft package as of May 2011 it sets an entry without a dot leader by actually requesting an entry with a leader but with a font dependent invalid spacing between dots The DVI renderer is supposed to reject the bad spacing and not set a leader at all With default font sizes in most DVI renderers that results in either the correct appearance no leader or an almost correct 28 appearance a single dot instead of a leader But with the larger sizes of the extsize
65. ity in this project for other rea sons anyway Thanks to Shriramana Sharma and the Metapost mailing list for discussion encouraging me to work on this issue S METATYPET infrastructure in general sometimes generates paths that some software layer possibly METATYPETs own code flags as degenerate triggering a fatal error Workaround is to fil ter things before rendering through the regenerate macro in intro mp which removes any very short path segments Possibly related the Fill macro will sometimes abort in response to some conditions on turning number that do not appear to actually be harmful Workaround is to use our dangerousFill in intro mp instead which is just a copy of Fill with the error checking removed Since the switch to bundled code based on METATYPE1 version O 55 this seems to be a less significant issue FANIA UE FontForge FontForges list of external dependencies continues to grow as does its emphasis on Python There has been talk of remov ing native scripting from FontForge altogether which would 23 be a disaster for Tsukurimashou Rather than waiting for Font Forge to become unusable and then panicking I ve forked the project FontForges source repository now includes a directory for FontAnvil a standalone interpreter of FontForges native scripting language Tsukurimashou s build system as of version 09 will look for FontAnvil and use it in preference to Font Forge if found My hope is that
66. ks trial and error needed to figure out how frequent they ought to be When FontForge saves an Adobe feature file what it writes may bear only a passing resemblance to what was actually in the font and in particular cannot subsequently be loaded by FontForge 24 sometimes cannot be loaded at all other times can be loaded but the loaded tables behave differently from the ones that were saved Workaround do not save feature files from FontForge As of October 2012 there has been some interest shown by FontForge developers in addressing this issue but it will likely be a gradual process instead of a single complete fix FontForge can segfault while writing a feature file because of dereferencing a pointer first and checking whether it was null afterward Fixed by the FontForge developers in December 2011 FontForge does not apply features of the DFLT language system as a default to Unicode ranges that have no explicit mention in the font file Adobe spec says it should Workaround explicitly list language systems to cover every Unicode range we care about FontForge may write up to 23 horizontal stem hints per glyph when writing a PostScript flavoured OTF font resulting in a PostScript stack overflow on loading if the glyph also has a non default advance width Went undetected for a long time apparently because only CJK fonts are likely to have so many horizontal stems and CJK fonts are likely to be monospace and thus wont
67. ku Square Gothic sans serif with squared stroke ends Q YELLA Tsukurimashou Maru Round Gothic sans serif with rounded stroke ends O FVELISPOE JLE Tsukurimashou Anbiruteki C Anvilicious 2 extra bold rounded sans serif O lEUXLEOX BoSX Tsukurimashou Tenshi no Kami CAngel Hair very thin hairline display font O FUEL M gt IR Tsukurimashou Bokukko Tomboy felt marker style fEUXLr58jSH Tsukurimashou Mincho Ming Dynasty mod ern serif The remainder of this document is reference material describing the special features of the fonts as well as instructions on how to build and customize your own font files including the styles that arent distributed in precompiled form 31 OpenlypeD 74 Fr Openlype Features OpenType contains a mechanism for defining what are called features in a specific technical sense rather than the more general usage of that word They are identified by four letter tags and generally correspond to extra data added to the font that compatible renderers can interpret to provide special typesetting effects The Tsukurimashou build system s configure script accepts an enable ot features argument which can be given a list of feature tags or all each optionally prefaced by an exclamation point to negate it These are processed left to right so that a setting like enable ot features all ccmp will enable everything except the ccmp feature T
68. l the hangul individual jamo code points whose layout is handled by OpenType substitution features and kerning would just make it too complicated and the special glyphs used by vertical fraction composition much the same situation as the hangul jamo The kerner C program reads that BDF font puts all the glyphs into a common bounding box big enough to contain any of them and finds the left and right contours basically the x coordinates of the leftmost and rightmost black pixels on each row for each glyph as well as the margins which are defined as the x coordinates of the leftmost and rightmost pixels on any row of the glyph There is some special processing applied to the contours to make them more suitable Consider what happens in a glyph like many horizontal rows namely those above and below the entire glyph and those in between the two lines contain no black pixels at all and so the leftmost and rightmost black pixels in those rows are formally undefined If we set it next to another glyph like does not then a kerning algorithm that looked only horizontally might let the period slide all the way under the equals and out on the other side There has to be some vertical effect to which only has ink bboy in rows where prevent that So the Tsukurimashou kerning program makes a couple of passes over each contour one forward one backward to enforce the following rules which except for glyphs containing n
69. lEX developers in replacing ICU with HarfBuzz which is known to handle this case correctly and may have other advantages too 9 x3IpX fontspec package in some versions has trouble with treat ing its WordSpace configuration option as a multiplier when font size changes in a way that is most noticeable when using monospaced OTF fonts complicated by the fact that it s hard to define just what should count as a monospace font Discussed at length on the X4lEX mailing list in February 2011 http tug org pipermail xetex 2011 February 020065 html and eventually resulted in three items and planned fixes in the fontspec is sue tracker https github com wspr fontspec issues 9 https github com wspr fontspec 1ssues 98 https github com wspr fontspec 27 issues 99 The main bug has now been fixed and it has also become less relevant to Tsukurimashou documentation since the introduction of the proportionally spaced fonts but still bears some watching If they implement all my suggestions then the current PunctuationSpace multipliers in Tsukurimashou doc umentation will become much too large and need to be reduced Earlier versions of Tsukurimashou toyed with workarounds based on the everysel package or on poking into the internals X4IlpX and fontspec do not apply by default some OpenType features that are supposed to be applied by default in particular limo and vimo Workaround manually request them w
