Feat: better clang support

[Noir01]

fixes #308

[Copilot]

Pull Request Overview

This PR improves Clang compiler compatibility for the Gibbon runtime system by adding compiler-agnostic macro wrappers and toolchain detection. The changes address issue #308 by ensuring the codebase can compile cleanly with both GCC and Clang.

Key changes:

• Introduces cross-compiler pragma macros (GIB_PRAGMA_MESSAGE, GIB_PRAGMA_UNROLL) to handle compiler-specific syntax differences
• Adds automatic detection of Clang compiler and selection of appropriate archiver tool (llvm-ar vs gcc-ar)
• Replaces GCC-specific inline assembly syntax with portable __asm__ __volatile__ form

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
gibbon-rts/rts-c/gibbon_rts.h	Adds pragma wrapper macros and updates inline assembly syntax for cross-compiler compatibility
gibbon-rts/rts-c/gibbon_rts.c	Replaces direct pragma usage with new wrapper macros throughout the runtime
gibbon-rts/Makefile	Adds -Werror flag and uthash include path configuration
gibbon-compiler/src/Gibbon/HaskellFrontend.hs	Removes redundant CPP conditionals around import statement
gibbon-compiler/src/Gibbon/Compiler.hs	Implements Clang detection and automatic archiver selection for build process

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ulysses4ever]

It's be wise to set up a test that runs in CI.

[Noir01]

It's be wise to set up a test that runs in CI.

What kind of tests should I add?

[vidsinghal]

It's be wise to set up a test that runs in CI.

What kind of tests should I add?

I believe all the tests run with gcc atm.
In gibbon-test-examples and the Benchrunner, you can add a gcc mode and a clang mode.
Run all tests with clang by changing the compile/run command and one with gcc.

[ulysses4ever]

You can set up a smoke test separately from everything else (e. g. a separate ci workflow but maybe not, needs thinking). It could compile some simple program with clang.

What Vidush suggests is bonus points (but sounds great!).

[Noir01]

I did what @vidsinghal suggested and now our tests run twice, once for gcc and once for clang. I also rebased my branch.

[vidsinghal]

@Noir01 can you rebase this with fixing_indirections? Tests should pass now

[vidsinghal]

builds failing

[vidsinghal]

@noir tests aren't setup properly?

    < [compiler] clang detected but llvm-ar not found; using '/usr/bin/ar' instead.
    ```

[Noir01]

I just looked at the logs, and what I guess is happening is that I didn't add llvm-ar in the workflow image, so every time it runs the tests and compares the output, the warning we added for different major versions of clang and llvm-ar triggers which is tripping up the simple diff tests.

[vidsinghal]

I just looked at the logs, and what I guess is happening is that I didn't add llvm-ar in the workflow image, so every time it runs the tests and compares the output, the warning we added for different major versions of clang and llvm-ar triggers which is tripping up the simple diff tests.

we need to update the workflow image to include llvm-ar

[ulysses4ever]

Regarding hPutStrLn stderr: I thought Gibbon already had some messaging infrastructure so that we have control over how things get printed.

[Noir01]

Regarding hPutStrLn stderr: I thought Gibbon already had some messaging infrastructure so that we have control over how things get printed.

Oh, I wasn't aware of that. How do I use the existing infrastructure?

[ulysses4ever]

I'm not sure there is. I might have mixed that up with err:

gibbon/gibbon-compiler/src/Gibbon/Common.hs

Line 366 in 40d511e

err = error

It's a good idea to grep for stderr to check.

[Noir01]

I did rg "stderr" and on a cursory glance, it doesn't seem like any infrastructure exists yet.

Also, some of the tests are failing with segfaults for the clang workflows. I will try to debug them tonight/tomorrow morning.

[vidsinghal]

I did rg "stderr" and on a cursory glance, it doesn't seem like any infrastructure exists yet.

Also, some of the tests are failing with segfaults for the clang workflows. I will try to debug them tonight/tomorrow morning.

oh wow, this is interesting, i would be interested to know why, if you can't figure out lmk, i might try to investigate. Otherwise we will have to mark these failing for Clang.