Use ruff_fix_validity to catch regressions in CI not detected by unit tests

[addisoncrump]

The fuzzers added in #4822 can be used to catch regressions for which there exists no unit tests (see: #4863 as an example of this; caught by ruff_fix_validity within ~30 seconds). PRs should include a short fuzz run (5-10m) before merging. Failure cases can then be added to the unit test suite.

Before that happens, the following issues (discovered by fuzzing currently and will cause this CI step to fail) should be resolved:

• [x] #4823
• [x] #4825
• [x] #4826
• [x] #4827
• [x] #4828
• [x] #4863
• [x] #4865
• [x] #4892
• [x] #4897
• [x] #4899
• [x] #4901
• [ ] #4924
• [x] #4925
• [x] #4973
• [x] #4975
• [x] #4991
• [ ] https://github.com/RustPython/Parser/issues/89
• [x] #5004
• [x] #5154
• [x] #5156
• [x] #6213
• [x] #6214
• [x] #9374
• [x] #9379

This change should include the following:

• Caching of the fuzzer corpus, fuzz/corpus/ruff_fix_validity (~20MB)
• Initialisation of the fuzzer corpus if not present (I will provide an initial corpus)
• Execution of timeout 10m cargo fuzz run -s none ruff_fix_validity -- -timeout=5 || [[ $? -eq 124 ]] after the fuzz build step

In the future, when parser idempotency is more stable or new fuzzers exist (such as for formatting), this should be included as well.

[addisoncrump]

I am happy to craft these changes. :slightly_smiling_face: I just want to make sure that this implementation is sane for the maintainers.

[T-256]

• I think this issued could be close since https://github.com/RustPython/Parser/issues/89 is solved by Ruff's internal parser at some point and https://github.com/astral-sh/ruff/issues/4924 is the only remained item which can tracking it separately.

[addisoncrump]

The list here is now very old. There are likely new bugs caught by this beyond what is tracked; the issue is about making sure that the fuzzer is not enabled constantly unnecessarily.

[T-256]

Correct, sorry for misreading discription.

PRs should include a short fuzz run (5-10m) before merging. Failure cases can then be added to the unit test suite.

I think it's better to use cron jobs workflow instead.

[addisoncrump]

As in, a scheduled workflow? Makes sense.