843075 - Firefox 19 fails to display Sudoku grids (regression)

Status: RESOLVED WORKSFORME

(Core :: DOM: Core & HTML, defect)

Description

[Loic]

STR:
Open these Sudoku websites:
http://www.sudokuhints.com/fr/sudoku-medium.html
or
http://www.extremesudoku.info/sudoku.html

Result: FF19 fails to load and display the blank Sudoku grid.

Error console log:
Error: TypeError: q2.frames.st.window is undefined
Source File: http://www.extremesudoku.info/sudoku.js
Line: 2

Regression range:
m-c
good=2012-10-15
bad=2012-10-16
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=942ed5747b63&tochange=8f145599e4bf

Reporter: http://forums.mozfr.org/viewtopic.php?p=722649#p722649

Comment 1

[Alice0775 White]

Regression window(cached m-i)
Good:
http://hg.mozilla.org/integration/mozilla-inbound/rev/224fddb79a38
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:19.0) Gecko/19.0 Firefox/19.0 ID:20121015103548
Bad:
http://hg.mozilla.org/integration/mozilla-inbound/rev/c350c6f28dd1
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:19.0) Gecko/19.0 Firefox/19.0 ID:20121015110148
Pushlog:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=224fddb79a38&tochange=c350c6f28dd1

Triggered by:
c350c6f28dd1	Boris Zbarsky — Bug 799875. Make the global scope polluter work even for qualified lookups. r=jst

Comment 2

[Loic]

Ty for the reg range. :D

Comment 3

[Boris Zbarsky [:bzbarsky]]

Hmm.  So q2.frames.st is coming out as an HTMLIFrameElement with id="st".  It used to come out as undefined in Firefox 18.

The relevant script bit looks like this:

    if (j6.all) {
        p6 = j6.frames["st"].document.body
    } else if (q2.frames && q2.frames["st"]) {
        p6 = q2.frames["st"].window.document;
        x8 = 0
    } else if (K("st")[j1]) {
        p6 = K("st")[j1].body
    }

We used to fall into the last branch, but now we end up in the second one.

Comment 4

[Boris Zbarsky [:bzbarsky]]

In Chrome, q2.frames.st seems to be the window inside the iframe.

Also in Chrome q2.frames[2].name == "st" while for us it's blank.

It looks like <iframe id="foo"> gives the window inside the name "foo" in WebKit.

The spec says the browsing context name (which is what .name on the window gets) is set from the "name" attribute of <iframe>.  The "id" attribute is not supposed to affect the browsing context name.

So as I see our options are the following:

1)  Back out bug 799875, reintroduce the difference for qualified access vs not somehow.
2)  Make an id set the browsing context name, get the spec changed accordingly.  Note
    that there's no guarantee that this will work, since the site sets p6 to the
    document, not the body, in that branch for some reason that presumably has to do with
    WebKit-onlyness.
3)  Leave spec as-is, file a bug on WebKit to fix their <iframe id> handling, evangelize
    the site to fix their script.

My personaly preference is for #3 given the evidence so far....

Comment 5

[Boris Zbarsky [:bzbarsky]]

Oh, the WebKit bug on them not matching the spec seems to be https://bugs.webkit.org/show_bug.cgi?id=11388

Comment 6

[Alex Keybl [:akeybl]]

Not enough user impact yet (a couple of small impacted sites) to track for release at this point.

Comment 7

[Boris Zbarsky [:bzbarsky]]

Hmm.  Have these sites been updated already?  The bug doesn't seem to be present anymore, and http://www.extremesudoku.info/sudoku.js now looks like this:

    if (d8.all) {
        n6 = d8.frames["st"].document.body
    } else if (l5.frames && l5.frames["st"] && l5.frames["st"].window) {
        n6 = l5.frames["st"].window.document;
        e6 = 0
    } else if (K("st")[s2]) {
        n6 = K("st")[s2].body
    }

which is checking for .window before jumping, so should work again.

Comment 8

[Loic]

Yes, fixed on website side.