Remove test times when running shtest-keyword-parse-errors test,
in order to prevent the previous executions from impacting subtest
order and therefore causing FileCheck to fail.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
Other than a tiny nit, LGTM.
llvm/utils/lit/tests/shtest-keyword-parse-errors.py | ||
---|---|---|
2 | @lebedev.ri has been marking these lines with a FIXME: ef4b3a4571e2. Perhaps it's good to do that consistently until someone implements a more comprehensive fix (like putting the cleanup or some command-line option in %{lit}). |
llvm/utils/lit/tests/shtest-keyword-parse-errors.py | ||
---|---|---|
2 | Will do, thanks! |
You're welcome. I'd love to try to figure out how to make the tests not depend on subtest order but my TODO is too long already.
Yes, it's not my highest priority either.
I think some sort of command-line option to disable this behavior is the right solution, and it can be embedded in %{lit} so that no test is affected accidentally. (I'd like to use that in clang/test/utils as well.) For tests that specifically check this behavior (reorder.py), an alternate substitution (e.g., %{lit-reordered}) can be provided.
Indeed. In fact, the first thing I did was to see --help if there's a way to disable these files. To be honest, I don't really like the idea of lit leaving cache files all over the place.
I think you're saying that users of lit (as opposed to maintainers of lit's test suite) might like such a command-line option too.
Ok, I've decided to spend 10 minutes and implement a lexical test order option: D107695.
@lebedev.ri has been marking these lines with a FIXME: ef4b3a4571e2.
Perhaps it's good to do that consistently until someone implements a more comprehensive fix (like putting the cleanup or some command-line option in %{lit}).