70. le WF Cea Eure Sieh aea oo AR NzZ7 V8 amp Korean Language Support Tay bh OBAAIOL T Regarding Font Names MeEYELESae HEYELES Building Tsukurimashou EMPATE Build SSI cr ABRA WS t i 7 7 Parasite packages ATV EH TATLO NDbUJNLE Building from a version control EOBERQUES c sos RAK GRRE REG RK RRR AA Y ILNMEB t Tool documentation o oo eee EJNLROZATLAO BIA Build system diagrams GUS Hamlog es irte A gt ILhHk Kerning 4v vpo Introduction BAZIDIH IE TZIPIM E EBIO E cA dS sIiEBBxcoa yvgv ar ILEDBEF I want to learn Japanese That s a large project likely to involve years of daily study to memorize a few tens of thousands of words It s clearly within the range of human instrumentality because many millions of Japanese people have done it but most of them started in infancy and it s widely believed that peoples brains change during childhood in such a way that it s much harder to learn languages if you start as an adult I also would like to be fully literate within somewhat less than the 15 years or so that it takes a native learner to gain that skill What about designing a Japanese language typeface family Typeface design is a difficult activity requiring hours of work by an expert to design each glyph A typeface for English requires maybe 200 glyphs or so one for Japanese requires thousands That makes the English language typeface about a year s full time work and
71. m the codes perspective the unrewritten Hamlog name is the true name of the font and all others are derived from it The source of the translation table for JK names is in txt jnames txt which is processed by Perl to change the UTF 8 into a hexadecimal form that Hamlog can handle the result going into hamlog jnames hl 5 EYELE TEUE Ux Building Tsukurimashou Please note that if you just unpack the Tsukurimashou distribution and look in the otf subdirectory you will find some ready made font files there If you are content with them then those are the only files you need you can safely delete everything else in the package maybe save the PDF documentation files if you want and ignore this section These notes on building Tsukurimashou are only for expert users who want the greater control wider style coverage and intellectual challenge of doing custom compilation First I will describe building from a source distribution package If you wish to compile from a version control checkout then also read the specific notes about that in Subsection Assuming you wish to embark on the adventure of compiling Tsukuri mashou you must have at least the following 9 A reasonably standard Unix command line environment I use Slackware Linux Anything branded as Unix should work MacOS X might work Windows with Cygwin might work A standard C language tool chain normally comes included with Linux Perl GNU Make
72. might also think it could be used to trace the origins of anonymously written PDF files and that forensic investigators dont have many other ways to do that Such people are wrong However my saying so wont cause them to change their minds If I distributed software that attached this information to generated fonts by default then someone who didnt read the documentation would eventually discover it and make a big fuss about it supposedly being a security hole Who needs that The feature is therefore turned off by default I recommend turning it on by passing the enable ot features all option to configure The default is enable ot features all pfed which enables everything else Regardless of the setting chosen the build system will place the same information in a file called fontlog txt in the txt subdirectory of the build tree After doing a build you can read that file to see what would have been put in the fonts The option setting just controls 38 whether or not the fontlog file will be added to the OTFs during the final packaging step AT ILE VEIL Small Caps smcp SMALL CAPS ARE AVAILABLE THROUGH THE OPENTYPE SMCP FEATURE This feature simply substitutes the 26 lowercase ASCII Latin letters with alternates that are encoded into the Private Use Area at the code points formerly used by Adobe for this purpose U F76l to U F77A At some point in the future other glyphs may be added to support small cap lett
73. n added paragraph clarifying that they may be embedded in documents See the files COPYING and COPYING GPL3 and note the following addition As a special exception if you create a document which uses this font and embed this font or unaltered portions of this font into the document this font does not by itself cause the resulting document to be covered by the GNU General Public License This exception does not however invalidate any other reasons why the document might be covered by the GNU General Public License If you modify this font you may extend this exception to your version of the font but you are not obligated to do so If you do not wish to do so delete this exception statement from your version The license means and this paragraph is a general summary not overriding the binding terms of the license that you may use the fonts at no charge you may modify them you may distribute them with or without modifications but if you distribute them in binary form you must make the source code available Furthermore this is where font embedding becomes relevant embedding the font for instance in a PDF file does not in itself trigger the source distribution requirement My plan is that at some point in the future when the fonts are in a more useful and complete form I will make precompiled binaries 8 available through commercial online channels That will serve several purposes it will allow me to make some money from my
74. n pattern for their arguments atom true if the argument is an atom This is implemented by rewriting the database entry for atom 1 on the fly when you call atom foo it magically changes to having been defined as either the single fact atom foo or nothing depending on whether foo is an atom atom finalCAZ A Z where AZ is an atom whose last character is Z and A is everything except the last character Used for building and tearing apart atoms like page00 This requires some careful handling in other interpreters because Hamlog has no concept of quotation marks and treats single digit integers 76 exactly the same as atoms whose names are the ASCIT digits real Prologs have more subtle type handling As with atom 1 this is implemented by magically creating appropriate clauses on the fly 9 2 This also creates clauses on the fly At least one of the two arguments should already be atomic when the goal executes though that is rarely difficult to guarantee in practice And that s it for built in predicates Note that goals in a clause body also can only be combined with comma for conjunction no semicolons for disjunction and without them parentheses become unnecessary and are not supported either There is also no syntactic support for negation However you can and the existing code does compute negation and disjunction using multi clause predicates cut and fail What you cant build is any kind of
75. n the current directory intended to be a checkout from my private SVN repository but it would probably work with one from the OSDN repository as well It scans all SVN controlled files for anything that looks like the project s standard copyright notice and then reports on any files for which the set of years mentioned in the notice does not match the set of years in which SVN has log entries for that file The idea is that once a modified version of the file has been checked into SVN in a given year it has been published in that year and the year should be mentioned in the copyright notice but when a new year comes we dont want to just go through and modify all files solely to insert another year in the copyright notice By running this every so often and manually updating the notices when appropriate the hundreds of copyright notices in the project should painlessly converge on correctly reflecting the years in which files were actually changed For completeness here is a summary of the functions of all the other scripts in the tools subdirectory Most of them are written in Perl 6 add flog attach the font log file to a FontAnvil SFD format font this requires re encoding it to UTF 7 format livecode chase file inclusions and determine all the live that is actually invoked macro definitions in a list of METAFONT language files The output of this is not because lines may appear in the wrong order actually
