[tlbuild] test errors on some archs

[Peter Breitenlohner]

On Wed, 24 Sep 2014, Norbert Preining wrote:

> I retested with current svn and shlib building on Debian, and that
> is what I get:
>
> Still 4 checks fail :
>
> On Mon, 15 Sep 2014, Peter Breitenlohner wrote:
>>> FAIL: omegaware/check.test
>>> FAIL: omegaware/ligkern.test
>>> FAIL: omegaware/vrepeat.test
>>> FAIL: omegaware/yannis.test
>
> For several archs, I haven't checked all of them:
>>> The failing archs I see are:
>>> 	hurd-i386
>>> 	armel
>>> 	armhf
>>> 	i386 (!!!!)
>>> 	powerpc
>>> 	sparc

Hi Norbert,

these are all 32bit system, right?

which compilers/versions?

> I did checks on i386 and can reproduce it in my own pbuilder.

strange, perhaps a problem with optimization?

> What happens is that wofm2opl segfauls. Running gdb the bin and
> a generated core file I see:
> Core was generated by `/tmp/texlive-bin-2014.20140923.35230/Work/texk/web2c/.libs/lt-wofm2opl omegawar'.
> Program terminated with signal SIGSEGV, Segmentation fault.
> #0  0x0804a360 in hashinput () at ofm2opl.c:738
> 738	    switch ( t )
> (gdb) bt
> #0  0x0804a360 in hashinput () at ofm2opl.c:738
> #1  0x0804f815 in mainbody () at ofm2opl.c:3036
> #2  0x080489c6 in ?? ()
> #3  0xf75d8723 in ?? ()
> (gdb)
>
> Interestingly, not *all* ofm2opl invocations fail. From testsuite.log:
> FAIL: omegaware/yannis
> ======================
>
>
> *** pltotf arabicr yarabicr
> I had to round some heights by 0.0279999 units.
> I had to round some depths by 0.0150003 units.
>
> *** opl2ofm arabicr.pl zyarabicr
> I had to round some heights by 0.0279999 units.
> I had to round some depths by 0.0150003 units.
>
> *** cmp yarabicr.tfm zyarabicr.ofm
>
> *** ofm2opl yarabicr.tfm xyarabicr
>
> *** ofm2opl zyarabicr zyarabicr

yarabicr.tfm = zyarabicr.ofm has no ligatures

> *** diff xyarabicr.opl zyarabicr.opl
>
> *** ovp2ovf arabic yarabic yarabic
>
> *** ofm2opl yarabic xyarabic
> Segmentation fault (core dumped)

yarabic.ofm does have ligatures

=================

All failing tests use .tfm/.ofm files with ligaures.  Seems there is a
problem with hashinput(), used to check for ligature loops.  Perhaps I find
something looking at the code.

Regards
Peter