Re: git: 8855efffd4fa - main - devel/llvm23: 23.1.0-rc3

On Fri, Aug 14, 2026 at 01:27:48PM -0700, Mark Millard wrote:
> On 8/14/26 02:17, Brooks Davis wrote:
> > The branch main has been updated by brooks:
> > 
> > URL: https://cgit.FreeBSD.org/ports/commit/?id=8855efffd4fa0467a5faff409d18444f76c0dbfb
> > 
> > commit 8855efffd4fa0467a5faff409d18444f76c0dbfb
> > Author:     Brooks Davis <[email protected]>
> > AuthorDate: 2026-08-14 09:16:40 +0000
> > Commit:     Brooks Davis <[email protected]>
> > CommitDate: 2026-08-14 09:17:12 +0000
> > 
> >     devel/llvm23: 23.1.0-rc3
> >     
> >     Attempt to fix compilation on 32-bit targets.
> 
> Hmm. When I previously tried the build that included .r2 in its
> /usr/ports ugrade, the build got:
> 
> . . .
> [02:40:33] [06] [02:39:47] Finished   devel/llvm23@default |
> llvm23-23.1.0.r2: Success
> . . .
> [02:41:09] Committing packages to repository:
> /usr/local/poudriere/data/packages/main-armv7-default/.real_1786620042
> via .latest symlink
> . . .
> 
> (It was armv7 poudriere-devel jail on aarch64 Windows Dev Kit 2023,
> which supports armv7 user space code. The jail is from an official
> upstream pkgbase distribution for armv7 main.)
> 
> So, what is sometimes broken about the compilation of llvm23-23.1.0.r2 ?

See files/patch-llvm_include_llvm_Support_Allocator.h

The failure on i386 was that this assertion failed:

  static_assert(MinAlign <= alignof(std::max_align_t),
		"MinAlign must not exceed the alignment of fresh slabs");

It sounds like I should narrow the patch to i386 only as it's down to the weak
alignment i386.

Thanks for the follow up.

-- Brooks