76. nese but version 0 5 also introduced a series of fonts called Mandeubsida designed for Korean hangul There is also some experimental code intended to eventually become a set of blackletter fonts but it is disabled by default and not currently usable The main goal of this project is for me to learn the kanji by designing glyphs for them and so far it does seems to be helping my learning process though the nature of computer programming is certainly having some unusual disruptive effects For instance in some cases I appear to have unintentionally memorized the Unicode code points of kanji without learning their meanings or pronunciations I may end up learning to speak Japanese just like a robot but by the time I m fluent the Japanese demographic collapse will be in full swing and they ll have replaced much of their population with humanoid interfaces anyway so maybe I ll fit right in Please understand that the finished product is not the point so much as the process of creation hence the name Furthermore although the fonts cover the MES 1 subset of Unicode and thus can in principle be used for almost all popular languages that use the Latin script and I know that at this stage most users are probably more likely to use the fonts for English than for anything else nonetheless these fonts are intended for eventual primary use with the Japanese language Some decisions on the Latin characters were driven by that consideration in pa
77. o black pixels at all which are removed from consideration fully define where the contour should be 79 69 The right contour cannot be any further left than the rightmost black pixel in the row 9 The right contour cannot be more than 10 font units one pixel further left than its value in the next or the previous row 9 If the right contour in the next or previous row is left of the left margin then the right contour in this row cannot be more than 3 font units further left than in the next or previous row 9 Subject to the above rules the right contour is as far left as possible The left contours rules are the mirror image of these These rules can be imagined as simulating something like the way letterpress printers kern type by physically cutting the metal type bodies at an angle to fit them more tightly together it s bad to cut off part of the actual printing surface you basically cut at a 45 angle but with glyphs that only have a small vertical extent so that the 45 angle would cut all the way across to the other side then you want to use a more vertical angle in this case 16 7 from vertical so you dont end up setting the next character actually earlier on the line Figure 6 shows a typical example of the contour computation The right margin is subtracted from the right contours and the left margin from the left contours to get for each glyph a vector of numbers describing its shape along th
78. onts look like and list the current kanji coverage Better documentation and some day Japanese language documentation will probably appear in a later version at the moment I m just more interested in designing fonts than in writing about them Of course all the typesetting in this manual is done with fonts from this package The name Tsukurimashou could be translated as Lets make some thing From time to time people ask how they can help with the project The one thing this project most needs is your attention I am not looking for code contributions because of the pedagogical goal part of the point is for me to learn kanji by doing all the work I am not looking for money as such although 1 could certainly use academic employment but public and third party attention would be very helpful to the project s long term success Use it share it and above all write about it At this point I m not even sure anybody but me has ever so much as run the Tsukurimashou build process and that really hurt my motivation to continue maintaining it as a published free software package when other demands in my life pile up The home pages for this project where you can download the latest releases browse the source control repository and so on are http tsukurimashou osdn jp index php en English http tsukurimashou osdn jp index php ja HA 55 SLUBRIUILES Matthew Skala mskala ansuz sooke bc ca August 17 2014 090 a3
79. operly updated I am releasing it as it stands because after a year we need a new version the kanji planned for inclusion in this version are complete at least and it may be a long time before I have another chance I mentioned in last version s release notes that I was unhappy about being single and unemployed It took me a year and a half to find an academic job During that time I lived mostly on savings with occasional short term contracts During that time my availability to work on Tsukurimashou was limited and development was slow I eventually obtained a postdoctoral position at the IT University of Copenhagen working for a project called Scaleable Similarity Search This project s subject matter is a close match to the kind of thing I wrote about in my doctoral dissertation and I am excited about getting to do more work on it As I write this text August 16 2014 I am in the middle of packing to move from Canada to Denmark Now I am rushing to get the Tsukurimashou O9 release out the door before I leave because if I dont it wont go out until much later if ever Very soon I will be shutting down and packing my main development computer so it can be shipped and it will probably be at least two months before I can use this computer again Once I arrive in Copenhagen I don t know if I will be spending a lot of time trying to learn to speak Danish and if so what that will do 12 to my Japanese studies Despite strict Danish gove
80. possible to just turn off Freelype rasterization on individual bitmap creation op erations but that option doesn t seem to be available from the scripting language and so the package has to actually be built without Freelype As of April 2012 seems not to be an issue with Arch Linux s packaged versions However as of a few minutes before the Tsukurimashou 0 8 release ouch this is a problem again with the lastest FontForge and the latest FontForge now requires Freelype so the workaround of disabling Freelype no longer works I ve reported it as FontForge GitHub issue number 685 Chttps github com fontforge fontforge issues 685 am hack ing my local copy of FontForge to generate distribution fonts and will hope to get it resolved properly in mainline FontForge as soon as possible FontForge attempts to modify the names of all glyphs that it thinks are related whenever a glyph name changes for any reason Related glyph renaming has wacky consequences when one glyph name is a substring of another and the Beikaitoru glyph naming scripts among other things in Tsukurimashou have to take it into account This issue was reported as FontForge Github issue number 523 Chttps github com fontforge fontforge issues 523 and patched in their mainline I dont understand FontForges release schedule if they have one but presumably distributed versions will someday contain the fix Until were sure that day has passed Tsukurimashou will
