I was wondering what are the risks of enabling the testUsdImagingGL tests for Apple. We could cherry pick the ones that are failing and skip only those. This will give some reassurance on changes made to the chain of hierarchy or other changes rendering related changes.
I tested locally and a big portion of the tests are already passing:
84% tests passed, 45 tests failed out of 290
Thanks for the question! Our intent is to turn those on (on all platforms!), but our focus has been to try to get the whole suite passing reliably. I’m also not sure whether we’ve optimized the perceptual tolerances in CMake for everything in that test suite yet (for the image diff tool). That said, that test suite is super useful to us in Pixar internal builds, as it covers a lot of the UsdImaging behavior set.
The risks are the usual: false negatives due to tolerances being too generous; false positives due to hardware differences (e.g. small float differences); and for things that don’t currently match up (like points rendering, which I think is a different shape in GL vs Metal) those tests won’t give you information unless you look at the image yourself. If you keep those in mind, running it may be helpful, though; and we’re always happy to take fixes as well. And the tests are mostly categorized by feature pretty well, although sphere rendering and instancing tend to creep in everywhere.