> assert cb.ax.get_ylim() == (1e-4, 1e6)
190
E assert (9.9999999999...05, 1000000.0) == (0.0001, 1000000.0)
191
E At index 0 diff: 9.999999999999999e-05 != 0.0001
192
E Full diff:
193
E - (0.0001, 1000000.0)
194
E + (9.999999999999999e-05, 1000000.0)
Is this something we should investigate for 3.6.1? The change becomes user-facing in existing versions of matplotlib when a new version of the responsible dependency gets released. If we need to fix this on our side, it is desirable to have a fixed version already out there to minimize user-impact.
Activity
tacaswell commented on Oct 2, 2022
This does look real:
timhoffm commented on Oct 6, 2022
Is this something we should investigate for 3.6.1? The change becomes user-facing in existing versions of matplotlib when a new version of the responsible dependency gets released. If we need to fix this on our side, it is desirable to have a fixed version already out there to minimize user-impact.
jklymak commented on Oct 6, 2022
Add some floating point tolerances to these tests so they don't fail is probably OK.
TST: Increase fp tolerance on more tests for new NumPy
TST: Increase fp tolerance on more tests for new NumPy
TST: Increase fp tolerance on more tests for new NumPy
TST: Increase fp tolerance on more tests for new NumPy
TST: Increase fp tolerance on more tests for new NumPy