conformance regression: program-test.html fails with "drawing with a valid program shouldn't generate a GL error "

[GoogleCodeExporter]

What steps will reproduce the problem?
1. on Windows, take a Firefox build with ANGLE r1226 and my fix in issue 351 
for the program-test.html crasher. Such builds are available here:
https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/bjacob@mozilla.com-15
02a8cbe063/
2. run the program-test.html page:
https://cvs.khronos.org/svn/repos/registry/trunk/public/webgl/conformance-suites
/1.0.1/conformance/programs/program-test.html

Expected: all green. Is all green with ANGLE r1041.
Actual (the URIs below refer to Mozilla's copy of the test suite, but it's 
almost identical to the official 1.0.1 conformance tests).

56342 ERROR TEST-UNEXPECTED-FAIL | 
/tests/content/canvas/test/webgl/test_webgl_conformance_test_suite.html | 
[conformance/programs/program-test.html] Test failed - getError expected: 
NO_ERROR. Was INVALID_OPERATION : drawing with a valid program shouldn't 
generate a GL error
FAIL: getError expected: NO_ERROR. Was INVALID_OPERATION : drawing with a valid 
program shouldn't generate a GL error
STACK TRACE:
testFailed@http://mochi.test:8888/tests/content/canvas/test/webgl/resources/js-t
est-pre.js:91
go@http://mochi.test:8888/tests/content/canvas/test/webgl/conformance/programs/p
rogram-test.html:264
@http://mochi.test:8888/tests/content/canvas/test/webgl/conformance/programs/pro
gram-test.html:320

Original issue reported on code.google.com by jacob.be...@gmail.com on 19 Jul 2012 at 3:07

• Merged into: #351

[GoogleCodeExporter]

Hm, though this could well be a driver issue as our test slaves have very old 
drivers. Trying with recent drivers now.

Original comment by jacob.be...@gmail.com on 19 Jul 2012 at 3:10

[GoogleCodeExporter]

The problem persists with the recent enough NVIDIA 296.88 driver:
 * Firefox Nightly with ANGLE 1041: program-test.html is green
 * build with ANGLE 1226 + patch for bug 351: program-test.html fails

That may still be a driver issue -- but in that case I'd say that having ANGLE 
1041 work around it was really nice for our conformance status ;-)

Original comment by jacob.be...@gmail.com on 19 Jul 2012 at 3:17

[GoogleCodeExporter]

Unfortunately there were lots of changes between these two ANGLE revisions. 
Program binary support? No-flip? Shader rewriting? Other ideas?

Original comment by kbr@chromium.org on 19 Jul 2012 at 5:06

[GoogleCodeExporter]

Priority-High because it's a conformance regression.

Original comment by kbr@chromium.org on 19 Jul 2012 at 5:06

• Added labels: Priority-High
• Removed labels: Priority-Medium

[GoogleCodeExporter]

Works fine in Chrome though.  Wonder what Chrome and Firefox are doing 
differently.

Original comment by dan...@transgaming.com on 19 Jul 2012 at 5:31

[GoogleCodeExporter]

Hmm, interesting.

Benoit, can you investigate that further?

Original comment by kbr@chromium.org on 19 Jul 2012 at 5:40

• Changed state: New

[GoogleCodeExporter]

I suspect this is highly related to Issue 351, which seems to be related to 
program binary support.

Original comment by dan...@transgaming.com on 19 Jul 2012 at 6:19

[GoogleCodeExporter]

For sure, my patch for Issue 351 is a clueless sweep-the-bug-under-the-carpet 
patch. It may well be the cause of the regression. But since I don't know 
another way of avoiding crashing in program-test.html, I can't test without 
this patch.

Original comment by jacob.be...@gmail.com on 19 Jul 2012 at 6:21

[GoogleCodeExporter]

Yeah I don't think that patch is correct.  In fact, if it's the draw call on 
line 263 of the test that is crashing for you, that explains why line 264 is 
returning an unexpected error (i.e. due to your patch). 

The proper solution is to make the previous link have the desired behaviour.
Request this be closed as a dup.

Original comment by dan...@transgaming.com on 19 Jul 2012 at 6:26

[GoogleCodeExporter]

Original comment by jacob.be...@gmail.com on 19 Jul 2012 at 6:30

• Changed state: Duplicate