glyf (i.e. contour) analysis reports on libre fonts.

Hi,

People who went to LGM 2016 London in April this year and heard my talk would hear that, IMHO, the most unique and valuable part of the Font Validator is its analysis on individual glyphs. That is broadly divided into the glyf table test on the glyph contours, and the rasterization test on the truetype hinting instructions. The latter wasn't open-sourced by Microsoft, but was re-implemented, much faster with broader scope and wider supported platforms, and superceding the old version in 2.0 a few weeks ago.

So I have turned my attention to the glyf test, and threw all my (many buggy) fonts to try to break it. When the glyf test itself breaks and cannot handle the situation, it is called "A1112 exception", and I filed them under https://github.com/Microsoft/Font-Validator/issues in the last few days.

Out of that activity, I collected the reports of the ~3000 fonts on Fedora (about ~2000 under /usr/share/fonts, and another ~1000 texlive). About half of them, (~1300 /~2000 for general libre fonts, and ~200 / ~1000 for texlive) have some level of glyph errors or warnings.

At the rate of 2/3, almost all libre font authors should have a look at their own reports and fix their fonts.

fedora-glyf-EW-reports-file-index-2016-08-11.txt is the file index and "fedora-glyf-EW-reports-2016-08-11.tgz" is the collection of reports under
http://htl10.users.sourceforge.net/tmp/FontVal-test-results-2016July/

The file index entries look like this, sorted by the "FileNameAndPath" field (i.e. this is the first entry, with "aa..."):

tmp6b009ab7.tmp.report.xml:<FontFile FileName="Comfortaa-Bold.ttf" FileNameAndPath="/usr/share/fonts/aajohan-comfortaa/Comfortaa-Bold.ttf">

and author of "/usr/share/fonts/aajohan-comfortaa/Comfortaa-Bold.ttf" should extract "tmp6b009ab7.tmp.report.xml" from the tar ball, etc.

Disclaimer: I am just the messager - I just ran the glyf test on all the fonts I have, and played no part in writing the test itself. People disagreeing or have questions with the reports' content should file upstream under  https://github.com/Microsoft/Font-Validator/issues .

People who want to make the glyf test even better can look at and help fixing the A1112 exception's - I honestly do not mind somebody else doing it! :-). People who want me to make it better, please feel free to make a donation ( https://sourceforge.net/p/hp-pxl-jetready/donate/ ).

Hin-Tak

P.S. It has been exactly a year ago today (baring time-zone arguments...), that I received and first saw the code bundle from Microsoft to get fixed up, to be publicly released a few months later. Hurray.
«1

Comments

  • Simon CozensSimon Cozens Posts: 195
    edited August 2016
    I'm not sure what the nature of the glyf test is, but is this really the fault of the fonts? What I mean is, I assume these designers are using FontForge or whatever to produce fonts, and regardless of what the designer draws on the screen, the font editor should emit a compliant font. If you're seeing malformed output, isn't that a bug in the font editor? And are there ways to isolate and fix the problem at source? (This is not a pointed question - maybe the individual fonts are doing something really funky. I don't know. But I tend to think that computer programs should produce well-formed output even with funky input.)
  • Even fonts made with professional font editors contain glyph related issues. Sometimes a designer didn't notice them, or decided such issues are either acceptable or intended.

    Is a contour made out of one or two points always an error which need to be fixed? Do you allow for local off-curve extremes? How about composite glyphs with overlapping contours?

    Calibri contains 157 unused glyphs, and 40 glyphs have contours with redundant points. It also contains 2 lookups which are not accessible.

    San Francisco contains 93 unused glyphs, 17 glyphs with off-curve extreme points. It has invalid PANOSE.
  • In the case of intersecting components of composite glyphs (W1110), recommendations seem to be contradictory. Microsoft explicitly allows them in their Hinting and Production Guidelines:
     
    "Composite glyphs should be used whenever a glyph image is identical to another. [...] Overlapping composite glyphs can be used." [1]
     
    [1] Hinting and Production Guidelines, Section 3.2.1,
    https://www.microsoft.com/typography/ProductionGuidelines.mspx#glyph
  • Hin-Tak LeungHin-Tak Leung Posts: 245
    edited August 2016
    I am just the messager - nobody (including me) has touched that part of the validator code since microsoft open-sourced it. My role here is simply that I wanted to see when that particular part crashes and bails :). (the A1112 exceptions). 

