Avoiding indirect linkage issues

[mahrud]

Description of issue

I have a formula on a tap that builds a rather large project using CMake. I'm wondering if anyone with experience using CMake and Homebrew could help me fix the following indirect linkage issue, or point me to other projects that use CMake and depend on libomp:

GitHub Actions / test-bot (macos-14)
`brew linkage --cached --test --strict macaulay2/tap/macaulay2` failed on macOS Sonoma (14) on Apple Silicon!
Indirect dependencies with linkage:
  icu4c
  libomp

GitHub Actions / test-bot (ubuntu-latest)
`brew linkage --cached --test --strict macaulay2/tap/macaulay2` failed on Linux!
Indirect dependencies with linkage:
  gcc

(The specific pull request that resulted on this is here.)

Surely there's a way to statically link libomp, icu4c, and gcc, but I can't figure out how.

[SMillerDev]

Surely there's a way to statically link libomp, icu4c, and gcc, but I can't figure out how.

Very few projects do static linking in Homebrew since it's actively discouraged. Your usage of std_cmake_args almost certainly means that it's using a dynamic linking build.

[mahrud]

Good point, but then what is the warning asking me to do?

Also, icu4c is particularly annoying because it's linking with a shared library in a cellar:

	libicudata.so.74 => /home/linuxbrew/.linuxbrew/Cellar/icu4c/74.2/lib/libicudata.so.74 (0x00007fe4e2000000)

Which causes breakages whenever it's updated.

[SMillerDev]

Good point, but then what is the warning asking me to do?

To declare the dependencies you have instead of hoping they will always be available with your current set of dependencies.

Also, icu4c is particularly annoying because it's linking with a shared library in a cellar.

It's also a versioned library so it'll almost always need a version bump when it's updated

[ipatch]

i came across this same error / warning with the homebrew-freecad tap this afternoon (snippets & links below). however the build process from test-bot still seemed to continue and upload a bottle'd artifact. so not sure if these are just warnings or if it will cause the app to crash at runtime, as i haven't tested the bottle'd artifact manually but the tests via the ci seemed to pass.

interestingly enough the indirect linkage is described as both a warning and failure so not sure of the severity here 🤷‍♂️

==> Testable dependents:

Warning: 1 failed step ignored!
brew linkage --cached --test --strict freecad/freecad/[email protected]_py310

https://github.com/FreeCAD/homebrew-freecad/actions/runs/9177500809/job/25235238110?pr=512#step:18:70

snippet grabbed from the above link.

==> brew linkage --test freecad/freecad/[email protected]_py310
==> brew linkage --cached --test --strict freecad/freecad/[email protected]_py310
==> FAILED
Full linkage --cached --test --strict freecad/freecad/[email protected]_py310 output
  Indirect dependencies with linkage:
    llvm
Truncated linkage --cached --test --strict freecad/freecad/[email protected]_py310 output
  Warning: Indirect dependencies with linkage:
    llvm
  
==> brew linkage --cached freecad/freecad/[email protected]_py310
==> brew install --only-dependencies --include-test freecad/freecad/[email protected]_py310
==> brew test --verbose freecad/freecad/[email protected]_py310

---

after your ci run went through test-bot step. did the workflow / run still manage to upload a bottle'd artifact for your repo / tap?

[ipatch]

well unfortunately with my use case the indirect linkage was to the upstream llvm formula, but if i declare llvm (latest version, ie. llvm.rb) directly in the [email protected]_py310.rb formula file then the build will fail. for pyside2 i have to use llvm@15 as i believe homebrew-core also uses llvm@15 for pyside2. i haven't checked ci run logs for homebrew-core to see if the indirect linkage warning is present as well, but it seems there could be some edge cases that would crop up and be problematic if this warning became and error and prevented the building of bottle'd artifacts. just my 2 cents though.

[Bo98]

If Homebrew is reporting it then otool/ldd is likely reporting that pyside2 is actually using llvm without you knowing.

If you do see it being incorrect let me know and I'll have a look. I'll also have a look at homebrew-core's pyside2.

[mahrud]

In my case, I know what things from gcc and icu4c are being used, but they are so tiny and incidental that it would be significantly more convenient if those small pieces were statically linked so that users didn't need to download gcc. Also, since it's a self-managed tap, rebuilding bottles every time icu4c updates has been a big hassle.

[Bo98]

Nothing has changed in Homebrew in that regards. Your formula was always indirectly linking to icu4c - it just wasn't telling you before.

How exactly that is addressed totally depends on your build script. I suspect Boost::regex's CMake script is inserting this in: boostorg/boost_install#47. You might be add a STATIC IMPORTED icudata CMake target etc that points to the static library. The default behaviour otherwise will be -licudata.

(The gcc one on Linux is a Homebrew bug however that has already been fixed.)

[PatTheMav]

I've ran across the same warnings just now (had not updated my custom taps in some time) and in this specific case it's FFmpeg which explicitly depends on libvorbis and theora, both of which depend on libogg which is somehow added as a direct dependency to FFmpeg itself.

Knowing FFmpeg I would not be surprised if this is the result of pkg-config information and the output for theora indeed includes ogg:

-L/opt/homebrew/Cellar/theora/1.1.1/lib -L/opt/homebrew/Cellar/libogg/1.3.5/lib -ltheora -logg

So is this something that I can fix as a formula author or this something that needs to be fixed in the theora formula (or, as a third possibility, I misunderstand the indirect linking issue)?

To wit: The Linux build of FFmpeg complains about libx11, libxcb, and libxext being linked indirectly.

[ipatch]

for reference / posterity purposes, snippet copied from the raw logs.

https://github.com/Macaulay2/homebrew-tap/actions/runs/9087675907/job/24975839771#step:8:42397

2024-05-15T01:21:14.4744247Z Emacs Lisp files have been installed to:
2024-05-15T01:21:14.4745444Z   /home/linuxbrew/.linuxbrew/share/emacs/site-lisp/macaulay2
2024-05-15T01:21:14.4746497Z �[34m==>�[0m �[1mSummary�[0m
2024-05-15T01:21:14.8198829Z /home/linuxbrew/.linuxbrew/Cellar/macaulay2/1.24.05: 21,017 files, 520MB
2024-05-15T01:21:14.8212869Z �[34m==>�[0m �[1mRunning `brew cleanup macaulay2`...�[0m
2024-05-15T01:21:14.9134376Z �[34m==>�[0m �[1mbrew linkage --test macaulay2/tap/macaulay2�[0m
2024-05-15T01:21:16.0712854Z 
2024-05-15T01:21:16.0714300Z �[34m==>�[0m �[1mbrew linkage --cached --test --strict macaulay2/tap/macaulay2�[0m
2024-05-15T01:21:17.2205199Z Indirect dependencies with linkage:
2024-05-15T01:21:17.2205902Z   gcc
2024-05-15T01:21:17.2455802Z �[31m==>�[0m �[1m�[31mFAILED�[0m�[0m
2024-05-15T01:21:17.2541769Z ##[group]Truncated linkage --cached --test --strict macaulay2/tap/macaulay2 output
2024-05-15T01:21:17.2553798Z ##[warning]Indirect dependencies with linkage:
  gcc