CI: caching: fix early termination expression check & cabal.project replacement

[Anton-Latukha]

Details are in the commit messages.

1. We not noticed the first specific of the CI, because the early cutoff behaved as it should. It was respected to return 'false' to terminate early, but that check always returned 'false'. When working in GHC 9.2 PR could not force the caching build to start, because check that I fix - always skipped the steps (always returned False). New check is done in the way how it is done in other if: ... != 'true' statements in the workflow.

2. The second specific is both Ubuntu-specific & maybe GitHub specific. Some OSes run alias cp='cp -i' & when I tried to rm manually - cabal.project appeared to be protected from simple rm by setinning (most probably) of a "sticky bit" on $HOME directory. I found that-out by working in GHC 9.2 PR (where index-state is actually different between .project files & noticed that CI cache key was with the same inferred index-state in all jobs). Now rm -f forced & that allows removing file with a sticky bit. log reports the situation of files changes there: https://github.com/haskell/haskell-language-server/runs/4597911492?check_suite_focus=true#step:14:460

[Anton-Latukha]

Ok. Now lets actually check the early termination & after that the new key (should go into full run).

[Anton-Latukha]

Early termination: https://github.com/haskell/haskell-language-server/runs/4599033379?check_suite_focus=true

[Anton-Latukha]

I advanced the index-state.

Which changes the key. So the old key of cache & new id for cache of the workflow does not align & so old storage is loaded, but the full build&cache process starts & then saves into the new key.

Resulting into the proper long caching run: https://github.com/haskell/haskell-language-server/runs/4599160496?check_suite_focus=true

[Anton-Latukha]

Ok, now I need to wait until this run completes & then drop the modifications. & have a 1 clean run on {test,bench}, it is honest to do at least because I changed test workflow.

[Anton-Latukha]

Ok, now run is complete https://github.com/haskell/haskell-language-server/pull/2520/checks.

By dropping the WIP commit - disabling caching & enabling {test,bench}.

[Anton-Latukha]

CI run completed successfully: https://github.com/haskell/haskell-language-server/runs/4599610364?check_suite_focus=true.

[Anton-Latukha]

Seems I checked everything about early termination this time.

& seem that cabal.project switcharoo now works.

I hope we solved the main problems there & hope that soon we would exhaust hidden from us bugs in the CI to become sure in configuration robustness.

[jneira]

lgtm, good catches, thanks

[Anton-Latukha]

Would note, that now because according .project are loaded & tested under - CI pipeline would ship according context & so uncover some related to that effects/bugs inside pull request processes.