I asked about this in the weekly meeting yesterday! There was mention that we keep Appveyor around because it tests conda, although I don't think anyone is enthusiastic about the frequent hangs.

A lot of the discussion went over my head though as I'm still new to the whole Python ecosystem.

I asked about this in the weekly meeting yesterday! There was mention that we keep Appveyor around because it tests conda, although I don't think anyone is enthusiastic about the frequent hangs.

I did port the conda test as well and now it looks like there must be some architectural difference (or running these tests now when we weren't before) because there are two minor 3d plot image comparison changes.

Not sure why the codecov/patch is reporting 0% diff, clicking into it does show the lines are being hit/missed same as before as expected.

Looking at the timings, all CI jobs finished in 23 minutes, versus ~33 minutes (or more) on the previous azure/appveyor versions.

Thanks for the review, I addressed all the comments and updated the docs to either remove azure/appveyor references or change to the more generic Windows CI.

Do we think that this will solve the general flakiness of AppVeyor timing out at the hour mark? Additionally, will this solve the problem where codecov generates false positives because it is missing a build?

I'm of the opinion that because it fails so much, the AppVeyor CI is something that we are all subconsciously ignoring.

I can review in more detail when I get back tonight; however, I don't know how valuable my review will be on most of this. That said, as long as we are properly testing conda, merging this PR has to be better than having a CI step that we ignore.

Not sure why the codecov/patch is reporting 0% diff, clicking into it does show the lines are being hit/missed same as before as expected.

It seems to have fixed itself with the rebases.

Do we think that this will solve the general flakiness of AppVeyor timing out at the hour mark? Additionally, will this solve the problem where codecov generates false positives because it is missing a build?

Well, this PR removes AppVeyor, so ... yes.

Well, this PR removes AppVeyor, so ... yes.

I… probably should have phrased that differently. I couldn't remember if the timeout issue was inherent to the Windows platform (and thus could show up in GitHub actions as well), inherent to conda, or something to do with AppVeyor itself.

Sorry, that was a bit snarky. I don't think we expect it to be Windows-specific, just that AppVeyor only has 1 core and a fixed time limit.