    Incidentally, I filed an issue about the used error and warning status not properly documented and explained some months ago when I was looking at the help file, so the full list of issues that the glyf test can flag, collected from the code itself, is listed in:

    https://github.com/Microsoft/Font-Validator/issues/15

    You can see the whole list of what is considered a problem in the glyph contours, and what should be done about them, towards the end, in the updated and merged help file with bits recovered from the old help file. I am just the messager, as I should repeat. I don't decide what is and what is not an error...
  • On the role of font editors being the "gate keeper" of font errors - that's a bit like asking the compiler to check your programming code for you, or asking your word processing software to check your grammar.

    Some of it is possible - e.g. word processor high-lights possibly spelling mistakes - but it is not possible nor convenient to check grammar on every 'save'. But there is a a place for dedicated and detailed external analysis. That's why you get proof-readers, software that analyses articles for plagiarism, etc.

    FWIW, the glyf test for the typical CJK font takes about an hour and half. It is certainly quite impractical for any font editors to do that every time per load or per save.

  • Even fonts made with professional font editors contain glyph related issues. Sometimes a designer didn't notice them, or decided such issues are either acceptable or intended.

    Is a contour made out of one or two points always an error which need to be fixed? Do you allow for local off-curve extremes? How about composite glyphs with overlapping contours?

    Calibri contains 157 unused glyphs, and 40 glyphs have contours with redundant points. It also contains 2 lookups which are not accessible.

    San Francisco contains 93 unused glyphs, 17 glyphs with off-curve extreme points. It has invalid PANOSE.
    No.
    In the case of intersecting components of composite glyphs (W1110), recommendations seem to be contradictory. Microsoft explicitly allows them in their Hinting and Production Guidelines:
     
    "Composite glyphs should be used whenever a glyph image is identical to another. [...] Overlapping composite glyphs can be used." [1]
     
    [1] Hinting and Production Guidelines, Section 3.2.1,
    https://www.microsoft.com/typography/ProductionGuidelines.mspx#glyph
    No.
    Here is microsoft's recommendation:

    https://github.com/HinTak/Font-Validator/blob/master/NewHelp/W1110.htm

    In the same directory there are others for the other error codes.

    These files make the chm file:

    https://github.com/HinTak/Font-Validator/blob/master/bin/FontValidatorHelp.chm

    At least for W1110, I think it is written quite clearly.

    I haven't read the others, and I really wish people read the bundled documentation - the chm file - before saying random things.


  • Why don't font-editors fix the errors themselves?  Like removing off-curve extremities?  And removing overlap from a glyph before writing it to a file?  Editing a font with overlapping parts is so much easier.
  • Mark SimonsonMark Simonson Posts: 788
    edited November 2016
    From the warning message regarding overlapping composites (W1110) in Font Validator:

    "Although Microsoft discourages this type of glyph design, for legacy reasons this overlap is allowed for composite glyphs."

    I really wonder about this statement. It seems to be saying that, for legacy reasons, overlap is allowed, which doesn't make much sense when you consider the statement just before it:

    "This type of glyph design introduces some ambiguity in terms of defining what is the 'inside' and 'outside' of a glyph shape, and may cause a rasterizer to reverse the overlapped portion. This problem is generally seen on older printers."

    It seems like maybe the comma was misplaced and it was meant to go like this:

    "Although Microsoft discourages this type of glyph design for legacy reasons, this overlap is allowed for composite glyphs."

