Vivien Nicolas (:vingtetun) (:21) - (NOT reading bugmails, needinfo? please) Reporter
	Description • 11 years ago

When the keyboard is triggered to be shown there is a lot of expensive reflows in it that makes it slower to load. The time in parenthesis are in ms.

I/GeckoDump( 2576):   interruptible Reflow (16.784668000182137) 
I/GeckoDump( 2576): Stack: updateTargetWindowHeight@app://keyboard.gaiamobile.org/gaia_build_index.js:105|initKeyboard/dimensionsObserver<@app://keyboard.gaiamobile.org/gaia_build_index.js:62|
I/GeckoDump( 2576):   interruptible Reflow (18.2800289997831) 
I/GeckoDump( 2576): Stack: updateTargetWindowHeight@app://keyboard.gaiamobile.org/gaia_build_index.js:105|initKeyboard/dimensionsObserver<@app://keyboard.gaiamobile.org/gaia_build_index.js:62|
I/GeckoDump( 2576): uninterruptible Reflow (32.56225600000471) 
I/GeckoDump( 2576): Stack: candidatePanelToggleButtonCode@app://keyboard.gaiamobile.org/gaia_build_index.js:51|kr_draw@app://keyboard.gaiamobile.org/gaia_build_index.js:13|drawKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:95|renderKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:97|doShowKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:193|gotText@app://keyboard.gaiamobile.org/gaia_build_index.js:195|
I/GeckoDump( 2576): uninterruptible Reflow (1.7089840001426637) 
I/GeckoDump( 2576): Stack: showCandidates@app://keyboard.gaiamobile.org/gaia_build_index.js:15|kr_draw@app://keyboard.gaiamobile.org/gaia_build_index.js:13|drawKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:95|renderKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:97|doShowKeyboard@app://keyboard.gaiamobile.org/gaia_build_index.js:193|gotText@app://keyboard.gaiamobile.org/gaia_build_index.js:195|

Making those reflows lighter will make it display faster.

Attached is the script I'm using to see those reflows. It will moves into a devtool at some point (bug 926371) but it is already good to start using it without waiting.

Tim, do you have some resources to work on it?

	Jan Jongboom [:janjongboom] (inactive)
	Comment 1 • 11 years ago

I have done quite some stuff about keyboard rendering (also trying to minimize the reflows) in bug 875963 (which is pending review).

	Tim Guan-tin Chien [:timdream] (please needinfo)
	Comment 2 • 11 years ago

I would redirect ni? to djf as he intend to take over keyboard UI from v1.3.

	Benoit Girard (:BenWa)
	Comment 4 • 11 years ago

Here's some more information I have from filing my dupe:

STR:
1) Profile with |./profile.sh start -p b2g|
2) Toggle the keyboard on the Homescreen using 'I'm thinking of...' input box
3) Capture profile using |./profile.sh capture| and open file at http://people.mozilla.org/~bgirard/cleopatra/

This will give you a profile like this:
http://people.mozilla.org/~bgirard/cleopatra/#report=9a1c537eb937bdd2132a91c760e4979864d66ae6&search=resizeUI

I'm not an export here but a first looks:
The script interleaves severals call to window.innerWidth/innerHeight/clientWidth/offsetWidth with DOM modificatins. This means that we have to force several sync reflows to answer these queries. Ideally we need to avoid all queries that require a sync reflow.

	Doug Sherk (:drs) (inactive)
	Comment 5 • 11 years ago

While trying to repro this, I noticed that the stacks for each reflow are gone, so they must be getting triggered by native code (or the interface has changed, which wouldn't make sense since it's spec'd).

I now count: 5 reflows on initial keyboard showing, 4 reflows on every showing after that. I can really only justify 2 in either case. The 2 should be caused by: a. drawing the keyboard, and b. resizing the frame that is being hidden by the keyboard.

I think BenWa's guess is reasonable, especially based on the stacks of the first two reflows that vingtetun posted. I'm going to poke at that.

	Jan Jongboom [:janjongboom] (inactive)
	Comment 6 • 11 years ago

Please note that the render code has changed around a bit in bug 875963, where we also tried to reduce reflows.

	Jan Jongboom [:janjongboom] (inactive)
	Comment 7 • 11 years ago

This is going completely crazy with APZ on again. Fine without it.

	Tim Guan-tin Chien [:timdream] (please needinfo)
	Comment 8 • 10 years ago

(In reply to Jan Jongboom [:janjongboom] (Telenor) from comment #7)
> This is going completely crazy with APZ on again. Fine without it.

Should be consider this as fixed now? Anything need to be worked on specifically?

	Vivien Nicolas (:vingtetun) (:21) - (NOT reading bugmails, needinfo? please) Reporter
	Comment 9 • 10 years ago

(In reply to Tim Guan-tin Chien [:timdream] (MoCo-TPE) (please ni?) from comment #8)
> (In reply to Jan Jongboom [:janjongboom] (Telenor) from comment #7)
> > This is going completely crazy with APZ on again. Fine without it.
> 
> Should be consider this as fixed now? Anything need to be worked on
> specifically?

This can probably be closed. Fwiw I expect bug 970093 to improve a lot the situation here.

	David Flanagan [:djf]
	Comment 10 • 9 years ago

Bulk edit to clear old and out of date needinfo requests that I never responded to. I'm assuming that these are no longer relevant. If you are still waiting for an answer from me, please set needinfo? again.