Failing CI on MacOS/Windows?

[susliko]

Haven't found a button to run the workflow manually, hence this PR
I suspect that CI is failing for MacOS and Windows

[susliko]

Indeed, it fails
I feel like test jobs for MacOS and Windows operate on some old version of grammar. Particulary, they fail to parse structural types in extend clauses brought by #285

cc @eed3si9n @ckipp01

[susliko]

Gotcha: the phase that generates parser out of the grammar is only run on Linux, and Windows/MacOS use the parser.c from the repo, which is stale

I think we should generate parser out of the grammar every time we run tests, regardless of the environment

[eed3si9n]

I think the bug is

tree-sitter-scala/.github/workflows/ci.yml

Lines 88 to 91 in cd3a29c

	- name: Parser tests
	if: ${{ needs.changedfiles.outputs.c }}
	shell: bash
	run: npm install && npm test

If I recall correctly, at some point @keynmol, @ckipp01, and I decided that running checks on Linux is good enough, except for the external scanner C code, which we run test on all 3 OSes.

[susliko]

If I recall correctly, at some point @keynmol, @ckipp01, and I decided that running checks on Linux is good enough, except for the external scanner C code, which we run test on all 3 OSes.

I think, running MacOS/Windows tests only when external scanner changes is okay as long as we run them using freshly generated parser

[eed3si9n]

as long as we run them using freshly generated parser

+1

[susliko]

Could you give me a hint on the purpose of the Check parser job? 🤔

[eed3si9n]

Could you give me a hint on the purpose of the Check parser job? 🤔

To prevent someone committing a malicious C code, we test that if C sources are checked in there should be no git diff output after running generate.

[susliko]

Is there a reason why it's disabled for Linux environment?

[eed3si9n]

Is there a reason why it's disabled for Linux environment?

When I first introduced it on 95c1437 it was Linux only, then it became non-Windows on fc77eb0, then should've been back to Linux only at 8aface5, but there was a typo on == vs !=, I think.

[susliko]

When I first introduced it on 95c1437 it was Linux only, then it became non-Windows on fc77eb0, then should've been back to Linux only at 8aface5, but there was a typo on == vs !=, I think.

The check fails for this PR, not sure why

[eed3si9n]

Closing this in favor of #288