    If you follow the link to Microsoft's TrueType spec referenced in the warning message, you'll find this:

    "If set, the components of the compound glyph overlap. Use of this flag is not required in OpenType — that is, it is valid to have components overlap without having this flag set. It may affect behaviors in some platforms, however. (See Apple’s specification for details regarding behavior in Apple platforms.)"
  • Paul van der LaanPaul van der Laan Posts: 171
    edited November 2016
    @Kristof Bastiaensen: the last thing I need is a font editor that automatically "fixes" things because of some spec. A validator that creates an analysis report is of much greater value.

    Truth is that there are many different rasterizers and font shaping engines out there that do not follow the latest specs, and sometimes we have to make fonts that need to work for those environments.

    Point in case: the very latest OT spec that includes variable fonts *does* allow overlapping contours.
  • Erwin DenissenErwin Denissen Posts: 71
    edited November 2016
    Even fonts made with professional font editors contain glyph related issues. Sometimes a designer didn't notice them, or decided such issues are either acceptable or intended.

    Is a contour made out of one or two points always an error which need to be fixed? Do you allow for local off-curve extremes? How about composite glyphs with overlapping contours?

    Calibri contains 157 unused glyphs, and 40 glyphs have contours with redundant points. It also contains 2 lookups which are not accessible.

    San Francisco contains 93 unused glyphs, 17 glyphs with off-curve extreme points. It has invalid PANOSE.
    No.
    My initial post had several questions and some examples of fonts with issues, so I wonder what you mean with your No.

    I just took another look at SFNSText-Regular, and besides the issues I mentioned earlier, it also has corrupt feature params concerning the character variants. If they are valid, let me know, as then I need to solve a bug in FontCreator!

    @Kristof Bastiaensen: the last thing I need is a font editor that automatically "fixes" things because of some spec. A validator that creates an analysis report is of much greater value.

    Truth is that there are many different rasterizers and font shaping engines out there that do not follow the latest specs, and sometimes we have to make fonts that need to work for those environments.

    Point in case: the very latest OT spec that includes variable fonts *does* allow overlapping contours.

    Like many font editors, FontCreator has its own font validation features. Most outline related issues can be automatically fixed, but this is optional.
  • @Kristof Bastiaensen: the last thing I need is a font editor that automatically "fixes" things because of some spec. A validator that creates an analysis report is of much greater value.
    Depends on the meaning of Fixing I guess.  There is nothing flawed in a design that doesn't have control points at extremities, and has overlapping paths.  Letting the computer do that work frees the designer from work, and gives him more freedom.
  • Depends on the meaning of Fixing I guess.  There is nothing flawed in a design that doesn't have control points at extremities, and has overlapping paths.  Letting the computer do that work frees the designer from work, and gives him more freedom.
    Depend on the meaning of Freedom I guess. To me that means that my tools do not get in my way, and that I will have full control over the result.

    If you’ve ever designed an italic you might have had to draw tiny shapes that can be defined much better without extreme points. Any attempt to add those kinds of points will actually lead to a worse result.
  • edited November 2016
    I can follow your point about having tight control over extreme points.  But how does removing overlap get in the way?
    EDIT: my program will give control over the approximation.  You can set a minimum size of curves, and choose for tight approximation with many curves, or loose approximation with few curves.  This should avoid problems like you mentioned.
  • A tool to add extreme points can be helpful but it should be optional. The OT spec does not dictate that outlines must have extreme points.

    Remove overlap is indeed helpful but should also be optional, in case you want to generate a variable font where they are needed to make interpolations possible.
  • @Paul van der Laan, my idea is to remove overlap when the target is an opentype font file, except when the glyphs are used for font variations.  I might add an option for signalling an error when the source contains overlap.
  • I am trying very hard not to engage in any specifics here - the basic "smoke test" of 6000 fonts: the rasterization part takes 9 hours; the glyf table test was about 9 or 10 days. Even just setting the tests going (and ignoring/not-reading the outcome) is a lot of time. I would not have any time to do any real work ( or even have a life) if I try to answer any question from any of the authors of even just the 6000 fonts I test things on.

