WebGLState: Use getParameter() to detect current scissor/viewport.

[Mugen87]

Related issue: Fixed #21824.

Description

As suggested in #21824 (comment).

[gkjohnson]

Looks good! gl.canvas is also used in the reset function here and here. Should those be changed to use gl.drawingBuffer* considering they seem to more accurately reflect the initial viewport and scissor settings anyway?

[Mugen87]

Using gl.canvas should not cause issues in reset().

[gkjohnson]

@WestLangley

Variables sized in device pixels should be assigned values in device pixel units -- not in CSS pixel units.

canvas.width and canvas.height should reflect the current size of the render buffer (unless the device cannot support the requested size), right? CSS dimensions would be retrieved with clientWidth and clientHeight.

@Mugen87

Using gl.canvas should not cause issues in reset().

If expo is setting gl.canvas to null as the MDN docs say its allowed to then calling reset will cause an error. And it seemed to me that the drawingBuffer* values were the more correct ones to use given my test in #21824 (comment). Maybe calling reset doesn't happen often enough for this to be a noticeable issue, though.

[Mugen87]

The usage of getParameter() makes sense in the initialization part of WebGLState and is an improvement.

However, IMO using the drawingBuffer properties is not the right thing to do. You have mentioned an issue in #21824 (comment) which was already discussed in #5194 and which is browser specific. I think we should not start to implement workarounds for this nor make the engine "smart". Coupling the viewport and scissor to the dimensions of the canvas seems the most natural thing to do (independently of Expo). Before edge cases like too large canvas sizes can be properly handled, browser implementations have to treat this issue in a common way.