testsuite: Add test for #9334

[mpickering]

This adds a test which checks that v2-run works with profiling.

cabal v2-run --enable-profiling Main.hs

which is the same one as is run in ghcup-ci.

This test fails on recent GHC versions on GHCUp CI so hopefully also on cabal CI.

https://github.com/haskell/ghcup-metadata/actions/runs/6369779959/job/17290326630#step:3:15214

[andreabedini]

👍

[ulysses4ever]

This test fails on recent GHC versions on GHCUp CI so hopefully also on cabal CI.

But it doesn't fail on our CI? I'm confused...

[mpickering]

This test fails on recent GHC versions on GHCUp CI so hopefully also on cabal CI.

But it doesn't fail on our CI? I'm confused...

I don't know.. it might only fail with 9.6.3 (which is not currently tested).

[mpickering]

I rebased onto master, where 9.6.3 tests are run on windows, so we will perhaps see if there is a problem there!

[ulysses4ever]

No luck so far: windows-9.6.3 succeeded

[gbaz]

One wild guess.

The failing line is cabal --verbose=0 run --enable-profiling ./main.hs -- +RTS -s

I don't think the verbosity flag is the problem, and in sufficiently recent cabals run is the same as -run, but note that the faling line A) miscapitalizes main.hs, and B) calls it with a "./" prefix. I wonder if either of these is the cause of the failure?

[mpickering]

I tried to recreate that situation in the test. The file I added is main.hs (like the ghcup CI script) and also call it in the same way (with the ./ prefix).

I will further modify the test to look for Main.p_o .. because I imagine that actually main.p_o is getting created. It's also possible this is fixed on master, and broken on 3.10 branch.

[gbaz]

hrm, maybe we should try to diagnose also by prs against ghcup's ci that tweaks the calls a bit? its very hard to see what might be happening. are they using the same or different windows images?

[mpickering]

I have added a second unit test which uses randomised testing to check the filepaths are valid.

[mpickering]

I deliberately rewrote my code away from the hlint suggestion so it was nicer to document the clauses. I don't intend to change it back.

[mpickering]

The test passes on windows, because % is in fact a valid character for a filepath. It just seems that some programs/contexts will interpret % as an environment variable marker. Do I augment my test with an ad-hoc check for % as well?

[ulysses4ever]

Re hlint: could you use {-# HLINT ignore "Use &&" #-} (or whatever the name of the suggestion, I haven't checked), so that the CI stays green, and you get your preferred style?

[ulysses4ever]

Do I augment my test with an ad-hoc check for % as well?

Yes, we live in a world where specs are not followed, and we have to acknowledge it, I think.

[Mikolaj]

@mpickering: would you like any help with that one?

[alt-romes]

Matthew thinks that this was ultimately not testing the right thing, so we are better off dropping the patch.