TST: Update test tolerances for MPL 3.5 #1938
Conversation
Matplotlib 3.5 caused some images to change slightly, so increase the tolerances on those tests since they were minor.
|
Do not merge yet, I pushed this up to see if CI was able to run with 3.5 yet or not. (Apparently the answer is no) |
|
Yeah, I'm currently working on getting matplotlib 3.5 builds on conda-forge (just waiting on some new fonttools packages across the architectures since matplotlib added that dependency). Then I'll work on getting the testing builds for 3.5 up. |
This update the contour test limits due to changed contouring with closed paths in Matplotlib 3.5. Additionally updates the Continuous Integration package to make sure we get proj>=8 and add color to the output.
|
OK, I think the failures now are due to conda inconsistencies with packages. Specifically having to do with installing the |
Don't run on forks of the repo.
|
@greglucas It turns out to be an old bug in |
|
Thanks for looking into that @dopplershift! It looks like that did it :) |
Matplotlib 3.5 caused some images to change slightly, so increase the tolerances on those tests since they were minor.
The three pcolormesh tests are all using the same image and looks to be very minor with a new tolerance update acceptable.
The
multiple_projections_520test appears to have had the entire axes stretched up (or down, I can't tell which image is which here) ever so slightly. So, the boundaries and line all are shifted a pixel or two in the new result image. A tolerance increase is probably acceptable again.The contours is a little more troubling. It looks like the contours now follow the boundary to close on each other and get smeared on the far right edge, whereas that wasn't the case before. @QuLogic, I know you had to update some of the contour open/closed paths on the Matplotlib side, so perhaps you have some insight on this one.
