[XeTeX] bug in polyglossia/sanskrit? (problem with diacritical marks)

[Steve White]

Hi,

I looked through the attachments to the previous postings, and some of the
images are consistent with the positioning feature being turned off.
That may now be fixed for FreeSerif.

However there were some that showed a completely wrong mark appearing,
far from the base mark.  I have no idea what that might be, but it's
hard for me to tell
just what fonts were causing which problems.

Could somebody please systematically make images of

    TL 2017  vs  2018
    fonts  FreeSerif  vs  Sanskrit 2003  vs  Murty Sanskrit

If anybody needs a recent build of GNU FreeFont,  I can provide that.

Thanks!


On Thu, May 16, 2019 at 10:43 PM Steve White <stevan.white at gmail.com> wrote:
>
> Hi Jonathan,
>
> You may have a point, although I don't think it's the whole story.
>
> (Most of what follows is font-techy. Upshot is: I did find suspicious
> issues in FreeFont, and have now fixed them.)
>
> The font feature in FreeSerif that positions generic marks like these is
>
> mark ('generic marks above')
>
> Indeed the 'dev2' script was different from the 'deva' script, being set
> nly for language 'dflt'.  That is formally a bug, but I don't see why it
> would matter here, as the "script" in which the problem happens is Latin
> (the base characters beinig from the Latin range).
>
> However, now that I look at the 'latn' script entries for that same feature...
> it lacks the 'SAN ' language.  As I understand the logic, that would
> result in the feature not being applied.
>
> Now in the case of FreeFont, that should result in the mark crudely
> appearing over the preceding character without GPOS placement, as these
> marks have negative right bearing.
>
> I have fixed these issues in FreeFont SVN.
>
> It will be interesting to hear if they alleviate the issue in the newer
> version of TL.  It might suggest triggers the effect.
>
> Cheers!
>
>
> On Thu, May 16, 2019 at 2:30 PM Jonathan Kew <jfkthame at gmail.com> wrote:
> >
> > On 16/05/2019 12:31, Ulrike Fischer wrote:
> > > Am Mon, 13 May 2019 17:52:17 +0200 schrieb François Patte:
> > >
> > >> I have just made a fresh install of texlive 2019 and encountered a
> > >> problem with the diacritical marks of romanized sanskrit.
> > >
> >
> > Do the 'dev2' script tables in FreeSerif include positioning features
> > for the various combining diacritics on Latin characters? I'm guessing
> > not, in which case it's unsurprising that the results will be poor if
> > you present it with accented Latin, yet request Devanagari shaping behavior.
> >
> > I'd expect romanized Sanskrit should be typeset with script=latn, or
> > with no explicit script specification.
> >
> > JK
> >
> > > (copied from f.c.t.t):
> > >
> > > One can see it with plain xetex too, and it is not the font (I'm
> > > forcing it to 2018), et it is not xdvipdfmx (it happens with tl19
> > > and tl18 version). So it is quite probably an xetex engine bug.
> > >
> > > A plain xetex example:
> > >
> > > \XeTeXtracingfonts=1
> > > \font\test="[C:/texlive/2018/texmf-dist/fonts/opentype/public/gnu-freefont/FreeSerif.otf]/OT:script=dev2;"
> > > \test  ā ī ū ṛ ṝ ḷ ḹ ṭ ḍ ṅ ś ṣ ḥ
> > > \bye
> > >
> > >
> > > Comparing the versions gives
> > > 2019:
> > > Compiled with ICU version 63.1; using 63.1
> > > Compiled with zlib version 1.2.11; using 1.2.11
> > > Compiled with FreeType2 version 2.9.1; using 2.9.1
> > > Compiled with Graphite2 version 1.3.13; using 1.3.13
> > > Compiled with HarfBuzz version 2.3.1; using 2.3.1
> > > Compiled with libpng version 1.6.36; using 1.6.36
> > > Compiled with poppler version 0.68.0
> > > Compiled with fontconfig version 2.13.1; using 2.13.1
> > >
> > > 2018:
> > > Compiled with ICU version 61.1; using 61.1
> > > Compiled with zlib version 1.2.11; using 1.2.11
> > > Compiled with FreeType2 version 2.9.1; using 2.9.1
> > > Compiled with Graphite2 version 1.3.11; using 1.3.11
> > > Compiled with HarfBuzz version 1.7.6; using 1.7.6
> > > Compiled with libpng version 1.6.34; using 1.6.34
> > > Compiled with poppler version 0.64.0
> > > Compiled with fontconfig version 2.13.0; using 2.13.0
> > >
> > > The first suspect is HarfBuzz.
> > >
> > >
> > >
> > >
> >