    So a few general points:

    - the bundled CHM file is auto-generated from the actual error code status list; it is actually quite difficult to add a test or add a few tests (as I did with CFF and SVG) without filling in a corresponding help section, because the compiler complains. So read the CHM file first, as it is written by the same people/person who wrote the test. Hopefully he/she explained why some issue is considered troublesome and warrant a warning. I am quite happy to include additional paragraphs from people if people feel like they can explaining issues better.

    - on W1110, I thought it was fairly clearly written - there is "what the spec allows", and there is "older legacy systems (printers)".  If you think your font will never ever be used for printing, or on older systems, then obviously you could ignore it. I don't really understand why the issue needs dragging on and out. But remember 'web font' does not mean "screen-rendering only" as the user's printer needs to be able to cope too, if he/she decides to print from the Web browser.

    - farming off issues to font engines or font editors seems irresponsible. Yes, font engines can cope with a lot of weird shit; and yes, many font editors can auto-correct or warn on issues.

    But - font engines (both on screen in the major ones, and printers) are not created all equal, or up to spec. There are part of the spec which are well-written and therefore well-implemented by most rendering engines; there are part which are ambiguous, and people have different opinions - for this a "font checker" should warn ; and there are undocumented behaviors (where different font engines are free to behave differently). yes, if you create a font that is meant to work with one platform - e.g. Windows, and *not* ever print from windows, then you can ignore a lot of that too.

    Relying on Font editor to auto-correct /warn is quite impractical - a font editor, particularly the interactive kind, need to have a guaranteed response time - it must save your work fully within say, 2 minutes after you press the "save" button; while it can and should check for common problems, a dedicated "font checker" that will spend as much time and do as thorough a report on any possible issues, without needing to worry about responsiveness, and in turn, you only run once in a while and let it takes as long as it will take to do a check, is a good thing.

    Out of the 300 fonts that windows ships, about 10% took longer than 20 minutes to run the full check. The longest is 7 hours with one of the CJK fonts. There are small changes e.g. In prep or cvt which affects every glyph - yet you do not want to wait 7 hours for your font editor to check, every time you touch either prep or cvt. so you run a check only when you think is needed.

    That's a good thing to separate the role of font editor vs the role of font checker - one optimises for responsiveness, the other for thoroughness.
  • On printing/printers, the cheaper ones might do better if you specialises in making shitty web fonts. The drivers for cheap printers basically do hires screen-dumps, while the more expensive printers take the fonts from the OS as they are.

    So if your font is crap and happens to work okay on screen, you get better result in cheap printers. By "better" I mean "closer to screen display".
  • My proposition is not to automatically fix flaws in the fonts, but to free the font creator from design limitations which are there only to satisfy the renderer.  It's a duplication of effort, and the time spend can be used better elsewhere.  If the program takes more than two minutes to write the font to file, it should update its algorithms (on a modern machine).
  • In Glyphs, overlapping components are decomposed and overlaps are removed by default. But there are options to disable this. For variable fonts, decomposition and removing overlap is disabled. 
  • Hin-Tak, I suggest changing the string to "Although Microsoft discourages this type of glyph design for legacy reasons, this overlap is allowed for composite glyphs." in your version :)
  • John HudsonJohn Hudson Posts: 1,092
    All change for variable fonts: overlapping outlines are presumed to be typical for both TTF and CFF2 fonts using variations technology. And yes, this means the Adobe rasterisers need to be updated to handle this (I believe they're the 'legacy reasons' in question).
  • Hin-Tak, I suggest changing the string to "Although Microsoft discourages this type of glyph design for legacy reasons, this overlap is allowed for composite glyphs." in your version :)
    I think arguing for the position of a comma is possibly bikeshed'ing. Besides, the meaning changed: I read the current wording as "Microsoft discourages it (now), in the past it was allowed and therefore is allowed", versus your suggestion "Microsoft discourages it in the past, it is allowed (now)". The meaning is completely different.