81. qualification because many projects with similar aims have existed in some form for a long time Here are some others going back a few decades that you might want to check out This is not intended to be a complete list in particular I m leaving out many sources of CJK fonts that are not METAFONT related and many academic papers that are not associated with publically available fonts There is also no doubt a great deal of research and development locked up inside commercial organizations or published in the Chinese language and thus inaccessible to me LCCD A Language for Chinese Character Design Tung Yun Mei 13 Stanford technical report STAN CS 80 824 He built a language similar in nature to METAFONT collaborating with Knuth and sharing ideas with the earliest versions of METAFONT and con structed 112 kanji for use in TAOCP Its interesting how little has changed since this early work A Chinese Meta Font John Hobby and Gu Guoan paper in TUGboat 5 2 1984 Proof of concept and discussion of some of the graphic design issues for parameterized CJK fonts They built 140 radicals and 128 characters using infrastructure very similar to Knuth s techniques for Latin fonts and the traditional Chinese stroke based analysis of characters High quality parameterized designs No apparent plan to actually turn these into a usable full coverage family it seems to have been meant as research into the techniques only http www
82. r they come up regardless of their level For instance when I added the gate radical PA it was easy to add many other kanji that consist of that wrapped around an easy pre existing kanji even though some of those are not in common use and one H isn t even a real kanji at all having been created by an error in the standards process But having just over a thousand in the main line roadmap makes the end of the kyouiku 17 kanji a good milestone for the first major version number That may be a few years from now Progress past that point will be somewhat dependent on how I feel about the project by then and what my personal career situation is My hope is that at that point or before I l have the chance to present this work at one or more conferences and that it will have attracted some attention Of course if I can figure out a way to get paid to do it that would be nice but attention is more important Although this is subject to change and cancellation my current thought is that the next major versions would be 2 0 with the jouyou kanji taught in high school a total of 1130 additional glyphs and 3 0 with the jinmeiyou kanji the name only kanji 983 additional glyphs I don t know how I d break those up into minor versions but presumably I d aim for a similar spacing of about 100 new characters per release At the 30 point with a little over 3000 kanji the fonts should be basically complete in the sense of being usa
83. rnment regulations limiting employment work hours I don t know how much leisure time I will really have to devote to Tsukurimashou and I don t know to what extent I will be able to fit continued Tsukurimashou development into my future career plans I am now six years post PhD My job in Copenhagen is to be at most two years long and given that it took me a year and a half to find this job and next time if I am to have a real academic career at all then my next job had better be at the faculty level it appears that I should already have been searching for my next job since something like a year ago This is an extremely stressful state to be in and it is destructive to my ability to pursue side projects like Tsukurimashou I am conscious that my life outside of work is also far from what it should be it is not clear whether I will ever be able to have the life that I was promised and have earned and that too is destructive The bottom line is that I cant predict much about possible future development of Tsukurimashou Zar09l RELAT PSU 077BHZzZh Other similar projects Maybe you shouldn t use this package It is designed for specific purposes that are relevant to its designer and although I certainly hope others will find it useful my goals may or may not be in line with yours Also although I sometimes describe Tsukurimashou as the first parameterized METAFONT family with Japanese language cover age that claim requires careful
84. rsion and time Figure 1 Growth of glyph counts but not including code in other languages and locations also excluding mande mp but still including a few other files from Mandeubsida plotted against the total number of glyphs in the main Tsukurimashou family 19 lines of code 60000 50000 40000 30000 20000 10000 0 md 500 1000 1500 2000 2500 3000 glyphs Figure 2 Lines of code per glyph 20 RDITRDIND Relevant bugs in other software Building Tsukurimashou and especially its documentation requires the use of some fairly advanced features of third party software Some of those features are not often used as a result I ve become an unintentional beta tester and in some cases a maintainer for the third party packages I ve previously noted bugs when they come up in code comments and the relevant parts of this document but as of the current version such bugs have become so numerous that it makes sense to also have a central list METAFONT723 METAFONT etc 69 Metapost maybe even METAFONT issue propagated to METATYPET the equation solver has a non renewable resource of independent variable instance serial numbers which are consumed as code executes Basically one is used up permanently every time an assignment statement executes Very old versions of Metapost either did not have these or allowed the counter to wrap around and so the solver would produce incorrect results in
85. rticular the simplistic serif design and weighting in the Latin characters of Tsukurimashou Mincho and the limited customization of things like diacritical mark positioning in Latin characters not used by English I biased some marks notably ogonek and ha ek toward the styles appropriate for Czech and Polish as a nod to my own ancestors but I cannot read those languages myself and I cannot claim that these fonts will really look right for them I m not interested in spending a lot of time tweaking the Latin to be perfect because it s not really the point I already know how to read and write English Some other notes e Since my learning the kanji is a big part of the goal of this project labour saving approaches that would relieve me of hav ing to look at all the kanji individually myself for instance by feeding a pre existing database of kanji shapes into my exist ing general font technology are not appropriate to the original intention of the project As of version O9 August 2014 I m not happy with the weight of punctuation marks nor the sizes of serifs in the Mincho Latin character range Both these things will probably change in future versions with consequent changes in spacing I would like to include at least some support for vertical script but it is not a high priority One obstacle is that I don t have access to competent vertical typesetting software whether the font could support it or not Tsukurimashou i
