Hello there!

Could you run clang-format -i /__w/wild/wild/wild/tests/sources/*.{c,cc,h} so the CI shows the real issue?

I'm also not sure if it should be using --no-gc-sections

Yeah, Wild unlike other linkers performs sections GC by default. For this test we need to disable it, to make sure the section is discarded even without GC.
I can look into it tomorrow.

Could you run clang-format

Done.

I've also pushed forcing --no-gc-sections and the more minimal test structure David suggested. I imagine turning it into a shared object would also remove the need to set up the runtime, have a _start symbol and some .text, etc?

Yep, if you wanted to go the shared object route, I think you'd add -shared as a link argument, then add //#RunEnabled:false. You could also move all the test directives into the .s file and delete the .c file. There are some other shared-object-only tests if you need something to reference. e.g. dynamic-bss-only.c. I'm not sure there are any others that are assembly-only, so that might be a first.

I think it might also be possible to remove the Arch line, since I think the assembly you've got is likely portable to other architectures as it doesn't actually use any assembly instructions.

The closest assembly only test would be wild/tests/sources/basic-comdat-1.s (found using rg "//#" wild/tests/sources/*.s).

Thanks! I've adapted the test to copy basic-comdat.s's structure, which looked very much like exactly what was needed.

section.got
wild Section missing
ld OK

Looks like GNU ld is generating a GOT section on aarch64. You could add //#DiffIgnore:section.got

Done. Hopefully this will pass CI now, but it doesn't like me enough to run automatically.

(thanks mati :)

Done. Hopefully this will pass CI now, but it doesn't like me enough to run automatically.

Now that you've had a PR merged, I think any subsequent PRs won't need approval for actions to run.

Thanks for your contribution!