Fix test reporting

[allanrenucci]

When running the tests, we were reporting the number of errors
encountered instead of the number of failed tests. For neg tests, errors
are not test failures

[Blaisorblade]

errorCount seems a trap, and reasonsForFailure seems to fall into it even if it's also called for neg tests by checkExpectedErrors.
Neg test shouldn't have failed, but did. Reasons: - encountered ${test.errorCount} error(s).

[Blaisorblade]

The original change made sense, in fact this errorCount field was a misfeature, so I went further and removed it. @allanrenucci can you review that?

[allanrenucci]

I would change error(s) for failure(s)

[allanrenucci]

@Blaisorblade I am not sure about the second part of this PR. I think errorCount is useful is some cases. E.g. compiling a neg file and failing because the compiler emits 2 errors instead of 3.

I would suggest to open a new PR with your changes

[Blaisorblade]

I am not sure about the second part of this PR. I think errorCount is useful is some cases. E.g. compiling a neg file and failing because the compiler emits 2 errors instead of 3.

That still works, and it's still visible in the meta test, in the output line

Wrong number of errors encountered when compiling out/VulpixMetaTests/neg/missing-error-annotation, expected: 0, actual: 2

The code you talk about is not affected — 565a2e5 changes some other behavior, the rest is purely meaning-preserving dead code elimination.

The behavior you describe involves another field named errorCount — see https://github.com/dotty-staging/dotty/blob/29f47602ee0d94bc0c89deab126dad32950316c0/compiler/test/dotty/tools/vulpix/ParallelTesting.scala#L754-L759 for the code involving the other errorCount, and see https://github.com/dotty-staging/dotty/blob/29f47602ee0d94bc0c89deab126dad32950316c0/compiler/test/dotty/tools/vulpix/ParallelTesting.scala#L714 to see where // error appears to be actually handled.

FWIW, I don't have a problem with splitting the other commits if there is plausible reason at all, this just isn't one.

[allanrenucci]

FWIW, I don't have a problem with splitting the other commits if there is plausible reason at all, this just isn't one

The reason is that it's safe to merge the first commit and I'd like to experiment with the other commits to convince myself that it is indeed dead code

[Blaisorblade]

The current PR was tested in http://dotty-ci.epfl.ch/lampepfl/dotty/4656, so merging.