86. ructural data extracted from Tsukruimashou GPLv3 Kleknev A coarse grained profiler for build systems such as the Tsukurimashou build This is still rather experimental It basi cally consists of a wrapper program that Make can invoke instead of bin sh which will when told to do so by a special environ ment variable collect a bunch of performance information that can be analysed by a separate reporting program to provide some insight into where all the time goes during a long build See my Web log article at http ansuz sooke bc ca entry 260 OCR Freely usable vector versions of the standard OCR A and OCR B fonts based on the work of Tor Lillqvist Richard B Wales and Norbert Schwarz License terms vary but the fonts should all be freely usable CZ aVBIHLATAISENN Building from a version control checkout 62 For the very latest public development version of Tsukurimashou and its associated tools you can check out the source code from version control Since the version control system stores only the source code as such not any files that can be automatically generated some additional steps are necessary to build from a checkout beyond those used to build from a distribution Building from a source control checkout requires some additional de pendencies not needed or not so strictly needed when building from a distribution package recent GNU Autotools including in particular Automake at least 113 the
87. s The actual kerning is done class to class using the class centroids rather than glyph to glyph That way we will end up with up to 40000 kern pairs instead of millions Openlype supports this kind of kerning pretty well There will be a feature file generated listing the contents of the classes and the distances for each pair of classes and that s much more efficient both in source and compiled form than specifying a distance for every pair of glyphs The number 200 up from 150 in earlier versions was chosen by educated trial and error Every glyph must be in a class for full kerning but no class can be too big or FontAnvil barfs That in turn limits the size of the average class to some maximum and so limits the number of classes to some minimum Using 200 classes seems to be enough for the current Mandeubsida fonts which are roughly 13000 glyphs each and it is reasonable to estimate that none of the fonts created in the planned scope of this project will have much more than that many glyphs To kern two contours together we can compute a closeness value for each row by saying if we positioned the margins of the glyphs this much apart how far apart would the contours be in this row That distance divided by a constant representing the optimal distance currently 230 font units and raised to a power representing how much extra weight to give to the closest points currently 3 represents 82 Figure 7 Closeness computation
88. s but in some cases the METATYPE1 pen stroke edge macros output does not well approximate the theoretical ideal shape that would be obtained by stroking the specified pen along the specified path A perfect result is not possible because the theoretical perfect curve is not in general a cubic spline curve and there are tricky topological considera tions in play too Some approximation is necessary and the one 22 chosen by METATYPE1 basically uses one control point in the out put envelope for each control point in the input path That may or may not be enough control points to produce a visually nice result It seems to especially often be a problem when there is a tight curve and not actually a sharp corner in the input path so curvature is great but not infinite or when there is an inflection point in the input path These cases are best simply avoided but Tsukurimashou s bundled version of METATYPE1 now contains added code that attempts to detect tight curves and inflection points and add extra control points automatically to reduce the likelihood of visual problems The detection rule the code uses was chosen by trial and error educated by dimensional analysis not by solid theoretical results and it will probably be incorrect in some cases The extra points and resulting topo logical weirdness necessitate FontForge postprocessing and may tend to trigger bugs in FontForge but FontForge postprocessing was already a dangerous necess
89. s classes in some renderers the resulting DVI file is so invalid as to cause the renderer to blank the rest of the page Reported to Will Robertson package maintainer he acknowledges it and says will fix by not requesting dot leaders when they re not wanted in next version A workaround involving poking into the package internals to set a smaller invalid spacing is implemented near the top of doc bkstyle tex PGF lower layer of TikZ shapes callouts library in version 2 10 only ellipse callouts just don t work causing fatal TEX errors if attempted apparently because a macro name was changed in the PGF development process and the change was not propagated to all files Workaround implemented in our build system consists of detecting version 2 10 and making a local modified copy of the buggy file with the macro name corrected The problem was apparently known to the maintainers and already fixed in their source code repository before I noticed it It presumably will not be an issue in any future releases but buggy 2 10 is the latest release and widely used as of late 2011 it will probably remain in the wild for a long time The ttx utility of TTX FontTools will emit an empty Format 2 cmap subtable containing no code mappings at alD if given an XML file that describes a Format 2 cmap subtable in which all the character codes are single byte codes Workaround is to include at least one double byte code Reported to their bug tracker Cht
90. s designed primarily for typesetting Japanese secondarily for English I have no immediate plans to support other Han script languages such as any dialect of Chinese nor put a lot of effort into tweaking the fine details of characters only intended for use in occasional foreign words Support of Korean is limited because of my limited knowledge of that language and learning it is not a high priority for me At this point the Mandeubsida fonts only support Korean hangul not hanja which are the Korean equivalent of kanji but just different enough that copying over the Tsukurimashou kanji would not be good enough I reserve the right to add features that I think are fun even if they are not useful Tsukurimashou is designed as a vector font assuming an output device with sufficient resolution to reproduce it In practice that probably means a high quality laser printer I have not spent time optimizing it for screens or low resolution printers and the hinting is automated 7 9 If it turns out to be too much work after all I might abandon the whole project Both building and use of Tsukurimashou require working around many bugs in third party packages some of which were men tioned here in earlier versions of this document The list has now grown so long it needs its own subsection which starts on page 21 of this document The Tsukurimashou fonts are distributed under the GNU General Public License version 3 with a
91. s that the user does or doesnt want based on knowledge built into select hl of what the different selection tags actually mean It is planned that in a future version the KANJIDIC2 file will be automatically translated to more Hamlog facts expressing which kanji are or arent included in the different grade levels then it will be possible to use options like enable chars kanji grade3 for finer grained selection of kanji In the case of a Prolog interpreter other than Hamlog there is some other code written in that interpreters own language to allow it to execute Hamlog programs and export something resembling this command line interface to the Makefiles Since Hamlog programs are also syntactically valid Prolog this support shouldn t be difficult in general See the swi ham pl file for what currently exists of this na ture The main advantage of using a non Hamlog interpreter is simply speed 78 eux Gave Kerning This is a summary of how the automated proportional spacing and kerning code works First the build system generates the future PS font as an OpenType file all complete except for widths and kerning It then calls bdf pe which sets all the bearings to 5O font units and makes a BDF format bitmap font scaled so that the reference kanji square 1000 font units takes up 100 pixels This script filters out glyphs that should not be subject to kerning which includes those with zero width mostly combining characters al
