GNU bug report logs - #78473
30.1; feature/igc: multiple crashes with pgtk_handle_selection_*

Previous Next

Full log

View this message in rfc822 format

From: Pip Cet <pipcet <at> protonmail.com>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: oliver.reiter <at> snapdragon.cc, 78473 <at> debbugs.gnu.org
Subject: bug#78473: 30.1; feature/igc: multiple crashes with pgtk_handle_selection_*
Date: Sun, 18 May 2025 11:20:00 +0000

"Eli Zaretskii" <eliz <at> gnu.org> writes:

>> Date: Sat, 17 May 2025 20:58:49 +0000
>> From:  Pip Cet via "Bug reports for GNU Emacs,
>>  the Swiss army knife of text editors" <bug-gnu-emacs <at> gnu.org>
>>
>> "Oliver Reiter via \"Bug reports for GNU Emacs, the Swiss army knife of text editors\"" <bug-gnu-emacs <at> gnu.org> writes:
>>
>> > Dear all,
>> >
>> > I have encountered multiple crashes on feature/igc with pgtk when
>> > copying from Emacs to somewhere else. All crashes seem to involve some
>> > form of 'pgtk_handle_selection_*' function (xbacktrace is always empty).
>> > These crashes appear to be reproducible for me, if you need further
>> > info.
>>
>> Thanks for the report!  Can you try adding -fno-tree-sra to your CFLAGS
>> and recompiling?  This looks like
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117423
>>
>> > #9  pgtk_handle_selection_request (event=0x7fffffffc780)
>> >     at /home/reitero/build/sources/emacs/emacs_debug/src/pgtkselect.c:585
>> >         local_selection_time = <optimized out>
>> >         dpyinfo = 0x55dee7e0
>>
>> This pointer has been truncated to 32 bits; it should be 0x555555dee7e0.
>> That's consistent with the gcc bug above (and
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119085, which has been
>> correctly marked as a duplicate).
>
> Is there any way we could change our code to avoid bumping into this
> problem?

I thought we had, to be honest.  Let's confirm this is 117423 first,
then we can figure out how to work around that.

> The bug exists in a relatively new version of GCC, so more
> users will hit it in the years to come.

Indeed, and that's if it is fixed at all.

> One way is to force the use of -fno-tree-sra (for the affected
> versions of GCC, if we can identify them), but is there maybe a less
> blunt instrument?

That'd be great, but I don't know how to do it.

>> > Configured using:
>> >  'configure 'CFLAGS=-g3 -ggdb -O3 -mtune=native -march=native
>> >  -fomit-frame-pointer' CPPFLAGS=-I/home/reitero/.local/lib/mps
>>
>> Using -fomit-frame-pointer will result in the compiler putting
>> references to GC-able objects in %rbp; since MPS usually uses setjmp to
>> spill the registers to the stack, and setjmp "scrambles" %rbp for
>> security reasons, that will result in references which are invisible to
>> MPS, which will cause other crashes.
>
> But configure.ac on the branch forces -fno-omit-frame-pointer.  Is
> that not working for some reason?

Both options are passed to GCC, and -fomit-frame-pointer wins because
it's last.

Pip

Previous Next

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