Use __builtin_va_copy instead.

Fix failed test case from pre-commit CI.

Update diff to trigger pre-CI rebuild.

LGTM.

Thanks for working on this version of the patch. It's always good to have something concrete to talk about instead of imaging how it's going to look like when everything is in place.
I'm still not convinced that dynamically allocating csect groups is the way to go though.
As you could see in the patch, the code are more complicated with dynamic allocation, and less efficient(doing more loopings to find the correct CsectGroup).
Also, we now lose control of how the CsectGroup are going to layout inside of a section, and how each section is going to layout in the object file (It now comes in a first come first serve order I believe).
This could bring some practical issues for us. For example, we may prefer to have layouts like: text -> data -> bss -> tdata -> tbss -> dwarf raw datas. But now, I think it's possible that we start having text -> data -> tdata -> bss -> tbss -> dwarf raw datas, which is not quite right when you have tdata cut in between of normal datas.
And I don't see a lot of benefit of having a single vector to contain DWARF sections and csect section. In D97184, we actually handles DWARF sections and csect sections quite differently (i.e. a large block of if/else statement for dwarf and csects).
I would still prefer only dynamically allocating DWARF raw datas, and have pre-defined csect sections. It's less of an issue to dynamically allocating DWARF raw datas because we do not really care the layout between different dwarf raw datas (as long as they are at the end of the object files).

Remove extra conditions because we only intended to target the headers for clang compilers.

@ldionne Hi Louis, what do you think about this patch? Do you have any ABI implication concern?

Making basic_filebuf's __open consistent with other __open.

@hubert.reinterpretcast Does the freelocale approach look good to you?

Use freelocale instead to avoid using thread local on Z.

Add back setlocale and use __ prefix in case of user macro expansion.

Use thread_local for newlocale variable.

Use "C" locale when NULL is passed.

Got it. Make sense. I removed the test case in case we want to have the declaration be more like what standard have in the text.

Ping for reviews from AIX's perspective. @xingxue @daltenty @hubert.reinterpretcast @Xiangling_L

In D97558#2635187, @ldionne wrote:

This LGTM but I would really really like to have a pre-commit node set up for this platform if we are going to support it properly.

Are all those people whose names I don't recognize AIX people? Any of them want to chime in with a "yes, this looks appropriate for our target"?

Rebase so that buildbot could build.

@Quuxplusone While waiting for the second reviewer, I added an extra change since you approved: do not build cxa_thread_atexit.cpp on AIX.
Could you take a quick look to see if that's okay?

Added do not build cxa_thread_atexit.cpp for libcxxabi on AIX.

FYI, I decide to review this patch before D97049 because I wouldn't know if the NFC make sense without the context of this patch.

I think what we usually do for test case is to upstream some pre-compiled binaries for testing purpose. But that's not the way community preferred anyway because those binaries are baked, and might be hard to reproduce later. Given this patch is small and pretty self-explanatory, I'm okay with skipping the test and let D97184 to "test" it.
FYI @hubert.reinterpretcast in case he think we should still upstream a pre-compiled binary for testing.
Otherwise, the patch LGTM.

Thanks @Quuxplusone for the review.
Ping for a second reviewer in libc++ review group.

Address Arthur's comment.

Gentle ping.

LGTM if the nit comment is addressed
There are two items that we need to follow up after this patch:

1. The corner case/bug in https://reviews.llvm.org/D95518#inline-919030
2. The possibility of addressing the FIXME for https://reviews.llvm.org/D95518#inline-918884

In D97558#2590597, @curdeius wrote:

It would be cool if this builder got added to the precommit CI. Any plans on that?

LGTM with minor nit.

LGTM with minor nits.

Thanks. LGTM.
Please allow some time for @hubert.reinterpretcast to go over again before you land.

Rebased on the latest parent.

ping.

Address comments.

Manually closing this because I spelled the Differential Revision wrong.

Address comments and rebase.

Address comments.

Address comments.