Re: git: 3a0793336edf - main - Merge llvm-project release/18.x llvmorg-18.1.6-0-g1118c2e05e67

From: Dimitry Andric <dim_at_FreeBSD.org>
Date: Thu, 30 May 2024 13:18:54 UTC
On 30 May 2024, at 13:53, Colin Percival <cperciva@tarsnap.com> wrote:
> 
> On 5/30/24 02:06, Dimitry Andric wrote:
>> On 30 May 2024, at 05:00, Colin Percival <cperciva@tarsnap.com> wrote:
>>> On 5/24/24 08:59, Dimitry Andric wrote:
>>>> commit 3a0793336edfc21cb6d4c8c5c5d7f1665f3e6c5a
>>>>     Merge llvm-project release/18.x llvmorg-18.1.6-0-g1118c2e05e67
>>>>          This updates llvm, clang, compiler-rt, libc++, libunwind, lld, lldb and
>>>>     openmp to llvm-project release/18.x llvmorg-18.1.6-0-g1118c2e05e67.
>>> 
>>> This broke the weekly snapshot builds for powerpc* platforms.  It looks like
>>> we have the same issue as https://github.com/llvm/llvm-project/issues/92991
>>> (failure is "Running pass 'PPC Merge String Pool' on module '/usr/src/crypto/openssh/sshkey.c'.") so hopefully importing the fix from that issue can fix
>>> them for next week.
>> I tried reproducing the crash with a "make buildworld TARGET=powerpc
>> TARGET_ARCH=powerpc64", but it worked just fine for me. Is there any way
>> to reproduce this on a non-PowerPC host?
> 
> There seems to be some indeterminacy going on here; all four powerpc*
> platforms broke for 15.x, but the stable/13 and stable/14 builds (which
> I think have the same llvm code) worked fine.  The weekly snapshot builds
> happen on amd64 hardware, so using a powerpc host is not required.
> 
>> That said, I will be happy to import the fix, but it would be nice if I
>> can verify that it is really solving the problem. :)
> 
> I have the files from the failed builds, so I can give you the logs and the
> preprocessed source and build script that clang reported.  That might be
> enough to reproduce it?  Let me know if you'd like those.

I have been able to reproduce it now, using the .bc file from
<https://github.com/llvm/llvm-project/issues/92991#issuecomment-2125387055>.

I can confirm that the upstream fix solves the crash, and will commit it
later today. This uses the default MFC timeout of 3 days, is that early
enough for the next snapshot builds?

-Dimitry