[XeTeX] wrong kerning and slow processing with xelatex in texlive 2010

[Vadim Radionov]

Hi, Vladimir,

> From you results I could deduce that something wrong goes with xetex
> from TL2010. You could enable kpathsea debugging and record used files
> (-recorder, -kpathsea-debug=1 or 2 or 3, but beware it will generate a
> lot of outout) to compare them.

I am not so sure now.  Before I start with kpathsea debugging, i tried
this example to verify the actual kerning (with TL 2010)

========
\font\b = "CharterITC:+smcp"

\hoffset-.5in\voffset-.5in
\baselineskip5.1pt
\newbox\tmp
\setbox\tmp\hbox{\b aт}\showthe\wd\tmp
\copy\tmp\noindent\b ат

\setbox\tmp\hbox{\b at}\showthe\wd\tmp
\copy\tmp\noindent\b at

\setbox\tmp\hbox{\b та}\showthe\wd\tmp
\copy\tmp\noindent\b та

\setbox\tmp\hbox{\b ta}\showthe\wd\tmp
\copy\tmp\noindent\b ta

\end
========

Letter widths (in fontforge): "Asmall" 532 "Tsmall" 509, the same for Cyrillic.
There are 3 kern lookups in the font dealing with certain pairs but
not these ones.
There is also 1 lookup with kerning classes; it specifies kerning -40
for the classes Latin letters belong to and -50 for Cyrillic (-32 and
-60, respectively, for transpositions)

This is the result of processing with xetex in TL2010 (and the pdf is attached)

> 10.41pt.
l.7 \setbox\tmp\hbox{\b aт}\showthe\wd\tmp
> 10.01pt.
l.10 \setbox\tmp\hbox{\b at}\showthe\wd\tmp
> 8.31pt.
l.13 \setbox\tmp\hbox{\b та}\showthe\wd\tmp
> 10.09pt.
l.16 \setbox\tmp\hbox{\b ta}\showthe\wd\tmp

For Latin letters everything works as expected.  But for Cyrillic,
10.41pt is a size of 2 letters without kerning, and, surprisingly, the
output is different with 2 ways used.  And with reverse order of
letters, kerning of -210 occurs.

I wonder what this all mean.

Vadim
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test2.pdf
Type: application/pdf
Size: 3087 bytes
Desc: not available
URL: <http://tug.org/pipermail/xetex/attachments/20101012/d81cfe42/attachment.pdf>