20:01 alyssa: dEQP-VK.pipeline.monolithic.cache.misc_tests.internally_synchronized_test failing with common vk_pipeline.. ummf
20:01 alyssa: I assume nvk will hit this when bringing up maint8
20:09 alyssa: wait what
20:09 alyssa: how does this pass for anyone
20:10 Sachiel: passes here
20:11 alyssa: Sachiel: CTS asserts that a pipeline cache can be deserialize->serialize roundtripped
20:11 alyssa: but Mesa pipeline caches are backed by hash-sets
20:11 Sachiel: ah, I always build cts release
20:11 alyssa: i.e.: set_foreach
20:11 alyssa: and set_foreach is nondeterministic order
20:11 Sachiel: so if it's trying to validate things with asserts, I won't see it
20:12 alyssa: no i mean it's part of the test
20:12 Sachiel: oh, ok
20:13 alyssa: although we might have multiple issues here
20:13 Sachiel: well, I don't know. I don't remember much about that anymore. When I was trying to do maint8, I think I saw those tests fail once, and then they always passed and coulnd't figure out why
20:15 alyssa: nondeterminism =D
20:18 Sachiel: Pass/Fail/NotSupported/AcceptYouCantControlEverything
21:12 alyssa: hehe
21:21 alyssa: filed https://gitlab.khronos.org/vulkan/vulkan/-/issues/4236 about the set_foreach thing
21:21 alyssa: but I suspect there's something else going wrong here
21:24 dj-death: alyssa: unset VK_ENABLE_PIPELINE_CACHE?
21:24 alyssa: dj-death: that makes the test pass
21:25 alyssa: ok confirmed it's the order thing
21:25 alyssa: unsure why I hit this for anv/radv doesn't
21:25 alyssa: maybe arm64 weirdness or something
21:27 zmike: probably y-inverted