2749 - Comments are incorrectly parsed: <!-- -- --> inside comment! <!-- -- -->

Status: VERIFIED FIXED

(Core :: DOM: HTML Parser, defect, P2)

Description

[Hixie (not reading bugmail)]

Comment delimiters are "--" while inside tags.

Thus: <!-- in --  -- in --  -- in -->
where "in" shows what is commented.

On the test page quoted, all is explained.

Comment 1

[John Morrison]

fixed the URL to what was intended

Comment 2

[Hixie (not reading bugmail)]

Eek! Thanks.

Comment 3

[Hixie (not reading bugmail)]

Incidentally, the test page doesn't (yet) test this:

 <A name=abc -- href="comment" --> </A>

That is, "--" delimiters also delimit comments inside element tags. Thus the
above should be parsed as:

 <A name=abc > </A>

...and the following:

 <A -- href="hello.my. -- name=fred -- /world.com"--> </A>

...should be parsed as:

 <A  name=fred > </A>

...and the following:

 <DEL datetime="1999-01-30T1325Z" --> Never mind. --> </DEL>

...should be parsed as:

 <DEL datetime="1999-01-30T1325Z" > </DEL>

Eventually I'll add these to the test page quoted above.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Are you sure about those examples above, Ian?  I think the -- only applies when
one is within SGML processing instructions, i.e, <! ... >, not any old tags.  I
don't have a copy of any SGML standard (I think it's on paper only!), so I'm
not sure.  But nsgmls (via http://www.htmlhelp.com/tools/validator/) agrees
with me.

Comment 5

[Hixie (not reading bugmail)]

If nsgmls says I'm wrong, then I'm wrong.

The SGML spec is indeed paper only, I was basing my assertion on what
I could remember of a tutorial I read a few months back. Please therefore
disregard anything I previously said about comments in any old tags.

Comment 6

[Hixie (not reading bugmail)]

See the end of the quoted test page:
  http://www.bath.ac.uk/%7Epy8ieh/internet/eviltests/comments-evil.html
Basically, "->" is terminating comments, instead of "--".

This is causing pages to break, for example:
  http://www.bath.ac.uk/%7Esu0bufs/
...has its comment visible at the top!

Comment 7

[leger]

Setting all current Open Critical and Major to M3

Comment 8

[rickg]

Fixed by update to comment parsing. We should now be handling common practice of
comments correctly now, but still not to spec. That won't get enabled until we
actually have a strict dtd.

Comment 9

[Hixie (not reading bugmail)]

Will not verify until strict parsing mode enabled.

Comment 10

[Hixie (not reading bugmail)]

Query: Is "->" supposed to end comments in NavQuirks mode?

Comment 11

[Jan Carpenter]

Reporter awaiting developer response.

Comment 12

[rickg]

We run in quirks mode by default -- and so yes, -> can close a comment. In
non-quirks mode you will have to use correctly formed comments.

Comment 13

[Jan Carpenter]

verified
1999-07-09-08-M8

Comment 14

[Hixie (not reading bugmail)]

janc: Strict mode parsing is not yet enabled. This bug cannot be verified until
it is. I am adding a dependency to indicate this and removing the verified
notation on this bug for now.

In any case, how exactly did you verify that the bug was fixed? Viewer still
fails two out of three of the tests on the test page for this bug. (It fails
them because it parses in quirk mode.)

Comment 15

[Jan Carpenter]

This was verified per the developer.
Clearing "Resolved Fixed" and re-opening bug.

Comment 16

[leger]

Clearing Fixed resolution due to reopen.

Comment 17

[leger]

Moving from M3 to M9 milestone since reopen.  rickg, please set to later
milestone if more appropriate.

Comment 18

[rickg]

I'll move this to REMIND, since we're working per Nav.

Comment 19

[Hixie (not reading bugmail)]

Ok, DOCTYPE-linked standard mode is now hooked up. We still fail these tests,
even though the test page's DOCTYPE should trigger standard mode.

Reopening.

Comment 20

[Masatoshi Kimura [:emk]]

*** Bug 12025 has been marked as a duplicate of this bug. ***

Comment 21

[harishd]

Strict comment handling has been hooked up to the noquirks.

Comment 22

[Masatoshi Kimura [:emk]]

Even if quirks mode, "->" should not terminate comments
because "->" do not end comments both on Nav4 and on IE.
reopen.

Comment 23

[Masatoshi Kimura [:emk]]

Attachment: Testcase for quirks mode coments

Comment 24

[leger]

Clearing FIXED resolution due to reopen.

Comment 25

[harishd]

Checked in a fix..

Comment 26

[Masatoshi Kimura [:emk]]

Verified fixed.

Comment 27

[Masatoshi Kimura [:emk]]

Oops. I found standards mode comment handling is still incorrect.
Reopening.

Comment 28

[Masatoshi Kimura [:emk]]

Attachment: Testcase

Comment 29

[harishd]

Good catch.

