[PATCH] D107799: [CMake] Enable LLVM_ENABLE_PER_TARGET_RUNTIME_DIR by default on Linux

Fangrui Song via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Jun 21 13:21:41 PDT 2022


MaskRay added a comment.

In D107799#3598878 <https://reviews.llvm.org/D107799#3598878>, @DavidSpickett wrote:

>> Having this on by default in some configs on linux and off on others seems like it'd be very confusing. Let's please either have it on or off uniformly per OS?
>
> I wasn't sure this met the revert bar but sure suits me: https://github.com/llvm/llvm-project/commit/90c9d41c8acc34fb68958f373eb23f53c1b1c840
>
> I am working to make it work but mostly from the point of view of not having broken builds. I don't have background on what this is trying to achieve overall and if arm/armhf will even fit that vision at all.

I think defaulting arm/armhf to LLVM_ENABLE_PER_TARGET_RUNTIME_DIR=off is fine. So I reverted your revert.

> So @MaskRay if you can summarise the goal and/or I can provide you with instructions to reproduce the armhf config and you can try to fix it up yourself.

See the summary for the goal. It is more important for `LLVM_ENABLE_PROJECTS` and `LLVM_ENABLE_RUNTIMES` to have the same behavior, and push target maintainers to fix CMake.
(Note that `LLVM_ENABLE_PROJECTS=libcxx;libcxxabi;libunwind` has been deprecated for quite a while now.)

Your arm failure link likely just exposed an issue that LLVM_ENABLE_RUNTIMES=compiler-rt (and probably libcxx;libcxxabi;libunwind) likely doesn't work (well) with arm/armhf.
Assuming that you maintain arm/armhf builds, you probably need to investigate how to make them work. I don't have an arm/armhf machine.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D107799/new/

https://reviews.llvm.org/D107799



More information about the llvm-commits mailing list