libharfbuzz0 1.7.6-1 update causing xwin-xdg-menu.exe to crash

[Jon Turney]

On 26/04/2018 16:40, Gilles Detillieux wrote:
> On 2018-04-26 09:03, Jon Turney wrote:
>> On 19/04/2018 22:15, Gilles Detillieux wrote:
>>> Has anybody else run into this problem? I've done two installations 
>>> of Cygwin/X on Windows 10 systems this week, and they both had 
>>> problems with the XWin Server dying just a few seconds after starting 
>>> up. I traced the problem back to xwin-xdg-menu.exe getting a 
>>> Segmentation fault, which then causes XWin Server to exit. I hacked 
>>> an alternate .startxwinrc file to prevent XWin Server from dying (it 
>>> ends with a "sleep infinity"), so I could debug it further.
>>>
>>> With the XWin Server running reliably, I then ran "strace 
>>> xwin-xdg-menu.exe" and saw that it got a segmentation fault just 
>>> after reading a TTF font from the Windows Font directory 
>>> (bahnschrift.ttf if it matters). I noticed there were two recent 
>>> library updates related to font handling, so I tried back out to the 
>>> previous version for each. It turns out that when I reverted to 
>>> version 1.7.4-1 of libharfbuzz0, xwin-xdg-menu.exe stopped crashing.
>>>
>>> If it matters, both these systems are the Fall Creator's Update 
>>> (1709) of Windows 10 64-bit, and I'm running the 32-bit version of 
>>> Cygwin.
>>>
>>> Hopefully someone can track down and fix this recent bug!
>>
>> Thanks for reporting this.
>>
>> I can reproduce this problem, but it only seems to occur with 32-bit 
>> cygwin.
>>
>> (Obviously you also need a recent enough Windows 10 to have the 
>> Bahnschrift font)
>>
>> The actual crash seems to be in fontconfig, e.g. 'fc-query 
>> /usr/share/fonts/microsoft/bahnschrift.ttf' fails in the same way.

Another possible workaround seems to be to blacklist this particular 
font, e.g.:

create a /etc/fonts/conf.d/local.conf containing:

<selectfont>
     <rejectfont>
         <glob>/usr/share/fonts/microsoft/bahnschrift.ttf</glob>
     </rejectfont>
</selectfont>

>> I didn't get very far investigating the problem, as rebuilding the 
>> fontconfig package with the current toolchain seems to be enough to 
>> make the problem go away.
> 
> Thanks for the follow-up and narrowing down the problem, Jon. 
> Interesting that rebuilding fontconfig clears up the issue. Although, if 
> it's a memory corruption issue, it could just be that the new toolchain 
> lays things out differently enough that the bug doesn't manifest itself 
> the same way. It could also be that the new gcc fixes a compiler or 
> optimizer bug that led to the problem. Perhaps you and Yaakov could 
> touch base on which toolchain versions you're using and see if an update 
> to his toolchain may be in order.
> 
> Are you using the test version of gcc (7.3.0-1) announced April 11, or 
> the older release. I've got gcc-core-6.4.0-5 on mine, which I assume is 
> the latest stable release.

The latest stable release, 6.4.0-5.



--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple