Repository navigation
10-20% performance regression between 5.9 and 5.10聽#20719
Description
Activity
- changed the title
[-]10% performance regression between 5.9 and 5.10[/-][+]10-20% performance regression between 5.9 and 5.10[/+]on Jul 9, 2024 I'll take a look.
(My first impression of the benchmarking suite: it sure is tied to the non-ES-modules way of doing things that we're trying to deprecate.)
Reacted by David TaylorThis should resolve
While our performance work isn't done, I do believe this particular problem is resolved
The problem was that we accidentally forgot to strip a bunch of code from internal glimmer debug stuff before publish / merge in to ember.
Going forward, we now have a dev asset size comment on every PR comparing the PR's change in asset sizes to
main-- while this isn't a direct indicator of performance, it'll at least be a reminder to check if the delta is significant (as it was in the 5.9 to 5.10 case)Reacted by David Taylor and Boris PetrovClosing based on @NullVoxPopuli comment on July 13
馃悶 Describe the Bug
emberperf (https://emberperf.discourse.org/, https://github.com/discourse/ember-performance) has identified a performance regression between Ember 5.9 and 5.10. This is in addition to the earlier (and more significant) glimmer-vm regression being tracked in glimmerjs/glimmer-vm#1590.
Running against individual commits between 5.9 and 5.10 shows:
This suggests that the performance regression was introduced in 53b2de8 (cc @ef4)
馃敩 Minimal Reproduction
Released ember versions can be tested at https://emberperf.discourse.org/. Enable the "Render Complex List (Glimmer)" test, and choose Ember versions to test.
The version of emberperf with per-commit tests can be found at discourse/ember-performance#7