cc @fabianschuiki

One confusing point is that the Github CI can pass this test without path substitution and the additional PATH variable, which seems to be an existing CIRCT binary path in the PATH.
https://github.com/llvm/circt/actions/runs/11998261862/job/33447711728 (this is the CI job for #7886)

Hmmm, this is interesting. The circt-test-runner-*.py scripts are installed alongside the circt-test executable. But you're right, if circt-test is expanded by lit instead of found through the PATH, then the scripts, which are searched for in the PATH, won't be found either. Can we tweak lit to add the binary build directory to the PATH, such that we don't have to muck with the PATH variable in RUN lines?

Ah sorry, I just now realized that this is for circt-opt, not the benchmark script. In the lit config, it should be possible to add the CIRCT tool build path to the config.path list, which will be used to populate the PATH variable. That might fix things.

Ah sorry, I just now realized that this is for circt-opt, not the benchmark script. In the lit config, it should be possible to add the CIRCT tool build path to the config.path list, which will be used to populate the PATH variable. That might fix things.

Yeah, it's a possible fix that I wasn't sure made sense before. Maybe we should just consider it a workaround to work with the Python script rather than using it to replace the substitution flow.

Would you like to merge this yourself, or do you want me to do it?

Would you like to merge this yourself, or do you want me to do it?

I'd be really grateful if you could help me merge them for now. Thanks so much!

Oh..sorry, my wrong rebase screwed everything up!