Bug fixes and speed up of replay function

[Louis-Pujol]

Hello,

I discovered some bugs in the previous implementation of the replay features, and fixed it.

Bugs were :

• mapping of indices between points before/after decimation was OK most of the time time but sometimes it failed : I write a function to tackle the issue of mapping points that are connected to the triangle mesh through edges but not involved in any triangle. Functional tests are in tests/test_map_isolated_points.py
• the mesh after replaying simplification wasn't exactly the same as after first simplification (vertices location slightly differed). There were no reason for this behavior and it was indeed an error in my implementation, I forgot to initialize the border attribute to the mesh. Now it is fixed, and test_replay pass with a tighter tolerance when comparing mesh after simplification and replay (indeed with the default tol of np.allclose).

In addition, I've optimized some parts of replay function where my previous implementation was too slow or memory inefficient. The interface didn't change.

It could be great if you can release this new version, I think it is now pretty stable. Perhaps some parts of replay function might be speeded up again, but for my use cases (< 1M vertices) I'm happy with the current running time.

[Louis-Pujol]

Hi, there was a problem for reading input arrays that were not C-contiguous (it was the case after a transpose for example).

I added a decorator to the functions simplify and replay to automatically call np.ascontiguousarray on non-contiguous arrays if needed.

I think the code is correct, the errror in the CI for building wheels on ubuntu seems to be related to Docker.

[akaszynski]

Looks like https://quay.io/pypa/manylinux2014_x86_64/ is down at the moment, see https://quay.io/ and https://status.quay.io/

[akaszynski]

Also, there are a handful of style fixes that should be resolved. Please see https://results.pre-commit.ci/run/github/434290916/1697214126.yTuVenK9REW8R9nk9A9q9g

[Louis-Pujol]

Thanks for the feedback. Now it seems that the line from pyvista import examples (used in test_replay.py) causes some trouble.

If I understand correctly the issue, we cannot import urllib3 because of a bad combination of openssl/python/urllib3 versions on the container that runs the tests (https://github.com/urllib3/urllib3/issues/2168). I tried to add sudo apt upgrade openssl to testing_and_deployment.yml but the issue remained the same. There's maybe a solution but I'm not skilled enough with CI to find a quick fix.

A simple workaround that made things work again was to skip the test that uses examples if from pyvista import examples raises an error.

[Louis-Pujol]

Thanks. I'm done with these changes, you can release this new version.