MGLMapView should be non-opaque by default

[1ec5]

In #3101 (comment), @mxcl argues that MGLMapView’s isOpaque property should be set to NO by default on iOS. In #3096, I originally defaulted isOpaque to NO but switched it back to YES out of concern that performance might be impacted. Assuming the flash of black is still a problem, I’m in favor of making the map view non-opaque by default. On macOS, we render the map in an NSOpenGLLayer, which like all CALayers is transparent by default, and I haven’t noticed any problems stemming from that choice.

Alternatively, we could mitigate the flash without impacting performance by defaulting isOpaque to NO at initialization, then switching it to YES as soon as the style finishes loading for the first time (unless the developer has manually set isOpaque in the meantime).

/cc @incanus @frederoni

[friedbunny]

I’m also in favor of moving ahead with the switch to MGLMapView.opaque = NO. Being that this change that would be trivial to revert in a patch release (and if it caused problems we could advise users to simply set the property back themselves), would there be any issues with putting this on the v3.4.0 or v3.4.1 milestones?

[1ec5]

I’d be fine with either v3.4.0 or v3.4.1. Any PR that changes the default should come with iosbench results comparing performance before and after.

[friedbunny]

Certainly we should run iosbench, but I suspect the results will be inconclusive as to the absolute effect — last time I ran it, we were maxing out at 60 FPS consistently. Perhaps an impetus to expand iosbench.

[incanus]

last time I ran it, we were maxing out at 60 FPS consistently

Hmm, this is in conflict with #5489. Are we sure about this, or am I misunderstanding what we're talking about?

[friedbunny]

@incanus iosbench doesn’t stress the map terribly these days — it doesn’t add any sort of annotation and focuses almost solely on core rendering performance. The goal has always been to expand the range of tests in iosbench, and perhaps this is a good opportunity.

[incanus]

Right, but my point is that even bare, we have not been hitting 60fps, hence the annotation sync problem root. Are we now reliably doing that?

[1ec5]

All iosbench does is render the same viewport over and over again. The framerate issues we were seeing over in #5489 are only triggered when panning, zooming, or rotating the map.

[1ec5]

Fixed in #7859 on the release-ios-v3.4.0 branch.