1152262 - Instrument the refresh driver to find out which pages eat CPU

Status: NEW

(Core :: Layout, defect)

Comment 1

[David Teller [:Yoric] - still alive but not very active]

Spinoff from bug 1149324.

If I understand correctly, the idea would be to instrument nsRefreshDriver::Tick() as follows:
- if we don't have a docShell, nothing to do;
- if we are not in a toplevel docShell, nothing to do;
- otherwise, simply measure the time spent executing this method and somehow charge it to the docShell (and/or its attached JSCompartment).

Is that correct?

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

(In reply to David Rajchenbach-Teller [:Yoric] (use "needinfo") from comment #1)
> If I understand correctly, the idea would be to instrument
> nsRefreshDriver::Tick() as follows:
> - if we don't have a docShell, nothing to do;
> - if we are not in a toplevel docShell, nothing to do;
> - otherwise, simply measure the time spent executing this method and somehow
> charge it to the docShell (and/or its attached JSCompartment).
> 
> Is that correct?

I think we'd want to find someplace appropriate to charge the time to no matter what -- the refresh driver should always have an mPresContext (unless it hits the early return 10 lines in), and the pres context should always have an mDocument, which should always have a window.  I guess if one of those isn't true I'm probably not too worried, although maybe there's an interesting case like refresh drivers for SVG resource documents.  (ni? :heycam to fill in the situation there, since I've forgotten)

Comment 3

[Cameron McCormack (:heycam)]

I don't recall how windows/docshells work for SVG image documents and their resource documents -- Daniel/Seth?

Comment 4

[Daniel Holbert [:dholbert]]

IIRC, the refresh driver should be disabled inside SVG-as-an-image. They update based on pings from their host document's refresh driver, via imgIContainer::RequestRefresh. (This lets them avoid painting in background documents)

SVG-in-opentype is one special case, though -- the glyphs there *do* have an internal refresh driver, because we haven't yet burrowed a RequestRefresh-type channel from the document through the font to the opentype glyphs. See Bug 1107252 comment 3 for more.

(I don't recall how this works for resource documents that are pulled in for filters/clip-regions -- whether those documents have their own refresh driver or what.)

Comment 5

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #4)
> the refresh driver should be disabled inside SVG-as-an-image. They
> update based on pings from their host document's refresh driver, via
> imgIContainer::RequestRefresh. (This lets them avoid painting in background
> documents)

s/painting/updating animations/.  This is how we throttle animated GIFs/PNGs, too, BTW. (This work happened for SVG-as-an-image in bug 1002632.)