10-20% performance regression between 5.9 and 5.10

[davidtaylorhq]

🐞 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)

🔬 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

[ef4]

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.)

[NullVoxPopuli]

This should resolve

• [BUGFIX beta] Upgrade glimmer-vm for some perf improvements #20746

[NullVoxPopuli]

While our performance work isn't done, I do believe this particular problem is resolved

Link: https://ember-performance-testing-prod-classic.pages.dev/report?benchmarks=%5B%22Render%20list%20(%40glimmer%2Fcomponent)%22%2C%22Render%20simple%20list%20(%40glimmer%2Fcomponent)%22%5D&clear=0&emberVersions=%5B%22ember-5-9%22%2C%22ember-6-2%22%2C%22ember-5-10%22%5D&timePerTest=5000

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)

Here is an example.