1181452 - Add memory reporting infrastructure for Servo to SpiderMonkey

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Nicholas Nethercote [inactive]]

I want to hook up SpiderMonkey's memory reporting to Servo. But I want to report much coarser-grained measurements than what Firefox reports, so there will need to be some aggregation. This bug is about exposing a new API that does that aggregation.

Comment 1

[Nicholas Nethercote [inactive]]

Attachment: Add memory reporting infrastructure for Servo to SpiderMonkey

I apologize in advance for the fact that this is a difficult patch to review.
It seems unavoidable :(

Comment 2

[Till Schneidereit [:till]]

Comment on attachment 8630883 [details] [diff] [review]
Add memory reporting infrastructure for Servo to SpiderMonkey

Review of attachment 8630883 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good. I mean, as much as it can with so much macro-ness and all.

I do wonder why it's preferable to have coarse-grained measurements only for Servo, though. Isn't it, if anything, more important to have detailed measurements because Servo uses the JS engine's GC much more than Gecko does? I know that Terrence was concerned about the increased GC pressure caused by allocating DOM objects on the GC heap. It'd be nice to get a handle of how much of an issue that is sooner rather than later.

Comment 3

[Nicholas Nethercote [inactive]]

Thank you for the fast review.

> I do wonder why it's preferable to have coarse-grained measurements only for Servo, though.

My guiding principle here relates to the display of the data.

Firefox has about:memory, which has automatic hiding of small sub-trees and support for collapsing/expanding trees manually. This gives lots of flexibility and the ability to support many measurements.

In contrast Servo currently has a very crude display system that just dumps measurements to the console every N seconds. So I'm being conservative with the granularity for now.

Comment 4

[Till Schneidereit [:till]]

(In reply to Nicholas Nethercote [:njn] from comment #3)
> > I do wonder why it's preferable to have coarse-grained measurements only for Servo, though.
> 
> My guiding principle here relates to the display of the data.

Good point, thanks for the explanation.

Comment 5

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/d8f5d0f09eec

Comment 6

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/d8f5d0f09eec