GL cleanup and optimizations.

[juj]

Optimizes GL emulation layer to avoid last detected duplicate redundant calls to GL. Cleans up GL library a bit by using 'GLImmediate' and 'GLctx' global names instead of routing though GL. and Module.

Still has a minor cleanup pending for used #ifdef macros, which needs #1969. A few places would like to use #if defined(FULL_ES2) || defined(LEGACY_GL_EMULATION) but the code had to duplicate that since the support for that isn't present.

[juj]

Running try builds now to check that no regressions are present..

[kripken]

Please comment here to ping me after tests passed.

[juj]

Overnight, OSX ran the default and other suites and passed, and Linux ran the browser suite and passed as well, so all looks good.

[kripken]

You removed quotes on Module.ctx['compressedTexImage2D'], however I think we need those because closure compiler mistakenly minified that name. Try to build a test with that call with --closure 1 to see if that changed. Ditto for Sub.

You can now remove GL.immediate = GLImmediate.

Why are matrix and matrixStack initialized to arrays of 16 zeros? Should be nulls since they will be references to objects (numbers tell the js engine to expect it to stay an array of numbers, which can confuse it), and why 16? Also, can't we initialize it to null in the JSON object definition, then set it to the right size in the initializer? Same with this.textureMatrixVersion = [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ];

I see you removed the unsafe check on // Potentially unsafe, since lastArrayBuffer might not reflect the true array buffer in code that mixes immediate/non-immediate - is that now guaranteed to be safe?

[kripken]

Please test this on BananaBread, with and without closure, to verify it does not regress anything.

[kripken]

One other thing missing here is var GLctx in the top level, unless I missed that. Without that, we pollute the global scope.

[juj]

Oh, I was puzzled about the Module.ctx['compressedTexImage2D'] and why only that function was closure-shielded, and thought that it was some leftover, so converted it to dot notation to be consistent.

I used [0,0,0, ..., 0] to avoid having to add dynamic initialization code (and defaulting on that number of texture coordinate attribute matrices). Agree on your points, I'll convert the initialization to use a dynamic init so that it's more consistent.

[juj]

Ok, updated the pull request. Works with Bananabread with closure on. How does it look now?

[kripken]

I don't have time to look at it right now, but let's merge it in, will take a look later. I'm sure it's fine though.