`ghcide` test failures in Gentoo Linux

[hololeap]

I'm getting a few test failures for ghcide-1.8.0.0 when they are run with Gentoo's network-sandbox disabled (so unrelated to #3126):

• incomplete entries

  FAIL (31.99s)
       test/exe/Main.hs:1653:
       expected: ["AAAAA"]
        but got: ["AAAAA","AAAA","AAAA"]
       Use -p '/incomplete entries/' to rerun this test only.
  

• Map

  FAIL (33.01s)
       test/exe/Main.hs:1807:
       expected: ["Defined in 'Data.Map","Defined in 'Data.Map.Lazy","Defined in 'Data.Map.Strict"]
        but got: []
       Use -p '/Map/' to rerun this test only.
  

• Module is symlinked

  FAIL
     Exception: /var/tmp/portage/dev-haskell/ghcide-1.8.0.0/temp/extra-dir-12909185275211/other_loc/Sym.hs: createFileLink:createSymbolicLink: does not exist (No such file or directory)
     Use -p '/Module is symlinked/' to rerun this test only.
  

3 out of 344 tests failed (2492.75s)

---

Full build log is here (raw file is ~15MB)

---

Your environment

GHC 9.0.2 on Gentoo Linux. All packages are installed through the system package manager.

ghcide-1.8.0.0 is being built with the ghc-patched-unboxed-bytecode flag turned OFF. (I hit other test failures when it was turned on.)

---

Other gentoo-haskell-related test issues:

• Multiple test failures in Gentoo's network sandbox #3126
• hls-hlint-plugin test failure in Gentoo Linux #3217
• hls-gadt-plugin test failures in Gentoo Linux #3218

[pepeiborra]

The problem is that Tasty runs test in parallel by default, and most HLS test suites do not handle this well. Set TASTY_NUM_THREADS to 1

[michaelpj]

The problem is that Tasty runs test in parallel by default, and most HLS test suites do not handle this well.

Isn't the number of threads a tasty option? Could we set it explicitly in the test suites to 1 using adjustOption?

[pepeiborra]

The problem is that Tasty runs test in parallel by default, and most HLS test suites do not handle this well.

Isn't the number of threads a tasty option? Could we set it explicitly in the test suites to 1 using adjustOption?

That would be great

[michaelpj]

I made an issue for this: #3245

[michaelpj]

Did Pepe's suggestion help?

[hololeap]

It seems that the Map test fails consistently even with @pepeiborra's suggestion and also using cabal-v1/cabal-v2 and network-sandbox/no-network-sandbox. The others are not as consistent.

I want to try to get a build environment set up that mirrors the CI exactly, hopefully get all the tests passing, then work backwards and see what is different in Gentoo compared to the CI environment (package versions, etc.)