Tweaked strict comment handling.  Marking FIXED.

Comment 30

[Masatoshi Kimura [:emk]]

Still incorrect using 1999102908 build. Reopening.

Comment 31

[Masatoshi Kimura [:emk]]

Attachment: Yet another testcase

Comment 32

[Masatoshi Kimura [:emk]]

Attachment: Proposed patch

Comment 33

[harishd]

The fix is already in.. which build did you check?

Comment 34

[harishd]

Oops, I'm sorry...I see what you're saying.  When I tweaked the strict comment
I, some how, forgot to comment out the line that you'd proposed...my bad :(

Will checkin the change Monday.  Thanx for catching it.

Comment 35

[leger]

Clearing FIXED resolution due to reopen of this bug.

Comment 36

[Hixie (not reading bugmail)]

Note. The eviltest has been updated to contain all permutations mentioned
herein. Thanks to [e-mail] for spotting them. :-)

   http://www.bath.ac.uk/%7Epy8ieh/internet/eviltests/comments-evil.html

Comment 37

[harishd]

Marking FIXED.

Comment 38

[Hixie (not reading bugmail)]

All tests on the evil test (a superset of the attachements) pass correctly.
Marking verified.

Comment 39

[Hixie (not reading bugmail)]

Reopening.
http://www.bath.ac.uk/%7Epy8ieh/internet/eviltests/comments-evil.html fails again.

Comment 40

[clayton]

Ian, would you try to identify the time frame in which the regression was 
reintroduced?  If it appeared recently it might be much easier to find and fix. 
 
How critical is this?  It says 'major' severity, it that accurate?

Comment 41

[Hixie (not reading bugmail)]

I would bet that this regressed when we turned off the Strict HTML parser.
->normal; this isn't major any more.

Comment 42

[harishd]

Ian, you're right. It did regress when the strict DTD got turned off. It's still 
possible to turn on strict comment parsing but I'm not sure if it's worth doing 
so because we now display exactly the way Nav4.x and IE do. Ian, what is your 
take on this?

Comment 43

[Hixie (not reading bugmail)]

If it is easy to turn on in standards mode, then we should. How easy is it?

I would say it's worth fixing, but not necessarily for mozilla0.9. If it is not
an easy fix, then I propose targetting mozilla1.0.

Comment 44

[harishd]

Ian, refer to bug 53011. How do we handle cases like this ( as in bug 53011 ) if 
the strict comment parsing is turned on?  evangelize?

Btw, it's a onliner to turn on strict comment parsing.

Comment 45

[Hixie (not reading bugmail)]

Yes, we would evangelise. This is just like we do with all the other "bugs" that
pages that trigger standard mode uncover, such as the inline box model issues.

Thankfully, very few broken pages trigger standard mode.

If it's a one line fix, I would recommend we do it for 0.9.

Comment 46

[rubydoo123]

I agree with Ian, if you can get this one in that would be a good thing. I was 
working with another bug and found this to happen:

original reduced test:
<html>
<head>
  <title>test</title>
<META NAME=GENERATOR CONTENT="Microsoft FrontPage 4.0">
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
</head>

commented it out, but placed the close comment as a single dash:
<html>
<head>
  <title>test</title>
<!--
<META NAME=GENERATOR CONTENT="Microsoft FrontPage 4.0">
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
->
</head>

and when I dumped the html, this is what the content was:
<head>
<title>test</title>
<!--
<META NAME=GENERATOR CONTENT="Microsoft FrontPage 4.0">-->
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
</head>
<body>
-&gt; 

Note the closing comment is placed after the first meta element -- why is that? 
In addition, note that the -> is escaped and within the body

Not sure if the resolution of this bug will resolve that -- Ian, is this similar 
to issues that you encountered?

Should this be a new bug? If so, I will gladly open one up

Comment 47

[Hixie (not reading bugmail)]

beppe: I'm not sure that's a bug, but file it or mail me directly so that we can
take the discussion out of this bug. (per HTML/SGML/XML, "->" is not a comment
delimiter, so it shouldn't close the comment, IMHO; hence why I say this is 
correct. I dunno, I might be wrong.)

Comment 48

[harishd]

Enabled strict comment parsing. Marking FIXED.

Comment 49

[Hixie (not reading bugmail)]

Horraahh! VERIFIED FIXED.

Comment 50

[Alix Guillard]

I still get it with version 3.5 for Mac see http://alix.guillard.fr/index2.php

Comment 51

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

(In reply to comment #50)
> I still get it with version 3.5 for Mac see http://alix.guillard.fr/index2.php

You're actually complaining about the opposite, i.e., bug 214476.  (This bug fixed us to behave as SGML specifies, or something like that, but that breaks your page.  Bug 214476 is a request to undo that, which we will when we switch to our HTML5 parser.)

It's an interesting case, since it involves commenting out IDN in the xn-- form, though.