Been using these changes for a while now:

Linux version 6.0.13-hardened1 (nixbld@localhost) (gcc (GCC) 12.2.0, GNU ld (GNU Binutils) 2.39) #1-NixOS SMP PREEMPT_DYNAMIC Wed Dec 14 10:41:06 UTC 2022

Haven't noticed any issues

Jobset: https://hydra.nixos.org/jobset/nixpkgs/pr-206907-gcc12

For now just x86_64-linux with low priority. It's likely to advance a lot while staging-next-* waits for *-darwin to catch up, but we'll see.

By looking at this the upgrade might go relatively easily: https://gcc.gnu.org/gcc-12/porting_to.html

Status:

Currently about ~1,000 breakages total on x86_64-linux, pushed some extra patches for packages and will keep at it for a bit while but this is almost ready for review.

The 2 biggest issues:

• vte no longer compiles with clang (which would require stdenv: aarch64-linux: gcc9 -> gcc12 #207135 to push aarch64-linux forward)
• xorg-server 1.20 doesn't compile with gcc12 and xorg-server 1.21 has some reported segfaults on NixOS (haven't experienced those through all this time using gcc12 by default)

Wow, so many -Wno-error= annotations! Thank you!

I feel uneasy about the use after free ones. Dx

Wow, so many -Wno-error= annotations! Thank you!

I feel uneasy about the use after free ones. Dx

Chances are those are false positives: https://gcc.gnu.org/PR104069. Worth a separate look as well, but it should not block a compiler update I'd say.

It's my understanding that some people still have bad issues with the new xorgserver, so why not decouple from PR #206805 by dropping the update from this PR? The old one is fixable trivially:

--- a/pkgs/servers/x11/xorg/overrides.nix
+++ b/pkgs/servers/x11/xorg/overrides.nix
@@ -878,6 +878,8 @@ self: super:
           "--disable-tls"
         ];
 
+        NIX_CFLAGS_COMPILE = [ "-Wno-error=array-bounds" ];
+
         postInstall = ''
           rm -fr $out/share/X11/xkb/compiled # otherwise X will try to write in it
           ( # assert() keeps runtime reference xorgserver-dev in xf86-video-intel and others

Dropped/decoupled xorg-server commit, I must have had a brainfart I thought there was a harder to fix error with gcc12 for xorgserver

Would you mind rebasing this to latest origin/staging? It makes it easier for me to test that #209870 still works regardless of whether it or this merges first.

It's intentionally based on recent master to avoid mixing up with (potential) regressions from master..staging. This makes comparisons easier, e.g. now https://hydra.nixos.org/eval/1789845?compare=1789809

Doing so would also cause a mass rebuild on hydra.

Not sure I should do that resource-wise.

Right, the rebuild on Hydra would seem wasteful to me at this point.

Bisect says cb57273 stdenv: gcc_11 → gcc_12 broke dmd test on master:

$ nix build -f. -L dmd
...
Test 'runnable_cxx/cppa.d' failed. The logged output:
g++ -m64 -c runnable_cxx/extra-files/cppb.cpp -o test_results/runnable_cxx/cppb.cpp.o -std=c++11
/build/dmd/generated/linux/release/64/dmd -conf= -m64 -Irunnable_cxx -checkaction=C  -fPIC -L-lstdc++
 -L--no-demangle  -odtest_results/runnable_cxx -oftest_results/runnable_cxx/cppa_0  runnable_cxx/cppa
.d test_results/runnable_cxx/cppb.cpp.o
/nix/store/0n4dckknrk1fyvh858hyqv9bk3nhn67s-binutils-2.39/bin/ld: test_results/runnable_cxx/cppa_0.o:
 in function `_ZNSaIiE10deallocateEPim':
runnable_cxx/cppa.d:(.text._ZNSaIiE10deallocateEPim[_ZNSaIiE10deallocateEPim]+0x5): undefined referen
ce to `_ZN9__gnu_cxx13new_allocatorIiE10deallocateEPim'
collect2: error: ld returned 1 exit status
Error: linker exited with status 1
...
FAILED targets:
- runnable_cxx/cppa.d