92. scriptions The glyphs look like this ANA TUS Official Euro Sign Fonts that include a symbol for the European currency unit Unicode U 20AC usually re draw it to match the style of the rest of the font and the Tsukurimashou fonts are no exceptions However there is an official glyph design that apparently was intended to be normative in theory you re supposed to use the official glyph which does not change 44 with the style of the rest of the font even if it clashes For those who want to do so the official Euro sign is included in the Tsukurimashou fonts at private use code point U F1710 It looks like this 45 INDY FILA Korean Language Support This user manual was built without the Korean fonts in particular Mandeubsida Dodum PS and as a result the section on Korean support cannot be typeset 46 vao mH vec Regarding Font Names First I will use the term font to refer to an OTF or similar file with a complete glyph set or as many glyphs as have been selected with configure This will generally be a few thousand glyphs for the Tsukurimashou family and about 13 thousand for the Mandeub sida family Because METATYPE is limited to producing Postscript font files with at most 256 glyphs each the system builds subfonts each corresponding to a page of the Unicode character set that is an aligned 256 code point block For instance the range U 0000 to U OOFF is the page containing ASCII an
93. siderably In a change from earlier versions the kern values are not otherwise rounded beyond being integers The table is written to a fragment of an Adobe fea file with subtable breaks on left class boundaries each time a subtable grows past 5OOO entries that means subtables actually end up a little over 5000 entries each That seems to be how big they can get without overflowing the OTF table size limits Bearings are written to a PE script The build system runs kern font pe which applies the output of the kerner program to the font Something else that kern font pe does is to add a hardcoded additional bearing of 40 on the left and 80 on the right to all Japanese script characters kana and kanji by trial and error this seems to make the results look better It seems to be simply a fact that Japanese characters need more space between them to look right than Latin characters do at the same type size Something similar should probably be done to Korean script characters but that has not been determined yet That is how the horizontal spacing of the font is currently computed It still isnt perfect but some progress has been made STA THILAGESATRLE 85
94. sidering but typically include all and none For example the specification mushroom purple would select purple mushrooms The specification fungi mushroom amp purple would select all fungi that are not purple mushrooms The specification mushroom red tomato would select all mushrooms that are not red as well as all tomatoes including red tomatoes tags are evaluated left to right and each one overrides the results of all previous tags The enable chars option uses the above syntax to describe which characters Unicode code points extended by internal private use al locations for some unencoded glyphs should be included in the fonts The default is all meaning all characters defined in the source code will be included in the fonts Other tags currently supported are none ascii latinl mesl and families of tags named like pagel27 unil234 and ul23ab with lowercase hexadecimal digits which corre spond to 256 character blocks of Unicode code points and to individual code points For example the specifier all pageO2 would include ev ery character it can except none in the range U 0200 U 02FF the specifier uni0041 would create very small fonts containing only the uppercase A from the ASCII range The default is all It is planned that in some future version this option should also be able to use KANJIDIC2 data to allow selection of kanji characters by criteria like school grade l
95. t through lt Z gt gt The choice of which ranges of numbers and letters to support is mostly not mine I just implemented what I found in the Unicode charts Unicode only has code points for the unvoiced versions of the enclosed katakana e g 99600 but not enclosed versions of 7j 7 and it has no code point for enclosed Y I ve added encoded to private use code point U F1711 but not the others If theres a demand for other enclosed characters though they are pretty easy to add I held off on just defining dozens or hundreds more because its not clear to me what the typical use of these characters actually is and a complete set of enclosed characters might be better seen as a font in itself rather than a series of special characters inside the font Note that for the enclosed katakana the substitution will accept either ASCII parentheses or wide parentheses U FFO8 and U FFO9 the others work with ASCII parentheses only LANS uE ET Limit PUA Code Points xpua This is not the usual kind of Openlype feature but is configured through the same interface as the others For software engineering purposes it is convenient to have a unique Unicode like code point for every glyph in the font family However some of those glyphs are not intended to be accessed directly by users 40 and more than one of them may correspond to the same Unicode character in ordinary text This feature removes the private use code points from many su
96. tionally spaced fonts to make this substitution work nicely When the ccmp feature is enabled as is default precomposed ac cented capitals which are taller than small cap plus combining accent combinations will be used wherever possible and the calt feature will only affect unusual combinations not covered by precomposed glyphs TUTO HACI Glyph De Composition ccmp This features general function is to join and split glyphs that can be thought of as combinations of other glyphs Exactly what that means depends on the script and the particular font Often the results of ccmp are needed as input for other features It is recommended that this feature should be turned on by default in fonts that have it and that is required by Microsoft s spec In all fonts this feature will substitute precomposed accented Latin letter glyphs for combinations of letter plus combining accent wher ever precomposed glyphs exist The precomposed glyphs generally look better because the accent in the precomposed glyph may change shape or size to work better with its base letter this issue is espe cially significant in the case of monospace fonts and capital letters because the generic accent would usually be too low and is not subject 34 to GPOS positioning When no precomposed glyph exists rendering will fall back on the unaccented letter and a generic accent with the capital letter issue in monospace fonts partially mitigated by the
