1711076 - Difficult to diagnose perf differences caused by GCs before a profile starts

Status: RESOLVED FIXED

(Core :: JavaScript: GC, enhancement, P3)

Description

[Steve Fink [:sfink] [:s:]]

When trying to figure out why a GC scheduling-related patch regressed things, I was looking at browsertime warm profiles of pre-patch and post-patch runs. They showed pretty much nothing of interest, just the changed behavior that was a downstream result of something happening before.

What was somewhat helpful was knowing that there had been more or fewer GCs before the profile. (The difference only showed up in warm profiles, not cold ones, which is consistent with something earlier changing things.) A simple way of exposing that is to record the major and minor GC numbers.

This doesn't seem universally useful enough to actually expose in the hover text for a profile, but I could still get it from the context menu by copying the full marker json.

Comment 1

[Steve Fink [:sfink] [:s:]]

Attachment: Bug 1711076 - Add minor/major GC counts to profiles

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/c0e7224e6b5b
Add minor/major GC counts to profiles r=jonco

Comment 3

[Iulian Moraru]

https://hg.mozilla.org/mozilla-central/rev/c0e7224e6b5b