On Monday, August 5th, 2024 at 16:20, Pip Cet <pipcet@protonmail.com> wrote:
> "Eval Exec" execvy@gmail.com writes:
> 
> > I wanted to help test the scratch/igc branch, so I compiled it from
> > commit 42731228d24.
> 
> Thank you for doing that!
> 
> > Then I build it by
> > ```bash
> > make extraclean
> > ./autogen.sh \
> > && ./configure \
> > --prefix=$(realpath ../emacs-build)\
> > --with-mps \
> > --with-imagemagick \
> > --with-modules --with-x-toolkit=lucid --without-compress-install \
> > --without-toolkit-scroll-bars --with-native-compilation --with-mailutils\
> > --with-tree-sitter --with-xinput2 \
> > --with-dbus --with-native-compilation=aot \
> > --with-file-notification=inotify\
> > && make -j20 install
> 
> 
> Like Eli, I'm suspecting the Lucid widget code...
> 
> > It works perfectly for the first hour, but then it crashes, and I
> > don't know how to reproduce the issue.
> > 
> > I think the crash is caused by "window size adjustments"
> > ```
> > (gdb) bt
> > #0 0x00007f47afea2efc in __pthread_kill_implementation () from
> > /nix/store/dbcw19dshdwnxdv5q2g6wldj6syyvq7l-glibc-2.39-52/lib/libc.so.6
> > #1 0x00007f47afe52e86 in raise () from
> > /nix/store/dbcw19dshdwnxdv5q2g6wldj6syyvq7l-glibc-2.39-52/lib/libc.so.6
> > #2 0x00000000004243a4 in terminate_due_to_signal (sig=sig@entry=8,
> > backtrace_limit=backtrace_limit@entry=40) at emacs.c:470
> > #3 0x00000000004248fc in handle_fatal_signal (sig=sig@entry=8) at sysdep.c:1800
> > #4 0x000000000056f568 in deliver_thread_signal (sig=8,
> > handler=0x4248f1 <handle_fatal_signal>) at sysdep.c:1792
> > #5 0x000000000056f669 in deliver_fatal_thread_signal (sig=<optimized
> > out>) at sysdep.c:1812
> > #6 <signal handler called>
> > #7 0x000000000069e35e in pixel_to_char_size
> > (pixel_width=pixel_width@entry=482,
> > pixel_height=pixel_height@entry=76,
> > char_width=char_width@entry=0x7ffced20da38,
> > char_height=char_height@entry=0x7ffced20da3c,
> > ew=<optimized out>) at widget.c:172
> 
> 
> That code reads:
> 
> static void
> pixel_to_char_size (EmacsFrame ew, Dimension pixel_width,
> Dimension pixel_height, int *char_width, int *char_height)
> {
> struct frame *f = ew->emacs_frame.frame;
> 
> 
> *char_width = FRAME_PIXEL_WIDTH_TO_TEXT_COLS (f, (int) pixel_width);
> *char_height = FRAME_PIXEL_HEIGHT_TO_TEXT_LINES (f, (int) pixel_height);
> }
> 
> Which gets a frame pointer from a structure (ew) that MPS doesn't know
> about; but frames can move, so that explains the bug, because MPS
> doesn't know to update the pointer in the EmacsFrame structure.
> 
> Unfortunately, the structure is allocated by XtCreateWidget, which calls
> malloc(), IIUC. So it's not immediately obvious (to me) how to fix it.
> My best idea so far is to create an ambiguous root covering the frame
> pointer and store that in the structure instead.

So I've done that, can you try the attached patch? This is a first stab (thus the ambiguous alloc), but it should prevent the crash we've seen.

TODO list for this:
- use precise alloc
- check framep, not *framep, in EmacsFrameInitialize

Pip