Which revision of LLVM is our llvm-libc based on?

[aheejin]

What revision of LLVM is https://github.com/emscripten-core/emscripten/tree/main/system/lib/llvm-libc based on?

I'm asking because we now need to update our libraries to LLVM 21 is our llvm-libc revision is likely old than that. Because a part of llvm-libc and libcxx should be in sync, I think it's desirable (if not necessary) to llvm-libc at least follow our LLVM regular library updates, in which update libcxx, libcxxabi, compiler-rt, and libunwind. (In case I was not clear, I think it's fine that llvm-libc is newer than our LLVM library version, but shouldn't be older than that)

And without knowing the base revision, it's not easy to extract our Emscripten-specific changes.

I'm not even sure whether all code in llvm-libc is based on the same revision at this point. These are the PRs we imported llvm-libc: #24507, #24530, #24604, #24710, #24765, #24818. In case you remember, what were the revisions you imported the code from? And do you remember any Emscripten-specific changes you made, in case you remember?
@sbc100 @google-yfyang

Going forward, I suggest the following:

1. To the extent possible, I suggest we import from a single revision for all llvm-libc code and write down that revision in https://github.com/emscripten-core/emscripten/blob/main/system/lib/llvm-libc/readme.txt.
2. If 1 is not strictly possible and we need to import a part of the code from a newer revision than our codebase, it should be written down in readme.txt as well.
3. When doing our LLVM library updates, in case the llvm-libc revision is older than the LLVM release version, we update llvm-libc to at least follow the LLVM releases.

Or, should it really be tot LLVM now? If we have most of what we need, can't it follow the same LLVM release update schedule?

[aheejin]

Another question: It looks #24604 undid several Emscripten-specific changes done in #24507, for example, things like

Was this intentional?

[google-yfyang]

Thanks for looking into this! I certainly agree that we need to make llvm-libc at least on par with the rest of the LLVM libraries.

Or, should it really be tot LLVM now? If we have most of what we need, can't it follow the same LLVM release update schedule?

I think we are fine with following the same schedule although potentially we might want certain upstream changes faster in the future.

In case you remember, what were the revisions you imported the code from? And do you remember any Emscripten-specific changes you made, in case you remember?

The imported llvm-libc change should come from a single revision. But unfortunately I have since repurposed that local repo and the exact revision the import was based on is kinda hard to obtain now. At least some if not all of the Emscripten-specific changes made to llvm-libc sources should be submitted upstream. llvm/llvm-project#154383. But it's been a while so I don't quite remember if that is all we need.

I can try working on updating the llvmlibc if you can just tell me the revision number that the rest of the library is based on.

[google-yfyang]

Another question: It looks #24604 undid several Emscripten-specific changes done in #24507, for example, things like

Was this intentional?

They should be. But it looks like the change in inline_memcpy.h might be unintentional as it leads to a behavior change.