Going to cherry-pick the bugfix from #356 in this PR and see if that makes the failing Linux tests pass. If it does, then we should merge #356 first, then this one after #356 is successfully merged.

Test Results

       8 files  +  4     416 suites  +4   2h 45m 57s ⏱️ +11s
   885 tests +  2     862 ✔️ +  2    23 💤 ±0  0 ❌ ±0 
4 052 runs  +12  3 918 ✔️ +12  134 💤 ±0  0 ❌ ±0 

Results for commit e857fae. ± Comparison against base commit a6a7041.

SaveAndLoad_10KRecords_CompletesQuickly

Main_NoConflictFileB4_ConflictsEncountered_HaveConflictFileAfter
Main_UnhandledMergeFailure_Returns1
Main_Utf8FilePaths_FileNamesOk

♻️ This comment has been updated with latest results.

Okay, now the tests are passing but the dotnet test run is still exiting with exit code 1 for some reason?

2024-10-29T11:46:47.1463558Z Passed!  - Failed:     0, Passed:   966, Skipped:    20, Total:   986, Duration: 25 m 49 s - LibChorus.Tests.dll (net6.0)
2024-10-29T11:46:47.2241280Z ##[error]Process completed with exit code 1.

I'll investigate why dotnet test doesn't like what we're doing. But it's probably related to running the tests in parallel, because the same code passes in PR #356.

Looks like this might be running into dotnet/sdk#29742 where running dotnet test --no-build after dotnet build is failing on ubuntu-latest runners. As a workaround, I'm going to try just running dotnet test without the --no-build parameter; this will build twice, but it should then stop complaining about DLLs being "invalid arguments" (?).

Talked about this with Robin, the issue is that Chorus.Tests only run against net461, so when we say dotnet test -f net8.0 it fails because Chorus.Tests does not support net8.0. To fix this we will split out our dotnet test command to be explicit about which test project we are running, then we can skip Chorus.Tests when we are not running our net461 tests. As for the tests filter which is why we split linux and windows, we can turn that into a github action expression, this will let us avoid having 6 commands, 3 per project and 2 per os.

As for the --no-build issue, it doesn't look like that's effecting us in our other tests, so I assume it's fine here.

Turns out we don't actually need the test filter to be different between Linux and Windows anymore, since we're skipping running net461 tests on Linux, and all the tests that needed the filter are net461 (and not likely to change since they use Windows Forms). So I chose to keep it simple and not have any filter differences (and just skip the SkipOnBuildServer tests everywhere).

nice work, though it looks like the Build Windows installers job is now failing.

Also it looks like no tests are actually running for net461, so I'm going to look into that