783064 - New tab performance analysis

Status: RESOLVED INCOMPLETE

(Firefox :: General, defect)

Description

[Neil Deakin]

The following is a log I made of reflows or marks for reflow that get made when a new opens. Tab animation causes a lot but that is a separate bug and not shown in this log.

--- start opening tab
<tab> inserted
 - mark box for reflow after insert into <box>
<notification> inserted
 - mark box for reflow after insert into <tabpanels>
 - flush style (due to getting docshell from containerBoxObject)
stop() is called within loadOneTab
 - flush layout (due to stop request, nsDocLoader::OnStopRequest)
 - reflow root box
 - flush child document style (due to stop request, nsDocLoader::OnStopRequest)
onLocationChange() called
 - mark block for reflow after remove <div>
 - mark block for reflow after insert into <div>
onSecurityChange() called
 - flush style (due to getting popup state)
 - mark for reflow after style change
 - mark box for reflow after remove <image>
 - mark box for reflow after insert <box>
 - mark box for reflow after remove <image>
 - mark box for reflow after insert <box>
isElementVisible() within focusAndSelectUrlBar() called
 - flush layout (get box object height)
 - reflow root box
 - flush layout (get box object width)
focus() within focusAndSelectUrlBar() called
 - flush interruptable layout (due to scroll into view)
select() within focusAndSelectUrlBar() called
 - flush style (due to select all)
 - mark for reflow after style change on <textbox>
--- end opening tab, the remainder is after a setTimeout
adjustTabstrip() called
 - flush layout (due to call to getBoundingClientRect())
 - mark box for reflow after insert <hbox>
 - mark box for reflow after insert <hbox>
 - reflow root box
adjustTabstrip() called (again) but within handleNewTab
 - flush layout (due to call to getBoundingClientRect())
 - mark box for reflow after style change <tab>
 - reflow root box
_fillTrailingGap() called
 - flush layout (due to call to get scrollLeft)
flush style from refresh driver
flush interruptable layout from refresh driver
flush interruptable layout from esm flush pending events
flush interruptable layout from willpaint
flush layout from scrollintoview
flush layout from querytextcontent
flush layout from scrollintoview
flush style from refresh driver
flush interruptable layout from willpaint
 - mark for reflow <textbox>?
 - reflow root box?
flush layout from stop request, nsDocLoader::OnStopRequest
flush child document style from stop request, nsDocLoader::OnStopRequest
flush child document style from load complete, DocumentViewerImpl::LoadComplete
 - reflow root box
flush child document interruptable layout from refresh driver
flush child document interruptable layout from esm flush pending events (DispatchSynthMouseMove)
flush interruptable layout from willpaint
flush child document interruptable layout from willpaint
flush interruptable layout from willpaint
flush child document interruptable layout from willpaint
flush layout from getting window.outerWidth
flush interruptable layout from willpaint
flush child document interruptable layout from willpaint
flush interruptable layout from willpaint
flush child document interruptable layout from willpaint
flush child document interruptable layout from esm flush pending events
flush layout from getting window.mozInnerScreenX
flush child document interruptable layout from esm flush pending events (mouseMoved)
flush child document interruptable layout from esm flush pending events (mouseMoved)
flush child document interruptable layout from esm flush pending events (mouseMoved)
flush interruptable layout from willpaint
flush child document interruptable layout from willpaint

Comment 1

[Dão Gottwald [:dao]]

(In reply to Neil Deakin from comment #0)
> select() within focusAndSelectUrlBar() called
>  - flush style (due to select all)

I suppose this style flush is in the input field's anonymous document, so it's probably not a big deal. We could however avoid this by calling select() only if the value is non-empty. Should we make this change?

Comment 2

[Avi Halachmi (:avih)]

FYI, possibly related, I filed bug 837535 (tabstrip-tab and newtab-button paint few gradients, possibly should paint only one). This affects tab animation as on each frame we might be painting more than we're hoping to.

Comment 3

[Tim Taubert [:ttaubert] (inactive)]

We got rid of the browser.stop() call and its consequences in bug 878747. Bug 876374 removed one of the sync reflows caused by adjustTabstrip() before addTab() finishes.

Comment 4

[Dão Gottwald [:dao]]

This is likely way outdated now.