We had some weird code in there, making us fail that test.
Created attachment 493454 [details] [diff] [review] fix readpixels
Comment on attachment 493454 [details] [diff] [review] fix readpixels erm wait, I just added that code in bug 612336 (was checked in with wrong bug #). We have to synthesize INVALID_OPERATION if both format and type are invalid; the test even says: // note that the GL error is INVALID_OPERATION if both format and type are invalid, but INVALID_ENUM if only one is. Need to update our copy of the test suite with latest from svn again looks like.
We're still failing this test (from svn). Also, it doesn't seem to have anymore the comment that you were referring to: https://cvs.khronos.org/svn/repos/registry/trunk/public/webgl/sdk/tests/conformance/gl-enum-tests.html Moreover I can't see where in the WebGL or GL ES spec it would prescribe this behavior. It seems to me that at some point in the past this test was wrong, and has been fixed since.
This test is now passing on all platform on the build slaves.