There are two cases that a gvr context is needed:
1. magic window mode (gvr context is needed to get sensor data)
2. presenting vr (either webvr presenting or ChromeVr mode)
The way we currently implement is:
1. For case 1, we create a non presenting gvr context and keep it around as long as there is a magic window in tab
2. After switch to case 2, we create a VrShell which is a GvrLayout, which contains a gvr context. And this gvr context takes control when presenting.
3. After exit vr presenting, the non presenting gvr context is used again by magic window.
Ideally, we just want one gvr layout. But currently, there is a gvr bug that results incorrect left/right eye rendering if gvr context is created when phone is in portrait mode. So create a gvr context for magic window(mostly in portrait mode) and separate gvr context for rendering (only in landscape mode) is the preferred way currently.
This could change in the future, and we probably also don't need to keep non presenting gvr context around when presenting (this might have interesting effect if we open a magic window in ChromeVr). So a refactor
is probably necessary. Creating this bug to track the process.