ICE in gcc 14.1.0 trying to compile emacs from source

[cyrilarnould]

Description / Steps to reproduce the issue

Trying to compile emacs from source, I get an internal compilation error. I do need to patch src/syswait.c to forcibly include <sys/wait.h>, as I get an implicit function declaration error otherwise (I've reported this upstream). The (hacky) patch:

syswait.patch

The steps to reproduce:

git clone https://git.savannah.gnu.org/git/emacs.git
cd emacs
git checkout emacs-29.3
git apply syswait.patch
./autogen.sh
./configure
make

Contrary to #20861 (which seems similar), I'm not compiling with -fno-omit-frame-pointer, just the default CFLAGS -g3 -O2 -gdwarf-2. OTOH, the issue does seem to go away when compiling with -O1.

Expected behavior

No ICE

Actual behavior

  CC       thread.o
during RTL pass: dwarf2
thread.c: In function 'Fthread_yield':
thread.c:725:1: internal compiler error: in dwarf2out_frame_debug_expr, at dwarf2cfi.cc:2112
  725 | }
      | ^
Please submit a full bug report, with preprocessed source (by using -freport-bug).
See <https://github.com/msys2/MINGW-packages/issues> for instructions.

Verification

• I have verified that my MSYS2 is up-to-date before submitting the report (see https://www.msys2.org/docs/updating/)

Windows Version

MINGW64_NT-10.0-19045

MINGW environments affected

• MINGW64
• MINGW32
• UCRT64
• CLANG64
• CLANG32
• CLANGARM64

Are you willing to submit a PR?

Sure

[lazka]

Thanks. Did this work before (with GCC 13.2.0)?

[lazka]

Your example also fails like the other issues if the dwarf flags are not given, so it might be related at least.

I'm going to reduce both cases and file them upstream.

[cyrilarnould]

Thanks. Did this work before (with GCC 13.2.0)?

Haven't personally tested it, but I guess the current mingw-w64-emacs release (also 29.3) is probably still built with GCC 13.2.

I'm going to reduce both cases and file them upstream.

Thanks!

[cyrilarnould]

I saw that -fno-fold-mem-offsets is suggested in your ticket as a workaround for the other issues, and it seems to solve the emacs compilation as well. I want to take another look at the -D_FORTIFY_SOURCE and -fno-optimize-sibling-calls that were added to mingw-w64-emacs in the past before creating a PR though.

Also, the sys/wait.h issue can be fixed using -Wno-error=implicit-function-declaration.

[lazka]

Reduced to this in the meantime:

// gcc -c -O2 repro.c -> internal compiler error: in seh_cfa_offset
// gcc -c -g3 -O2 -gdwarf-2 repro.c -> internal compiler error: in dwarf2out_frame_debug_expr
void a() {
    __builtin_unwind_init();
}

[cyrilarnould]

Ah, I guess there's no need to add the -fno-fold-mem-offsets flag if the upstream patch works, so we'd only need to add -Wno-error=implicit-function-declaration.

In the meantime, I've confirmed that #17550 and #17343 still are a problem, so the -D_FORTIFY_SOURCE and -fno-optimize-sibling-calls modifications are still necessary.