Comments

101 Comments

At least one of the gating tests is not compatible with RPM's signature requirement. I'm going to waive it.

This update has been unpushed.

BZ#2403510 rpkg: Teach pre-push check to ignore auto-generated source files
BZ#2382126 binutils: GAS crash with --sframe

@siosm If that's the case, I don't think we can ever again introduce any Lua scripts into glibc. That's not great. (Forcing users to update on particular timeline, first to pick up the rpm-ostree, and then start shipping the new glibc a few weeks later does not seem to be workable).

@adamwill Is it safe to push this now?

karma

I'm going to unpush this because none of the failing tests block promotion of the build to stable.

Retriggered the tests because of a 404 HTTP error when accessing kojipkgs.

@davidlt Aren't we in the release freeze now? I suppose I could prepare a Fedora 42 update and have it wait in the queue.

The problematic GCC commit has been reverted upstream and reimplemented in a completely different way. The crash is gone. We'll do a glibc rawhide sync from upstream glibc, using the new GCC snapshot soon.

I forgot that Fedora 42 uses GCC 15, too. Same regression as in the rawhide update: FEDORA-2025-6616d58c7a

And malloc/malloc.os is the file that is miscompiled.

Bisected down to GCC's r15-7400.

@adamwill Yes, I'm looking into it. Bisecting GCC now. The crash does not reproduce when building current glibc sources with gcc-15.0.1-0.7.fc42.x86_64.

karma

I can reproduce this. It also happens with glibc-2.41.9000-2.fc43.x86_64 if rebuilt against the current buildroot. The original -2 build is fine. There was an update from gcc-15.0.1-0.7.fc42.x86_64 to gcc-15.0.1-0.9.fc43.x86_64.

“indicates that glibc has a per-package gating policy with different requirements for testing vs. stable”

I'm not sure this is intentional. Do we need to add more items here?

decision_contexts:
  - bodhi_update_push_stable
  - bodhi_update_push_stable_critpath

Is this the reason why this update appears stuck?

I re-reported the rpminspect segfault here: https://gitlab.com/testing-farm/general/-/issues/94

I'm going to waive the failure, hoping that it's unrelated to the present builds.

Ugh, looks like I submitted an older build by mistake without checking NVR. Not sure what happened to the -13 build.

Thanks. I need to do another update anyway with a security fix (which this build here introduces into Fedora 40 for the first time). So I suppose we got lucky in some ways.