Ensure that Path.arc works for any full circle. #8994
Conversation
|
It's never that easy, is it... The failing test seems possibly both wrong and right. The top left circle is fully 0-380 now, instead of the expected 0-20 (this is debatably correct now.) The expected bottom centre circle was supposed to look like the bottom left; this is probably a precision issue as it's inputting an arc that's ~180 degrees wide. |
These trig bits were necessary to properly scale by major and minor axes for an ellipse, but this method only supports circular arcs.
Depending on the inputs, the range may get squashed down to ~0 when trying to remove extra 2pi revolutions. Fixes matplotlib#8992.
These are a bit unfortunate since they're mostly invisible.
| eta2 = np.arctan2(np.sin(theta2), np.cos(theta2)) | ||
| eta2 -= twopi * np.floor((eta2 - eta1) / twopi) | ||
| eta1 = theta1 | ||
| eta2 = theta2 - 360 * np.floor((theta2 - theta1) / 360) |
There was a problem hiding this comment.
There are a few edge cases where it's not the same, but I'm not sure if that's the reason. This is just a rearrangment of existing code changed from radians to degrees (because the trig is not necessary here.)
|
I forgot to comment after pushing, but I found that there's some trig that's unnecessary which produces some small floating-point inconsistencies. It's there to scale by ellipses major & minor axes, but |
|
Can you add a note to the docstring that this always draws from theta1 -> theta2 clockwise and winding at most once? I don't think this is a change of behavior, but there is some ambiguity as to what it should do (for example it could wind multiple times (which I don't think any one wants) or it could always connect them via the shortest route (which I also think no one wants)). |
|
That's correct; it's the existing behaviour, and I've added some lines explaining it. |
Depending on the inputs, the range may get squashed down to ~0 when trying to remove extra 2pi revolutions.
This condition is part of implementation linked in the docstring for
Path.arcbut is not included here for some reason.Fixes #8992.
PR Checklist