97. tps sourceforge net p fonttools bugs 53 January 2014 HarfBuzz hangul shaper deliberately does not handle the case of multiple conjoining jamo in the same category that could better be expressed as a single cluster for instance the case of U 1100 U 1100 kiyeok kiyeok instead of U I101 ssangkiyeok These two sequences are not canonically equivalent in Unicode and apparently the longer one is forbidden by the Korean stan dard KS X 1026 1 nonetheless Mandeubsidas GSUB features were intended to handle them both nicely Worked around this and various other issues some of which were probably bugs in Mandeubsida by redesigning the GSUB features January 2014 29 Thanks to Dohyun Kim and Jonathan Kew for discussion of these points 30 MEYELESIO IRA Using Tsukurimashou Fonts appear in Openlype format in the otf subdirectory of the package If you just unpack a distribution there will be four fonts there corresponding to the monospace and proportional versions of Tsukurimashou Kaku and Mincho If you run the build process that will create more For a quick start all you need to do is install the font files in whatever way is standard for installing OpenType fonts on your system It is safe to delete everything else in the package though you might want to keep the PDF documentation files Samples of what the different styles look like are in the file demo pdf which see Heres a brief summary O EJLA Tsukurimashou Ka
98. ude the standard all and none and the short name tokens and Hamlog atoms for all the families styles weights and spacings currently supported The token proofs may also be used as a shortcut for enabling all the styles needed to generate proof documents See Section for more about the naming of fonts and note that these are the pre rewriting names so that for instance Anbiruteki might be selected via tsuku amp mg amp eq The enable parasites option allows this kind of include exclude selec tion for parasite packages see next subsection The supported tokens are all none and the names of the parasites currently beikaitoru fontanvil genjimon icemap idsgrep kleknev and ocr The default parasite setting is auto note not a supported token itself this enables a special bootstrapping mode in which the configure script will conditionally enable fontanvil and icemap as necessary for the local configuration Normally that means FontAnvil will be built if there is no FontForge or FontAnvil already on the system Icemap is not required for building from a source distribution but will be built in a version control checkout situation Mostly In the case of white on black reversed glyphs and some fractions precomposed by FontAnvil instead of by METATYPE1 you must include all the parts that FontAnvil will assemble in order to get the combined glyph made by assembling thos
99. ue that it shares a name with a widely used programmer s IDE so it is not safe for the configure script to actually execute an executable called eclipse if it finds one If something that might be ECL PS CLP is detected the configure script puts up a warning and the user is free to enable it explicitly The rest of this subsection can and probably should be skipped by anyone who isnt both a Perl and a Prolog hacker Still with me The way Hamlog works is sort of fun and so I m going to spend a few pages describing it for those who are interested if only as an example of something you should Not Try At Home The idea is to use regular expressions for the operation of finding a clause whose head matches the current goal Hamlog reads its program into a Perl hash where the key is the functor and arity like foo 3 and the value is a newline separated pile of clauses in more or less plain text When it tries to satisfy a goal it takes the goal which starts out as plain text and converts it to a regular expression that will match any appropriate lines in the database Variables in the goal turn into wildcard regular expressions ground terms turn into literal text and then when theres a match parenthesized sub expressions are used to extract the body of that clause 73 It is because of the use of regular expressions that Hamlog doesnt do compound terms and in turn is likely not Turing complete though I haven t thought caref
100. ular kanji in atoms but refrained because of the desire for Hamlog code to be easily readable by nearly all Prolog interpreters 75 I tried to keep the number of built in predicates to an absolute minimum partly because any that are not standard Prolog have to be re implemented in Prolog and probably once for every supported Prolog to build the shell that executes Hamlog programs on a Prolog interpreter Heres an exhaustive list e cut This is implemented by string substitution as well when a clause body gets added to the to satisfy stack there s an additional regular expression substitution pass that converts any instances of O in the body into 1 where the argument is an integer identifying the current height of the choice point stack If at any point in the future we attempt to satisfy a 1 goal then the stack gets popped back to that point discarding any choice points created between the time the got its argument and the present time For this reason should not be used as an atom or functor for any other purposes than as the cut even if to do so would otherwise be valid Prolog fail which causes immediate backtrack useful in conjunction with cut to implement negation true which is not actually used so maybe I should delete it var true if the argument is a variable not yet bound to an atom This is important in Hamlog because many predicates need to be written to accept more than one instantiatio
101. ully through all the possibilities of using recursive predicates or atom splitting to build stack like data structures As all the world knows it is impossible to write a regular expression to match balanced parentheses Current versions of Perl actually bend the theory with experimental extensions that do allow the matching of balanced parentheses so that in a certain important sense Perl regular expressions are not regular expressions anymore at all but even I am not quite twisted enough to actually deploy such code Things in Hamlog that look like compound terms such as the sub goals in a clause body are handled as special cases but the point is that arguments to a functor that will be used as a goal have to be either atoms or variables This also means Hamlog doesn t do Prolog syntactic sugar things that expand to compound terms such as square brackets for lists Once theres a match it does string substitution on the matching head the current partially completed goal and the body to get a new modified body for the clause taking into account any variables assigned by the head match The new clause body gets substituted into the current partially completed goal which is a string as a replacement for the head that just matched So the partially completed goal is a sort of stack of comma separated heads that grows from right to left and implicitly contains all the assigned variables Because of the simplistic way variables are given the
102. valid code one could run but the output of livecode is intended to have the property 67 that its checksum will almost always change when changes in the input necessitate recompilation and will seldom change otherwise Therefore the build system can use the checksum of livecode s output for fine grained change detection to skip recompiling some files that under GNU Makes timestamp based scheme would otherwise appear to require recompilation As a special feature intended for maintainer use only livecode also supports a find dead code mode If its first command line argument is d then instead of its usual output it will print a list of names of macros it thinks are not invoked Running the dead code check against the entire code base may be a way to find superfluous METATYPET infrastructure that could be removed make ass generate the assemble font pe script which in turn does most of the work of assembling 256 glyph subfonts to create full coverage OTFs make book make decisions on how to split the multi thousand page proof document into smaller books The output is a set of TEX files which then get processed further to generate the actual PDFs make cfghl translate configure command line strings into Hamlog for inclusion in config hl which then feeds back to control the build system make cover generate colour covers for the proof books make eids scan the proof files and generate an EIDS format dictionary
