503632 - [HTML5][Patch] Script containing <!-- in a string never ends up closed

Status: RESOLVED FIXED

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

Description

[Boris Zbarsky [:bzbarsky]]

Attachment: Testcase

STEPS TO REPRODUCE:
1) Enable HTML5 parser.
2) Load attached testcase

EXPECTED RESULTS: The "There should be text here" text shows up.

ACTUAL RESULTS: Blank page.

ADDITIONAL INFORMATION: DOM Inspector shows that all content after the <script> is consumed as script text.  The testcase is reduced from the site in the url bar.  Our old parser shows the text, as do Safari and Opera.

I assume this will need a spec change?

Comment 1

[-fullmetaljacket-]

dup of bug 502984?

Comment 2

[Boris Zbarsky [:bzbarsky]]

Not at all; this particular script doesn't have anything like "-->" or "-- >" at the end.

Note that this bug causes this very bug page to misrender, since bugzilla puts the bug summary in a string in a script... which is ending up not terminated as a result.

Comment 3

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

How is this dealt with by shipped in-browser HTML parsers? I'd like to avoid introducing back-and-forth parsing of comments and CDATA escapes. I suppose I could try digging deeper into the escape hole and make ' or " until ' or " respectively escape the start of an escape.

Comment 4

[Boris Zbarsky [:bzbarsky]]

The way the shipping Gecko parser deals, I think, is http://mxr.mozilla.org/mozilla-central/source/parser/htmlparser/src/nsHTMLTokens.cpp#608

It doesn't look happy, does it?

I don't know what others do.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Note that in that code, aIgnoreComments is set to theTag != eHTMLTag_script.

Comment 6

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

(In reply to comment #4)
> It doesn't look happy, does it?

It's not happy. :-(

If this pattern is something that need to be supported, I don't currently have better ideas than tracking things that look like string literals so that they don't affect the <!-- ... --> escape sections. That still leaves regexp literals, which seem harder to track without a more comprehensive EcmaScript parser integrated into the HTML parser. However, in that case, stray quotes would open up a whole new can of worms.

Considering that this bugzilla page breaks in Safari and Opera, I'm semi-hopeful that supporting this pattern isn't critical for Web compat.

Comment 7

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

Another possibility is letting <!-- have the escaping effect only if the there have been nothing but whitespace and optionally // or /* before <!-- after the previous line break or semicolon. This seems safer than paying attention to quotes.

Comment 8

[Boris Zbarsky [:bzbarsky]]

The original site I ran into this on (see url field) works fine in Safari and Opera...

Comment 9

[Jesse Ruderman]

See also bug 504941.  I don't think the position of <!-- within the script should matter.

Comment 10

[shirish]

Why is there no voting or ability of CC List to this bug ?

Comment 11

[-fullmetaljacket-]

(In reply to comment #10)
> Why is there no voting or ability of CC List to this bug ?

see note at comment #2

Comment 13

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

Note to people who are searching for dupes before filing bugs:
If you see this in the wild, please note the URL of the page here.

Comment 14

[Jesse Ruderman]

FWIW, Bugzilla is getting fixed in bug 503980.

Comment 15

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

I wrote up a relatively radical proposal for a fix:
http://wiki.whatwg.org/wiki/CDATA_Escapes

Comment 18

[Henri Sivonen (:hsivonen) (away from Bugzilla until 2026-04-13)]

http://hg.mozilla.org/mozilla-central/rev/3f5418384820