1045970 - TabChild::RecvUpdateDimensions trigger PresShell::ResizeReflowOverride all the time in B2G::reftest

Status: RESOLVED INVALID

(Core :: Layout, defect)

Description

[u459114]

In reftest, widget/window size keep still. Supposedly, we should only receive TabChild::RecvUpdateDimensions once in a whole test run. File this bug to dig out why reftest content receive this message from chrome all the time in B2G::reftest run

Comment 1

[Kan-Ru Chen [:kanru] (UTC+9)]

(In reply to C.J. Ku[:cjku] from comment #0)
> In reftest, widget/window size keep still. Supposedly, we should only
> receive TabChild::RecvUpdateDimensions once in a whole test run. File this
> bug to dig out why reftest content receive this message from chrome all the
> time in B2G::reftest run

Are you sure it's TabChild::RecvUpdateDimensions that triggered he resize reflow? Per bug 997101 comment 58 before-first-paint is sent for each reftest, maybe it's the reason.

Comment 2

[u459114]

I am sure... that you are right..

TabChild::Observe explicit triggers resize once
  http://dxr.mozilla.org/mozilla-central/source/dom/ipc/TabChild.cpp?from=TabChild.cpp&case=true#795
And indirect tirgger resize again
  http://dxr.mozilla.org/mozilla-central/source/dom/ipc/TabChild.cpp?from=TabChild.cpp&case=true#805
  http://dxr.mozilla.org/mozilla-central/source/dom/ipc/TabChild.cpp?from=TabChild.cpp&case=true#805

And thanks for your info.

Comment 3

[u459114]

OK, then, everything make sense to me.

reftest-content sets fullZoom in onload event handler
http://dxr.mozilla.org/mozilla-central/source/layout/tools/reftest/reftest-content.js?from=reftest-content.js&case=true#592

Which may come later then first-paint, so we may receive two SetCSSViewport before or after SetFullZoom.

Close this one.

Comment 4

[Kan-Ru Chen [:kanru] (UTC+9)]

(In reply to C.J. Ku[:cjku] from comment #3)
> OK, then, everything make sense to me.
> 
> reftest-content sets fullZoom in onload event handler
> http://dxr.mozilla.org/mozilla-central/source/layout/tools/reftest/reftest-
> content.js?from=reftest-content.js&case=true#592
> 
> Which may come later then first-paint, so we may receive two SetCSSViewport
> before or after SetFullZoom.
> 
> Close this one.

Thanks. The redundant SetCSSViewport is indeed a performance problem and we should fixed if we could know why the reftest failed in bug 997101.