Closed martinfleis closed 6 months ago
Once everyone is happy with the approach I took here, I'll mirror it to branca.
I'm OK with this approach but I'm afraid our tests are not robust enough to catch any failures. The way we use numpy here is unchecked and will produce passing status even when it is wrong. As you confirmed in https://github.com/python-visualization/branca/pull/163.
Not sure if I would run the tests a second time. I'd think that makes the output harder to understand. Or is this a temporary addition? In that case it doesn't matter as much.
Not sure what your suggestion here is then. Split it into two separate runs? Generally, it is better to test against nightly versions of our dependencies, assuming the test will catch issues, so I would try to keep it as permanent. In other projects I maintain, we have dev as another environment alongside the others (latest, oldest...)
@martinfleis I can add this later but, if you want to tackle this now, let's add an experimental label to that matrix item to make it easier to find in the logs. I did something like that in https://github.com/Unidata/netcdf4-python/pull/1319/files#diff-83620a0d7a95a32e599b5b5b89eabf1419cb0e3ffb5212d15f87f5bd3a452730R17
Looks good, let's do it!
Closes #1937