[XeTeX] segfault with xetex

[Jonathan Kew]

On 3 Sep 2006, at 11:54 am, Pablo Rodríguez wrote:

> Jonathan Kew wrote:
>>
>> Of course, xetex shouldn't crash even if you do have a bad font; and
>> the "problem" font might not really be wrong at all, it could simply
>> be a xetex bug. But until we can isolate which font is triggering the
>> problem, it's hard to be sure.
>>
>> So if you're willing to spend some time tracking this down, I'd
>> suggest removing "extra" fonts from your font directories (half at a
>> time, perhaps), and re-testing, to try and narrow down the problem.
>> What you *should* get with "junk" names like this is a simple "! Font
>> not found" error.
>
> Jonathan, just accidentally I found out which font causes the
> segmentation fault when not finding the font name for any font in my
> font directories (http://www.orthodoxia.gr/media/athena.zip).
>
> Don't forget that the segmentation fault happened after I removed some
> Type1 fonts when xetex was running. Because of that, it might be a bug
> in fontconfig.

I don't think so. I just installed this font on a Linux system, and  
easily reproduced the segfault with xetex.

> Just in case it helps, Pango 1.14.x seems to have problems with this
> font (it makes Firefox crash when browsing a page that loads the font
> [bug already reported]).

Reported as a bug in Pango, Firefox, or the font? I believe it's  
primarily the font at fault here.

It looks to me as though the OpenType tables in this font are  
invalid, and that's why xetex (and other programs) have problems when  
they try to interpret them. It's a broken font. I notice that it is  
dated 1997; there were probably very few applications at that time  
that could actually test the tables, so perhaps this isn't too  
surprising.

IMO, the actual glyph design in this font is also not very good. The  
Greek outlines are poorly drawn (could they be auto-traced from  
something?), and many of the Latin letters have poorly-placed  
accents. And the Greek and Latin don't harmonize at all. It looks  
like it was an early attempt at broad character coverage, but could  
use extensive work to improve the glyphs, in addition to apparently  
needing OpenType bug-fixing. (Actually, the Latin glyph shapes may  
well have been copied from an existing Baskerville font, and then  
lots of composites auto-generated; I'd be concerned that this could  
be a copyright/license violation, depending on the original source,  
but I don't know the details.)

So my suggestion would be to simply abandon it and use a different  
font. But I will also try to identify more precisely where xetex  
crashes, and make it more robust if possible. In principle, a bad  
font shouldn't crash the program, it should just fail to work as  
intended.

JK