llvm lld flang 19.1.1

[carlocab]

• Have you followed the guidelines for contributing?
• Have you ensured that your commits follow the commit style guide?
• Have you checked that there aren't other open pull requests for the same formula update/change?
• Have you built your formula locally with HOMEBREW_NO_INSTALL_FROM_API=1 brew install --build-from-source <formula>, where <formula> is the name of the formula you're submitting?
• Is your test running fine brew test <formula>, where <formula> is the name of the formula you're submitting?
• Does your build pass brew audit --strict <formula> (after doing HOMEBREW_NO_INSTALL_FROM_API=1 brew install --build-from-source <formula>)? If this is a new formula, does it pass brew audit --new <formula>?

---

• llvm 19.1.1
• lld 19.1.1
• flang 19.1.1

[cho-m]

Based on https://github.com/orgs/Homebrew/discussions/5631, we probably should disable EH since we now ship Flang.

homebrew-core/Formula/l/llvm.rb

Line 111 in 77f6529

-DLLVM_ENABLE_EH=ON

This is how other repositories that ship flang build llvm. We are doing a bit of a hack by setting -D LLVM_ENABLE_EH=OFF which gets overridden when running find_package(LLVM ...) which allows us to bypass the error while still linking with exception symbols.

Other References:

1. https://github.com/llvm/llvm-project/blob/llvmorg-19.1.0/flang/CMakeLists.txt#L17-L23
2. flang build not enforcing style guidelines llvm/llvm-project#59353
3. https://discourse.llvm.org/t/compiling-with-flang-new/66808
4. https://github.com/spack/spack/blob/develop/var/spack/repos/builtin/packages/llvm/package.py#L846-L854

---

EDIT: If we want to keep it on, may need to add at minimum some caveats or perhaps some workarounds as our build is non-standard.

[carlocab]

I'd just follow what Fedora are doing. LLVM's lead maintainer and one (out of two) of the LLVM release managers maintain the Fedora packages.

But if we want to turn off EH and/or RTTI (though IIRC RTTI requires EH -- but maybe it was the other way around), I would suggest saving it for LLVM 20 because changing those settings will require recompiling everything that links to libLLVM.

[cho-m]

Actually my comment was wrong and Fedora does disable. I confirmed by downloading RPM¹² and checking LLVMConfig.cmake

set(LLVM_ENABLE_ASSERTIONS OFF)

set(LLVM_ENABLE_EH OFF)

set(LLVM_ENABLE_FFI TRUE)
if(LLVM_ENABLE_FFI)
  find_package(FFI)
endif()

set(LLVM_ENABLE_RTTI ON)

Footnotes

1. https://koji.fedoraproject.org/koji/buildinfo?buildID=2549315 ↩

2. https://kojipkgs.fedoraproject.org//packages/llvm/19.1.0/1.fc42/aarch64/llvm-devel-19.1.0-1.fc42.aarch64.rpm ↩

[carlocab]

That's weird, since their spec files show LLVM_ENABLE_EH=ON:

https://src.fedoraproject.org/rpms/llvm/blob/3b5b15045a8c9449d53582df0842c21465e99e4f/f/build.spec.inc#_70
https://src.fedoraproject.org/rpms/llvm/blob/3b5b15045a8c9449d53582df0842c21465e99e4f/f/build.spec.inc#_241

Probably worth investigating further.

[cho-m]

That's weird, since their spec files show LLVM_ENABLE_EH=ON:

Looks like it may be arch-specific. x86_64 rpm does have it enabled:

set(LLVM_ENABLE_ASSERTIONS OFF)

set(LLVM_ENABLE_EH ON)

set(LLVM_ENABLE_FFI TRUE)
if(LLVM_ENABLE_FFI)
  find_package(FFI)
endif()

set(LLVM_ENABLE_RTTI ON)

Anyway, I think the upstream CMake check (llvm/llvm-project@f7e58c9) wasn't tested on Standalone mode which is why it currently works.

There should probably be a caveat as CMake LLVMFlang can't handle our flang which is usually a bad sign.

[carlocab]

We could just create a wrapper script that always adds -l[std]c++ 🤷 Or perhaps a config file next to it (when installed in libexec) that isn't linked to bin.

Setting LLVM_ENABLE_EH could be a big change. Probably worth opening an issue/discussion for feedback, plus a note in caveats linking to the issue/discussion. Then if no one complains we can set it to OFF for LLVM 20.

CC @Bo98 for thoughts.

[Bo98]

Setting LLVM_ENABLE_EH could be a big change.

It can affect ABI at least. Though I also agree we should try disable it & RTTI.

LLVM 20 is definitely a good time to make changes. Somewhat relatedly: perhaps worth considering if libcxx probably should be split to a separate formula. We've had regular bug reports about it since the LLVM C++ headers got prioritised again.

[carlocab]

Or perhaps a config file next to it (when installed in libexec) that isn't linked to bin.

Testing creating a libexec/"bin/flang.cfg" that contains only -lc++. Seems to work. Will add it in a follow-up, plus some caveats. #192611

[carlocab]

afl++ bakes in Cellar paths. Will have a fix separately.

[github-actions]

@carlocab has requested bottles to be published to this PR.