As we all know, this case draw a top-left triangle in a square, the size of square is 2*2. 
I had run the case on my chromebook, if I changed the size to 3*3 or 4*4, the case passed. It seemed case only failed when the size is 2*2, and the red channel values of four pixels are 0, 0, 255, 0.
@dongseong Could you help to check whether it is a CMAA regression? thank you.

The CMAA shader implementation in Chromium only processes the color attachments, as documented here: https://www.khronos.org/registry/gles/extensions/INTEL/framebuffer_CMAA.txt
Thus, depth buffers are not processed by CMAA. Maybe this causes the regression?

@belgun, the failure case is sub-case of this html. In this case, it only verifies whether anti-alias feature is valid, and it has disabled the depth, stencil features, etc.
The drawing content is on color attachments, so I think it directly related with CMAA. When creating context, the source code is as below,
if (antialias)
   shouldBeNonNull("gl = getWebGL(2, 2, { depth: false, stencil: false, alpha: false, antialias: true }, [ 0, 0, 0, 1 ], 1, 0)"); 

I verified the regression came from r403821："gpu, cmaa: Don't blend isolated L shape because it's not a closed geometry."

Dongseong, can you take a look?

Thank you for reporting. Let me look immediately.

Hi, IMO, the fixed webgl test is wrong. 

I submitted the WebGL test PR. zmo@, could you review?
https://github.com/KhronosGroup/WebGL/pull/1977

PR #1730 [1] changes the test to pass CMAA wrong behavior.
As the CMAA bug [2] is fixed, this test needs to be more reasonable. See the PR description.
[1] https://github.com/KhronosGroup/WebGL/pull/1730
[2] https://chromium.googlesource.com/chromium/src/+/9df37c48526a8f6b350687e34e75b2eca1961736

Fixed at the WebGL test. https://github.com/KhronosGroup/WebGL/pull/1977