webgl/2.0.0/conformance2/rendering/clipping-wide-points.html started to fail 100% after the switch to ANGLE on some hardware/software configurations (I think that it fails on all IG models, but didn't check carefully). https://results.webkit.org/?suite=layout-tests&test=webgl%2F2.0.0%2Fconformance2%2Frendering%2Fclipping-wide-points.html The diff appears to say that it's actually passing on IG and failing elsewhere? @@ -1,15 +1,5 @@ This test runs the WebGL Test listed below in an iframe and reports PASS or FAIL. Test: ../../resources/webgl_test_files/conformance2/rendering/clipping-wide-points.html -[ 1: PASS ] getError was expected value: NO_ERROR : Should be no errors after setting up program -[ 2: PASS ] should be 0,255,0 -[ 3: PASS ] getError was expected value: NO_ERROR : Should be no errors after running one test case -[ 4: FAIL ] should be 0,255,0 at (0, 0) expected: 0,255,0 was 255,0,0 -[ 5: PASS ] getError was expected value: NO_ERROR : Should be no errors after running one test case -[ 6: PASS ] should be 0,255,0 -[ 7: PASS ] getError was expected value: NO_ERROR : Should be no errors after running one test case -[ 8: FAIL ] should be 0,255,0 at (0, 0) expected: 0,255,0 was 255,0,0 -[ 9: PASS ] getError was expected value: NO_ERROR : Should be no errors after running one test case -[ 10: PASS ] successfullyParsed is true -[ FAIL ] 2 failures reported +[ PASS ] All tests passed
<rdar://problem/58257017>
The test used to pass before r253926, but the change added a failing expectation for it. So what this means is that the test actually regressed on Macs with AMD graphics. Here are some of the affected models: MacPro6,1 MAcPro7,1 iMac15,1 iMac17,1 iMac18,2 iMac18,3 MacBookPro11,5 MacBookPro13,3
Unfortunately this test uncovers some longstanding behavioral differences between AMD GPUs and other vendors'. It's been failing in Chromium for a long time: see https://bugs.chromium.org/p/chromium/issues/detail?id=642822 . Likely we'll have to remove this test from the WebGL 2.0 conformance suite, since the emulation path to make this work on AMD GPUs would likely be too slow.
Let me take this bug to follow through on it.