GNU bug report logs - #79116
31.0.50; Crash on IGC build

Previous Next

Full log

Message #32 received at 79116 <at> debbugs.gnu.org (full text, mbox):

From: Pip Cet <pipcet <at> protonmail.com>
To: Helmut Eller <eller.helmut <at> gmail.com>
Cc: Sean Devlin <spd <at> toadstyle.org>, 79116 <at> debbugs.gnu.org
Subject: Re: bug#79116: 31.0.50; Crash on IGC build
Date: Wed, 30 Jul 2025 12:51:24 +0000

"Helmut Eller" <eller.helmut <at> gmail.com> writes:

> On Tue, Jul 29 2025, Pip Cet via "Bug reports for GNU Emacs, the Swiss army knife of text editors" wrote:
>
>> Thanks for the report. I'm afraid we don't have very much to go on: this
>> assertion in trace.c in MPS failed:
>>
>>     AVER(RefSetSub(ScanStateUnfixedSummary(ss), SegSummary(seg))); /* <design/check/#.common> */
>
> This may or may not be relevant: the GC literature defines a weak and a
> strong tricolor invariant.  I wonder if this code checks those
> invariants.
>
> In fix_frame, we do something that may not be kosher: we trace the glyph
> pool.  We can access the glyph pool because it lives outside the GC heap
> and has no memory barriers (and the world is stopped).  However, MPS may
> assume that we scan such regions as roots because there are no memory
> barriers on them to enforce the tricolor invariants.  I don't understand
> the invariants well enough to tell if what we do is harmless or not.
> Perhaps someone who does could think this through.

Good catch!

I think you're absolutely correct. What we do there definitely isn't
correct: if one of the frame's pools is behind a read barrier, there's a
tiny race condition, but a write barrier would effectively be ignored,
which might well cause problems like this one.

But we should fix it even if it didn't cause this particular crash.
igc_xpalloc_ambig shouldn't be hard to do, but it's probably better for
performance to turn glyph pools into their own IGC object type.

Pip

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.