Re TART regressions and Gavin's concerns - as always, we should not trust the numbers blindly.
The first thing we need is probably taking few windows machines with different performance characteristics and compare tab animation perf on those machines, especially on the cases where TART shows regression (TART measures 10 cases of animation - like when opening a new tab, closing a tab, in several DPIs, etc), and preferably listen to subjective assessments on this from more than one person. We should also remember that tests tend to be more reliable when the stuff they measure is in their "comfort zone". The more specialized the test is - the more it expects the test subject to behave within tighter constraints, and vice verse - the higher level the test is, the more it could treat its subject like a black box, and cares less about internal details. TART happens to be quite specialized and OMTC is a major shift in graphics implementation. Even if TART is already running and providing useful results with OMTC on OS X, it could still be out of its "comfort zone" with OMTC on windows. This is true about all tests. The more specialized the test is - the more it needs to be kept aligned with test subject, and the more it could produce less reliable results when it isn't. So, we should take the regression numbers with a grain of salt, and try to make sure first that the results are still good, and if they are, see what we can do about it, etc. _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
