Hypothesis: Add tests for rotate()

[nicoonoclaste]

Based off #62
Thanks to @astronouth7303 for writing the initial version of the test.

@pathunstrom If you confirm that the issue uncovered by Hypothesis is a bug, I can add commits fixing it to this branch.

[nicoonoclaste]

@astronouth7303 @pathunstrom Actually, Hypothesis finds some annoying numerical instabilities even without the rounding (due to the way we compute rotate and angle).
Would it be reasonable to provide a more stable implementation in this PR, i.e. add tests that expose the problem (already done) then fix it?

[AstraLuma]

I would say create a separate PR for it. If you discover more bugs, this PR could become large.

[nicoonoclaste]

@astronouth7303 So, ship as-is and send another PR to fix the bug exposed by the test?

[AstraLuma]

In the other order, yes.

[nicoonoclaste]

@astronouth7303 Weird, but OK.

[nicoonoclaste]

#67 contains the first 2 commits, which fix the bug exposed by Hypothesis.

[nicoonoclaste]

Had to pull in #69: 5c6c1a9 wasn't a correct test of linearity -- it didn't test multiplication by a scalar -- and the corrected test had overflow issues with large scalars and large vectors.

[AstraLuma]

The blocker for this right now is that I've found a clear failure of the linearity test.

[nicoonoclaste]

@astronouth7303 I rebased and cleaned up the history a bit (the resulting diff is equivalent, though).

Also, I noticed that the comment on the example you provide says “Rotation must not be an multiple of 90deg”, while you set angle=45.

[AstraLuma]

Yes, in order to trigger the test case, the angle must not be axis-aligned (0, 90, 180, 270, etc). So the demo data uses 45deg as the sample.

I listed out the rules of how to trigger the bug, as best as I could discover them.

[nicoonoclaste]

Oh, yes, I'm very good at reading and missed the “not” there. Sorry

[nicoonoclaste]

Finally fixed <3

[AstraLuma]

Sweet! Is there anything else that needs to happen?

[nicoonoclaste]

@astronouth7303 Not for this branch? Tests pass :D