A week ago, some deqp-gles2@functional@shaders@keywords@reserved_keywords tests started failing. Here is the result of the bisect: 34f7e761bc61d3086c1e4e42285c31678b256107 glsl/parser: Track built-in types using the glsl_type directly - 2017-10-30 18:27:09 deqp-gles2@functional@shaders@keywords@reserved_keywords@sampler2drect_vertex: pass -> fail deqp-gles2@functional@shaders@keywords@reserved_keywords@sampler2drect_fragment: pass -> fail deqp-gles2@functional@shaders@keywords@reserved_keywords@sampler2drectshadow_fragment: pass -> fail deqp-gles2@functional@shaders@keywords@reserved_keywords@sampler2drectshadow_vertex: pass -> fail 747c057530a1da32860f3881ca73a0d648e8f317 glsl/parser: Return the glsl_type object from the lexer - 2017-10-30 18:27:09 deqp-gles2@functional@shaders@keywords@reserved_keywords@double_fragment: pass -> fail deqp-gles2@functional@shaders@keywords@reserved_keywords@double_vertex: pass -> fail Sorry for the delay, the egl regression prevented the machine from making progress (ezbench issue). It is now catching up on 2 weeks of work and found this issue.
Ian, it looks like this really is supposed to be a reserved keyword in #version 100, but we're only marking it as reserved in #version 300 es and later. It looks like it's been wrong since the Paul Berry era, but must've worked by luck somehow. We should probably fix it (really trivial to do). It looks like these tests are blacklisted in dEQP, not because of any known test bugs, but because a bunch of vendors failed them, and Google hasn't yet made them required for some reason. Martin, I'm not sure it makes much sense to run dEQP tests in your CI system that aren't in the mustpass list (android/cts/master/gles2-master.txt). A lot of them are blacklisted because of test bugs, and neither Google nor Khronos require them for conformance... Cc'ing Mark as a heads up.
It seems to me that detecting regressions on blacklisted tests is desirable, even if we are not required to pass them. It may be that some blacklisted tests are wrong, but this bug is a data point that indicates tracking pass->fail transitions will identify driver issues. Tracking status for all the blacklisted tests will muddy the water for the mesa CI, as we rely on CI test results to confirm that we pass conformance. Ideally, we want the config files listing test failures to be empty. Martin's ezbench system is a nice way to get this test coverage.
(In reply to Kenneth Graunke from comment #1) > Ian, it looks like this really is supposed to be a reserved keyword in > #version 100, but we're only marking it as reserved in #version 300 es and > later. It looks like it's been wrong since the Paul Berry era, but must've > worked by luck somehow. We should probably fix it (really trivial to do). That was my thinking. I asked Martin to submit the bug so that I wouldn't forget about it.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/1647.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.