attn @mdboom @pelson

OK, I changed it to be a subclass of TransformedPath.

One more minor change; I forgot to re-cache the updated path when it changed.

I'm 👍 on this change @QuLogic. Just out of curiosity, have you had a look at how this impacts Cartopy, which to my knowledge is the biggest user of matplotlib's non-affine transforms? Also, it may be that the example in https://github.com/SciTools/cartopy/blob/master/lib/cartopy/examples/favicon.py#L19 no longer needs to recompute the clip path with this change.

It's certainly possible. Cartopy seems to have 2 or 3 "background" patches though, so I wasn't sure which one to use.

So "proposed next" is after 1.5, right?

No, 2.1

On Sun, Sep 13, 2015, 14:42 Elliott Sales de Andrade <
notifications@github.com> wrote:

So "proposed next" is after 1.5, right?

—
Reply to this email directly or view it on GitHub
#4920 (comment)
.

👍 from me, but it should have a test, ideally something that tests whether the clipping path updates when the transform does.

Test added for both the new TransformedPatchPath and the older TransformedPath which didn't really have one directly.

Thanks. I'm happy to merge this, but first have @QuLogic and/or @pelson answered the question about any ill effects on cartopy?

Unfortunately, it is difficult to say with any certainty, because Cartopy's tests for MPL integration do not pass with 1.4+. But at the very least, nothing has broken.

After updating Cartopy's tests to work with 1.5, it seems that there is one test with a failure, though it is small:

When the tolerance is turned down in matplotlib, a few more test images here might fail too if the same thing happens. It seems to be a rounding error at half-pixels. If I make the same plot interactively and resize, it alternatively comes and goes. But I'm not sure why, since the calculation should be about the same between the two.

👍 - thanks @QuLogic
Would you mind adding another PR which puts a whats new entry in the docs.

Thanks!