1340224 - stylo: Test whether Stylo improves terrible restyle performance on cleopatra

Status: RESOLVED WORKSFORME

(Core :: CSS Parsing and Computation, defect, P5)

Description

[(no longer active)]

Cleopatra is almost unusably slow if you increase the Gecko Profiler buffer size (I use 90MB for example).

I was wondering why that is today.  It seems like our restyling is extremely inefficient on this page.  For example, see this profile: https://perfht.ml/2lSZEeH.  We are spending 3.8 *seconds* restyling, in three chunks of roughly the same size.  This is easily visible if you for example try selecting a range in the timeline.  Each click has a dag of seconds to have any effect.

The profile in question from which the measurement was obtained is <https://perf-html.io/public/55385c23113278e8db31a2fee58338fdbf04bfd1/calltree/?search=topicListener&thread=1>.

Comment 1

[(no longer active)]

Bobby, if stylo is far enough, I'm very curious to know how well it performs on this page...

Comment 2

[Bobby Holley (:bholley)]

It isn't far enough to provide a useful answer to this question. I'll add this as a dep so we can remember to measure when we're getting close to landing on nightly.

Comment 3

[Greg Tatum [:gregtatum]]

For anyone in the future looking at this, you may need to run the project locally, and check out the version from this date in case our implementation changes: https://github.com/devtools-html/perf.html

That top header may be re-worked.

Comment 4

[Markus Stange [:mstange]]

That rework has landed. https://github.com/devtools-html/perf.html/pull/168
You'll need to check out a commit before that PR if you want to reproduce this.

Comment 5

[Chris Peterson [:cpeterson]]

status-firefox57=wontfix unless someone thinks this bug should block 57

Comment 6

[Mayank Bansal]

This is long fixed.