    This is the sort of changes I'd prefer people filing the issue upstream and let the Microsoft folks explain what they mean by those words. Sorry.
  • My proposition is not to automatically fix flaws in the fonts, but to free the font creator from design limitations which are there only to satisfy the renderer.  It's a duplication of effort, and the time spend can be used better elsewhere.  If the program takes more than two minutes to write the font to file, it should ..update its algorithms (on a modern machine).
    This seems to be a mis-conception - you seem to assume there is a one-and-only-renderer to satisfy. Printers have their own renderers. Even if you restrict to windows and never to print, there are different generations of microsoft windows.
  • The font validator warnings were supposedly written around 2000, but newer Microsoft fonts like Calibri in fact do use overlapping components, so probably this warning can safely be ignored now.

    I think I wrote about this, but it hasn't registered in some people's mind: Microsoft substantially refactored/rewritten the glyf test between 2003 (the last public binary) and 2009 (the opened code). So this part of FontVal can be as "young" as 2009.

    I am fairly tired of people wanting to weaken tests. The way I see it, font checks should only get stricter, more warnings and more errors, as there are more old/odd/different, and different versions of renderers over time, and more problems to trip over. It should rarely if ever get weaker. If you just want some software to "okay" rubber-stamp your font, maybe just writing "my font is okay" on a piece of paper and put it up on your own wall is easier.

     If you think a warning/error message does not apply to  you, and you do not want to support old/odd renderers, please ignore it. No need to tell me.

    Many practices are "allowed" (i.e. tolerated), but not encouraged. I think that was the meaning of Microsoft's message. And between "allowed" (tolerated) vs "encouraged", there is room for a warning message.
  • @Hin-Tak Leung, are you saying there are renderers where having overlap is not only allowed, but desirable? I am not talking about variable fonts, because thats different technology.
  • The font validator warnings were supposedly written around 2000, but newer Microsoft fonts like Calibri in fact do use overlapping components, so probably this warning can safely be ignored now.

    I think I wrote about this, but it hasn't registered in some people's mind: Microsoft substantially refactored/rewritten the glyf test between 2003 (the last public binary) and 2009 (the opened code). So this part of FontVal can be as "young" as 2009.
    It is already in the old public FontVal version, so it cannot be from 2009.
    I am fairly tired of people wanting to weaken tests. The way I see it, font checks should only get stricter, more warnings and more errors, as there are more old/odd/different, and different versions of renderers over time, and more problems to trip over. It should rarely if ever get weaker. If you just want some software to "okay" rubber-stamp your font, maybe just writing "my font is okay" on a piece of paper and put it up on your own wall is easier.

     If you think a warning/error message does not apply to  you, and you do not want to support old/odd renderers, please ignore it. No need to tell me.

    Many practices are "allowed" (i.e. tolerated), but not encouraged. I think that was the meaning of Microsoft's message. And between "allowed" (tolerated) vs "encouraged", there is room for a warning message.
    The discussion is not about weakening tests, but finding out what the intent of the wording actually was.

    Especially when dealing with font size optimizations (web fonts), using overlapping components can be a huge gain. So I think it is fair to question the intentions of this particular check.
  • @Hin-Tak Leung, are you saying there are renderers where having overlap is not only allowed, but desirable? I am not talking about variable fonts, because thats different technology.
    I am saying neither it is only tolerated. That's what 'for legacy reason ... allowed' means.
  • Hin-Tak LeungHin-Tak Leung Posts: 245
    edited December 2016
    It is already in the old public FontVal version, so it cannot be from 2009.
    You still fail to read - I wrote "refactored/rewritten". The current glyf test code can be as young as 2009 because it was rewritten. Why are we even debating meanings of words like "rewritten"? Does "rewritten" means different things to different people? Sigh.
  • It is already in the old public FontVal version, so it cannot be from 2009.
    You still fail to read - I wrote "refactored/rewritten". The current glyf test code can be as young as 2009 because it was rewritten. Why are we even debating meanings of words like "rewritten"? Does "rewritten" means different things to different people? Sigh.
    I’m not talking about the glyf test code. The wording of the warning message W1110 is exactly the same in the help file from 2003 and 2016. So logic dictates that it was written in or before 2003.
Sign In or Register to comment.