538193 - [HTML5] HTML5 parser messes up betty crocker Russian Tea Cake recipe page

Status: RESOLVED FIXED

(Core :: DOM: HTML Parser, defect, P1)

Description

[Boris Zbarsky [:bzbarsky]]

BUILD: 2001-01-05 nightly

STEPS TO REPRODUCE:
1)  Load http://www.bettycrocker.com/recipes.aspx/russian-tea-cakes/3af8664b-6c3e-4022-b686-cd961521e59b
2)  Note that there are two copies of the "save/print/e-mail/share" toolbar and that the actual recipe is off in the weeds below the ad on the left.
3)  Disable the HTML5 parser and watch the problem go away.

I saved the page "complete" while not using the HTML5 parser, and the saved version shows similar problems, even in offline mode (so seems to not depend on network resources).  Might be easier to debug with that...

Comment 1

[Boris Zbarsky [:bzbarsky]]

Attachment: Zip with the page to reproduce; load foo.html

Comment 2

[d]

Is this confirmed to be a problem with the parser, and not poor coding which contradicts the new HTML5 specs parsing rules? In that case, this is a case for the evangelism team, but anyhow, a reduced test case would be good, so that we can find out what is causing it.

Comment 3

[Boris Zbarsky [:bzbarsky]]

It's not confirmed to be anything other than a case in which the HTML5 parser breaks a webpage.  It might be a bug in our HTML5 parser code, it might be a bug in the HTML5 parsing spec, or it might be something we have to evangelize.  Since the spec wants to be web-compatible, this last is least likely.

You are of course free to work on reducing the attached testcase...

Comment 4

[Henri Sivonen (:hsivonen)]

It seems this problem is related to the site doing:
document.write("<noscript>");

Comment 5

[Boris Zbarsky [:bzbarsky]]

Attachment: Testcase using document.write("<noscript>") that the HTML5 parser clearly gets wrong

Comment 6

[Henri Sivonen (:hsivonen)]

Guess: Tokenizer state migration doesn't work right when it comes to copying the name of the current element.

Comment 7

[d]

Bug 538754 looks like it might be related to this one. This also has document.write (or similar) combined with <noscript>. Inspecting with Firebug shows some odd things.

Comment 8

[Henri Sivonen (:hsivonen)]

Attachment: Copy over the content model element when loading tokenizer state from another tokenizer

Comment 9

[Henri Sivonen (:hsivonen)]

Pushed:
http://hg.mozilla.org/mozilla-central/rev/606c6a340c3f

A reftest directory for the parser would be good to have in order to add tests for document.write(), since document.write() isn't covered by the html5lib test suite.

Comment 10

[Boris Zbarsky [:bzbarsky]]

Please feel free to add a parser reftest directory as desired!

Comment 11

[Boris Zbarsky [:bzbarsky]]

I can confirm that the changeset in comment 9 fixes the site.