On Mon, 26 Aug 2024 14:00:22 GMT, Claes Redestad <redes...@openjdk.org> wrote:
>> #14404 caused some startup regressions, where the main cause of startup >> increase in that change was due the use of - and the not very optimized >> state of - runtime bootstrapped switches. This was remedied by a series of >> optimizations to the switch bootstrap methods and finally a desugaring of >> the switch added by #14404. Now @shipilev reports that there appears to be >> some lingering startup issue from #14404. >> >> One thing that stands out is a couple of caches which are being eagerly >> initialized regardless of user locale. This PR makes those caches lazily >> initialized and slightly more efficient (no need to implement >> `UnaryOperator` after inlining some code). Avoids loading 4 classes if >> running the HelloStream startup test in a locale that's in the set of >> constant base locales, 2 otherwise. This doesn't really do much to move the >> needle for startup on my setup, but tentatively it's the only remaining >> inefficiency I've found that came from #14404. > > Claes Redestad has updated the pull request incrementally with one additional > commit since the last revision: > > final @shipilev - mind testing this on your setup? I suspect there are other regressions not related to #14404 that has happened between 23-b13 and current RC, but it's an open question if I should leave JDK-8338897 open or close it with this PR. None of the code in Locale/BaseLocale gets executed a lot in the HelloStream tests, but a small dependency reduction like this might help. ------------- PR Comment: https://git.openjdk.org/jdk/pull/20713#issuecomment-2310305847
