328541 - iframe drawn to canvas doesn't show immediately preceding DOM changes

Status: VERIFIED FIXED

(Core :: Graphics: Canvas2D, defect)

Description

[Myk Melez [:myk] [@mykmelez]]

If a script modifies the DOM of a document in an iframe (f.e. by adding an element via appendChild) and then immediately draws the iframe onto a canvas, the canvas doesn't reflect the modification.  However, if the script instead draws the iframe onto the canvas via a setTimeout() call, even one with a delay of zero milliseconds, the canvas does reflect the modification.  Seems like it should reflect the modification in both cases.  Testcase to be attached shortly.

Comment 1

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Interesting.. not sure where I'd look, bz or roc probably have an idea.

Comment 2

[Myk Melez [:myk] [@mykmelez]]

Attachment: testcase

Here's a testcase that demonstrates the problem.  It's a XUL doc with an iframe, a canvas, and two buttons.  The first button appends a text node to the iframe and then immediately draws the iframe to the canvas.  The second button appends a text node to the iframe and then sets a timeout to draw the iframe to the canvas.

If you press the first button, the text "sample text" appears in the iframe but doesn't get drawn to the canvas, demonstrating the bug.  If you instead press the second button, that text appears in the iframe and does get drawn to the canvas, demonstrating the correct behavior.

Note that you have to load this as chrome to satisfy security requirements.

Comment 3

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I think we need to do some flushing in drawWindow before we paint.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Yeah.  Before actually painting you want to call FlushPendingNotifications(Flush_Layout) on the relevant document.

Note that doing that can destroy any presshells or prescontexts (and when XUL is involved can actually destroy docshells) that are around.  So if you're holding any pointers to any of the above objects before you flush, you want to reget them.  I recommend flushing before you start looking for any of those.  ;)

Comment 5

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: fix

As suggested --- this fixes it.

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

checked in. We should probably take this on 1.8.1 branch too.

Comment 7

[Myk Melez [:myk] [@mykmelez]]

Works great; thanks guys.

Comment 8

[Mike Shaver (:shaver emeritus)]

Comment on attachment 213277 [details] [diff] [review]
fix

a=shaver for branch.

Comment 9

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Checked into 1.8 branch.