Improve PanZoom camera performance when non-+z direction is used

[os-gabe]

Fixes #1681 by keeping a MatrixTransformation that gets modified on each iteration rather than overwritten

[djhoese]

By removing the if statement aren't you now adding a penalty to the +z case when it isn't needed?

My other thought is, is it possible to modify either the rect or the main transform once whenever up is set and then apply that to all calculations from then on? Or would that require a Rect class that understands orientation? Just throwing ideas out there I haven't actually looked at the math very much.

[os-gabe]

I went back and forth on leaving the if statement and decided the computation was cheap enough that it was a good trade-off for simplifying the logic by removing the if statement.

I like where you are going with only computing the up transform whenever it changes but I'm not seeing a straightforward way to do that at the moment.

[djhoese]

I'm having a hard time being ok with the if statement being removed. Could you add it back. I can see how only creating the MatrixTransform once is a huge performance gain (it stops shaders from being recompiled every time), but the other calculations in there aren't nothing. Especially when you consider they aren't needed at all for +z.

I spent 30 minutes looking in to a different approach to this and I think it could be done, but has so many possible points of failure I'm a little worried. The up property isn't actually used anywhere in the base camera and neither is self.transform. That should mean all of the changes can be limited to the panzoom.py module. I originally though self._real_rect was the camera's rectangle oriented for the self.up property, but it isn't and is only used in the _update_transform method in question. So besides this update method the other use of the transform for the camera is viewbox_mouse_event and it is only used to get the current distance the mouse moved or the current pointing position of the mouse. I'm a little confused at why the mouse wheel event uses the Scene's transform but a mouse move uses the camera's transform. We'd have to make sure that the transform isn't linked via event handling to some other piece of code. Otherwise I don't see why it can't be handling this orientation stuff. I'm sure I'm missing something. Just brainstorming, don't feel like you have to do anything about this.

[os-gabe]

What do you think about getting this merged @djhoese? I don't think my change is causing the failed CI test is it? I don't really understand the error there.

[djhoese]

Darn, I forgot about this. I really wanted to come up with a better overall solution that didn't need the extra logic for changing up. I think I'm ok merging this, but am leaving for the weekend and have been kind of swamped with other projects' CIs failing. I restarted the failing job which was due to an HTTP timeout so that should pass. I'll plan on merging next week.

[djhoese]

Sorry, forgot about this again. Thanks.