[XeTeX] misplaced combining diacritical marks 2

[Alexander Schultheiß]

I forgot the attachments :(

2010/9/1 Alexander Schultheiß <aschulth at googlemail.com>:
> Hey,
>
> I'm starting this second thread because I'm unable to figure out how
> to reply to a single message within a digest :(. I turned digest mode
> off by now but decided to open a second thread in order to reply to
> everybody in one mail. Thanks for all the help.
>
> @ Ross Moore
>
>> Please show the complete preamble for the coding that you are using.
>> We need to see what packages you use, and how the font is loaded.
>
> The preamble I' using is:
>
> ----------------------------------------------------------------------
> \documentclass[a4paper,12pt]{article}
> \usepackage{fontspec}
> \usepackage{xltxtra}
> \usepackage{xunicode}
>
> \fontspec{Junicode}
> \setmainfont{Junicode}
> ----------------------------------------------------------------------
>
> I don't think it's a problem with the preamble, though. This is the
> same preamble David suggested except for the command
> \defaultfontfeatures{Mapping=tex-text} which I haven't included
> because I don't know what it does and haven't really bothered yet to
> find out. But even with this command included the problem persists.
>
>> Also provide a screenshot of what you see.
>> For example, the attached image shows what various fonts
>> (including styles of Junicode) should produce.
>
> I have attached an image (see below).
>
> @ Fr. Michael Gilmary
>
> Thank you for your suggestion. I have to admit I don't understand what
> the code really does. Also, I'm not merely trying to get a certain
> glyph working and I'm hesitant to define every combination in latex
> and kern them by hand as it should be latex's job to do this. Even
> more so, if anchors are provided and the program doesn't have to
> guess. I keep it in mind, however, as a last resort.
>
> @ David J. Perry
>
>> I happen to have the Unicode values for the diacritics memorized, so
>> I just typed them in
>> [...]
>> la\char"0304 \char"0323 m
>> hy\char"0304\char"0301 s
>
> I tried it and it works somehow but the results are not as good as if
> I were using the standard latex commands \d{} &c. I've attached an
> image where you can see in line 4 that neither the macron nor the
> acute is placed correctly. I think this due to the fact that xelatex
> just puts the glyphs one after the other and since comb.diacritical
> marks have negative witdh they roughly happen to be above the a.
>
> @ Khaled Hosny
>
>> In an ideal world, fonts should have proper GPOS anchors for
>> positioning arbitrary combining marks, and sometimes this is only
>> way to get proper output as there isn't a pre-composed form (and, as
>> a matter of policy, Unicode is not adding any new pre-composed
>> forms), but almost all fonts fail on that, fortunately TeX can still
>> provide workarounds for that.
>
> My suspicion is, that xelatex does not honor anchors at least when it
> comes to latin script. Rather it relies on pre-composed glyphs and
> the combining diacritical marks.
>
> 1. Observation:
>
> As noted before, if there are no pre-composed latin glyphs like
> a+acute at the apporpriate unicode codepoints xelatex prints a blank,
> which is confirmed if you look at the second image I've attached,
> where I deleted the a+macron glyph from the Junicode font.
>
> 2. Observation:
>
> Oder of commands does play an important role. If you look at line 3 of
> the second image, you'll find that there is at least an a+macron as
> opposed to lines 1 & 2. The commands used in line 1 and 2 look for the
> unicode codepoint whereas the commands in line 3 try to assemble the
> glyph based on the combining diacrtical marks, it seems.
>
> 3. Observation:
>
> Xelatex does not honor mark-to-mark anchors in latin script. This
> seems to be confirmed by the glyphs a+macron+acute in line 3 and 4
> where, according to Obs. 2, the combining diacritical marks are used
> but not stacked correctly as they should (Junicode has mark-to-mark
> anchors. The macron has an "Anchor-3 base" and the acute has the
> corresponding "Anchor-3 mark").
>
> Summary
>
> If there is no pre-composed glyph xelatex _never_ assembles the glyph
> correctly. Look at image 1 for example. a+acute and a+macron in line 1
> are correct because pre-composed glyphs do exist. The a+macron+acute
> does not exist in unicode, yet the distance between the macron and the
> acute is too big. According to the mark-to-mark anchors within
> Junicode the distance between the acute and the macron should be 109
> (2048 em-size) while the distance between the macron and the a is 234,
> more than double. This confirmes obs. 3 above. In line 2 the distance
> between the acute and the macron is even bigger. How can this be?
>
> In addition in line 3 the acutes are off to the left a bit. While,
> interestingly, similar is true for the dots in line 2. Further, in
> line 2 the dot of a+acute+dotbelow is even more off to the left than
> the already misplaced dots in the remaining two glyphs!?
>
> As a summary we can say that in line 2 both, the accents and the dots
> are placed incorrectly while in line 3 the dots are correctly placed
> as well as the macrons (making the middle glyph the only correctly
> assembled one!) while the accents are off. This seems to be due to the
> order of commands. The first applied diacritic is placed correctly
> _because_ there is a pre-composed glyph for it (if not see image 2).
>
> So it really seems that xelatex doesn't care about the GPOS for latin
> script and just assumes that there are pre-composed glyphs. I wonder
> why this is the case. Wouldn't it be easier to make xelatex recognize
> anchors so that users can compose arbitrary glyphs? Believe it or not
> some humanities use weird combinations :).
>
> Thanks for the help, Alex
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image1.png
Type: image/png
Size: 25332 bytes
Desc: not available
URL: <http://tug.org/pipermail/xetex/attachments/20100901/879001e5/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image2.png
Type: image/png
Size: 26144 bytes
Desc: not available
URL: <http://tug.org/pipermail/xetex/attachments/20100901/879001e5/attachment-0003.png>