Over in pandas we use hypothesis seldomly in a pytest test suite, and I noticed that one of our hypothesis tests is consistently our slowest running test on a particular CI build
============================= slowest 30 durations =============================
8.37s call pandas/tests/io/sas/test_byteswap.py::test_int_byteswap[True-uint16]
6.40s teardown pandas/tests/io/parser/dtypes/test_dtypes_basic.py::test_dtype_all_columns[python-False-str]
...
Using AI to help profile test_int_byteswap, a significant amount of time is spent in constants_from_module trying to extract local constants. pandas does have few locally defined constants, but IMO they are not generally applicable to test in a hypothesis context. This feature request is echoing @oscarbenjamin's suggestion in #4364 (comment) to allow a flag to opt out of scanning the source tree for local constants.
Over in pandas we use hypothesis seldomly in a pytest test suite, and I noticed that one of our hypothesis tests is consistently our slowest running test on a particular CI build
Using AI to help profile
test_int_byteswap, a significant amount of time is spent inconstants_from_moduletrying to extract local constants. pandas does have few locally defined constants, but IMO they are not generally applicable to test in a hypothesis context. This feature request is echoing @oscarbenjamin's suggestion in #4364 (comment) to allow a flag to opt out of scanning the source tree for local constants.