103. ved from the Hamlog name after rewriting long names are generally mixed case ASCII and the elements are separated by spaces and may also contain internal spaces As with short names normal weight and monospace translate to nothing and are removed when forming the long name An example long name might be Tsukurimashou Tenshi no Kami PS note that that would correspond to the short name tsuku mg el ps as if Tsukuri mashou Maru Extra Light PS because of rewriting There are also some places notably in the filenames of final OTF fonts where a long name is used with all the spaces and hyphens removed 50 There are also JK names which are constructed just like long names but with elements written in Japanese or Korean and thus generally in non ASCII Unicode characters For Mandeubsida fonts these should be in Korean but the occasional Japanese or English word may appear because of a yet unimplemented translation feature and have spaces between the elements otherwise JK names are in Japanese and have no spaces he main use of JK names is for populating the font metadata tables An example Japanese name might be fEUZ Ux XfEof PS corresponding to Tsukurimashou Tenshi no Kami PS Since there is some information loss in transliteration to English and many of the names come from Japanese and Korean words anyway it may be best to think of the JK name as the truest name of the font for human purposes even though fro
104. verage font but as of 2012 it doesn t appear anyone has done that Availability and licensing terms are unclear no released code or fonts seem to be available but there have been papers published about it http www hang yph com 9 IPA Mincho and IPA Gothic fonts 2003 onward Note IPA in this case stands for Information technology Promotion Agency not the International Phonetic Alphabet and these fonts do not cover that IPA Free high quality fonts for Japanese Truelype 15 format not parameterized http ossipedia ipa go jp ipafont index html X3lEX SIL International 2004 onward TEX engine extended to handle Unicode and modern font technologies Used to compile this document and one of the main compatibility targets for Tsukurimashou http scripts sil org xetex H ng Zi project by Javier R Laguna Aimed to be a parame terized METAFONT family for Chinese The last release which was in 2006 contained 125 characters No infrastructure for addressing issues like METAFONT s 256 glyph limit or radicals changing shape depending on their context Probably abandoned However it did make several releases of code that you can still download and compile http hongzi osdn net The KanjiVG Project coordinated by Ulrich Apel current in 2011 Still under development but already has basically complete cov erage and is deployed in several important applications This is not a font family but a
105. versions of FontAnviD to have much hope of building fonts that can pass fontlint Running make check at the top will also recursively run make check on any parasite packages that have been enabled and the parasite package test suites in particular the one for IDSgrep may be of more practical use than the main Tsukurimashou fontlint tests If you try to do make distcheck it will fail if any of the tests fail and the tests probably will fail because of the bugs in FontAnvil spline geometry A workaround used in my Tsukurimashou 0 8 distribution packages is to override the options for fontlint pe as follows make DISTCHECK CONFIGURE FLAGS FONILINT FLAGS w 2 3 5 23 distcheck The DISICHECK CONFIGURE FLAGS variable specifies additional flags ac tually a command line fragment that make distcheck will pass to 58 configure Then inside that the FONILINI FLAGS variable specifies flags to pass to the font integrity checker The codes 2 3 5 and 23 are the errors that I wasn t able to correct with the version of FontForge installed on my development system as of the time of the release Dont bother with disable dependency tracking that is an Autotools thing meant for much larger and more softwary packages It applies only to code in languages supported by Autotools which at present means only the C kerning program whose dependency is trivial The dependency tracking for METATYPE1 code is completely separate
106. width as monospaced Latin characters Otherwise with wide input or more than one digit in the numerator or the denominator the fraction will be the width of a wide ideographic character J121 gt als abd A E E 134 56l gt 2l 34 56l 5 amp l As of version 0 5 this feature works in both the monospace and proportional fonts However in order to make it work in proportional fonts it was necessary to exclude the glyphs involved from the kerning table as a result only precomposed fractions will benefit from kerning Fe TRAE JULIET Capitals to Small Caps c2sc 33 This feature replaces capital letters with small caps as might be desired for typesetting all caps abbreviations like TEPCO and IAEA It works much like the smcp feature Similar limitations apply ARIAS TE Contextual Alternates calt The calt feature in general replaces glyphs with other glyphs that look better in certain contexts In the Tsukurimashou family it is used only by monospace fonts to replace Latin capital letters with small caps when they are followed by combining accents The accents which do not shift vertically would otherwise usually collide with the upper part of the letters Accents in proportionally spaced fonts are positioned using the GPOS table and will shift vertically when applied to capital letters Small caps in monospace Tsukurimashou fonts are the same height as lowercase letters a little shorter than small caps in propor
107. xtension Building a subfont involves creating a temporary driver file that includes all the relevant source files for that font s short name For instance to build tsuku mi ps 00 pfb the build system would create and execute a driver file that loads tsuku mi mp tsuku ps mp and tsuku OO mp in that order Some of those files might themselves load other files outside this naming scheme Almost all file inclusion is done via conditionals in METAFONT to prevent any file from being included more than once but double inclusion should also be both avoided by design and usually harmless should it occur by accident The file inclusion macros in preintro mp also support a queue of late includes which will be processed after the others to allow for instance code associated with a style to override code associated with a weight In the case of the family short name token not being tsuku the build system will use source files that match the subfont s family token where those exist and default to tsuku when such files don t exist For instance to build bll co 01 pfb Blackletter Lolita Cosette page OD it might load bll co mp which exists but then load tsuku 01 mp because bll Ol mp doesn t exist Thus the tsukurimashou family can be thought of as a generic ancestor from which any others inherit Fonts also have long names which are created by a similar transla tion process Long names are always deri

Download Pdf Manuals

Related Search

Related Contents

建築・設備工事検査実施要領      [输入书名]  pdf 0.74M - 中央大学 理工学部 経営システム工学科  Vulnérabilité des bâtiments  Philips HP4885 Hairdryer  Lavante-séchante Waschvollautomat Waschmachine  SERVICE MANUAL - MS  

Copyright © All rights reserved. Failed to retrieve file