Use getLocalBuildInfoM not getTestEnv

[philderbeast]

Fixes #9725 using getLocalBuildInfoM similar to the way it is done in the following snippet;

cabal/cabal-testsuite/PackageTests/ExtraCompilationArtifacts/test.hs

Lines 14 to 19 in 2acae63

	lbi <- getLocalBuildInfoM
	let installedLibPath = libdir $ absoluteInstallDirs (localPkgDescr lbi) lbi NoCopyDest
	shouldExist $ installedLibPath </> "extra-compilation-artifacts" </> "ghc-plugin-X" </> "data-dir" </> "content-A.txt"
	shouldExist $ installedLibPath </> "extra-compilation-artifacts" </> "ghc-plugin-X" </> "data-dir" </> "content-B.txt"
	shouldExist $ installedLibPath </> "extra-compilation-artifacts" </> "ghc-plugin-Y" </> "content-Y.txt"

The previous getTestEnv method was expecting a path that didn't include the ABI hash suffix.

When the test runs it logs the library path (unchanged by this fix);

library-dirs:
    /.../CopyHie/setup.cabal.dist/usr/lib/x86_64-linux-ghc-9.8.1-eb95/hie-local-0.1.0.0

Include the following checklist in your PR:

• Patches conform to the coding conventions.
• Is this a PR that fixes CI? If so, it will need to be backported to older cabal release branches (ask maintainers for directions).

[alt-romes]

Thanks @philderbeast, but working around this issue in this particular test doesn't seem to be the right solution. Shouldn't getTestEnv and the functions it depends on be updated to properly account for the abi tag?

[mpickering]

If we look at testLibInstallDir then it seems that this should already account for the ABI tag so it seems we should investigate why that produces the wrong result.

[mpickering]

I can reproduce this, it seems the issue is when you build the tests with 9.8.1.

How testCompiler is set should really depend on the version of the Cabal library the test is using, this is the reason for the mismatch because the path is computed using Cabal-3.10 but the test itself uses Cabal-3.11. Will be improved by my other patch which allows you to run the testsuite against different CAbal library versions.

[philderbeast]

Thanks for taking a look @mpickering.

• getLocalBuildInfoM uses getTestEnv but it looks like TestEnv doesn't have the data read from PackageTests/CopyHie/setup.cabal.dist/work/dist/setup-config
• in testLibInstallDir, showCompileIdWithAbi is taking the NoAbiTag branch because that's what ScriptEnv0.hs has
• dist-newstyle/build/x86_64-linux/ghc-9.8.1/cabal-testsuite-3/build/global-autogen/Test/Cabal/ScriptEnv0.hs is where the compilerAbiTag is read from as a field of Compiler, isn't it?

cabal/cabal-testsuite/src/Test/Cabal/Prelude.hs

Lines 99 to 102 in 2acae63

	getLocalBuildInfoM :: TestM LocalBuildInfo
	getLocalBuildInfoM = do
	env <- getTestEnv
	liftIO $ getPersistBuildConfig (testDistDir env)

cabal/cabal-testsuite/src/Test/Cabal/Monad.hs

Lines 560 to 568 in 2acae63

	-- | The absolute path where library installs go.
	testLibInstallDir :: TestEnv -> FilePath
	testLibInstallDir env = libDir </> compilerDir
	where
	platform@(Platform _ os) = testPlatform env
	libDir = case os of
	Windows -> testPrefixDir env
	_ -> testPrefixDir env </> "lib"
	compilerDir = prettyShow platform ++ "-" ++ showCompilerIdWithAbi (testCompiler env)

[philderbeast]

I'm going to park this for now. Feel free to close it if you feel it is the wrong approach.

[Mikolaj]

Let me take the liberty of turning it into a draft so that next time I do a sweep, I don't panic it's bitrotting.