This PR showcases one use case (simplified here):

[
    da_haskell_test(
        name = "data-dependencies-{}".format(suffix),
        srcs = ["src/DA/Test/DataDependencies.hs"],
        args = test_args,
        tags = extra_tags,
    )
    for (suffix, extra_tags, test_args) in [
        # Test all LF versions on the main branch, after merging
        ("all", ["main-only"], []),
        # Don't test LF v2 when running the CI on a PR
        ("no-v2", ["pr-only"], ["--pattern !(v2)"]),
    ]
]

That is: I want to only run the non-v2 tests against PRs, and all tests against main. Ideally I'd say "if running against PR, then pass the --pattern !(v2)" flag to the test. But there is no way to express "if running against PR". Instead we can only say "define two tests: one with flag main-only and no pattern, and one with flag pr-only and some pattern". Without "pr-only" I would run the v2 tests twice on main.

I'm not sure how else to express this.

Without "pr-only" I would run the v2 tests twice on main.

Wouldn't caching make the second run instant? Or would Bazel consider them separate tests and thus not reuse the cache?

If I'm not mistaken they would be considered separate runs: it's the same binary, but with different flags: in one case the flag is "run all test cases", in the other case it's "run a subset of the test cases". But bazel is not aware of the semantics of the flags, so it doesn't know that one is the subset of the other.

What I could do is have two tests that run exclusive subsets instead I guess.