We should have a test in which a fragment shader tries to read from a varying that the vertex shader did not define. I suspect that this case works correctly, but we have an incorrect assertion that will get tripped.
Created attachment 198437 [details] Patch
Comment on attachment 198437 [details] Patch Looks great! Just a couple of nits. View in context: https://bugs.webkit.org/attachment.cgi?id=198437&action=review > LayoutTests/ChangeLog:7 > + don't lead to unexpected crashes due to linking errors. You might want to add that "The current implementation does not crash, and this test will verify that it doesn't crash in the future." > LayoutTests/css3/filters/custom/custom-filter-unavailable-varying.html:5 > + #ohmymy { Could you change this name to "#filter-target"? > LayoutTests/css3/filters/custom/custom-filter-unavailable-varying.html:9 > + -webkit-filter: custom(url('../resources/unavailable-varying.vs') mix(url('../resources/unavailable-varying.fs') normal source-atop)); Your vertex shader looks a lot like the implementation's default shader. Maybe you use the "none" keyword instead of "unavailable-varying.vs"?
Created attachment 198442 [details] Patch
Comment on attachment 198442 [details] Patch Clearing flags on attachment: 198442 Committed r148581: <http://trac.webkit.org/changeset/148581>
All reviewed patches have been landed. Closing bug.