Closed ilevkivskyi closed 3 weeks ago
According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅
Nice, thanks for fixing!
By the way, there's another similar follow imports related discrepancy in daemon https://github.com/python/mypy/issues/15677
Fixes https://github.com/python/mypy/issues/18019 Fixes https://github.com/python/mypy/issues/17775
These two are essentially variations of the same thing. Instead of adding e.g.
types
toSENSITIVE_INTERNAL_MODULES
(which would be fragile and re-introduce same crashes whenever we add a new "core" module) I add all stdlib modules. The only scenario when stdlib changes is when a version of mypy changes, and in this case the daemon will be (or should be) restarted anyway.While adding tests for these I noticed a discrepancy in
--follow-imports=normal
in the daemon: the files explicitly added on the command line should be always treated as changed, since otherwise we will not detect errors if a file was removed from command line in an intermediate run.Finally the tests also discovered a spurious error when cache is disabled (via
/dev/null
).