199959 - Attribute.specified isn't true when attribute set through Attribute.value='string'

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect, P4)

Description

[David "liorean" Andersson]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.3) Gecko/20030312
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.3) Gecko/20030312

Minimal testcase:

var attr=document.createAttribute('title');
attr.value='titleAttributeIsSpecified';
alert(attr.specified); // => false

It doesn't make any difference if I append it to an element using
setAttributeNode either. Haven't tried to append through
NamedNodeMap.setNamedItem, but I assume the same is the case then.

Reproducible: Always

Steps to Reproduce:

Comment 1

[Boris Zbarsky [:bzbarsky]]

sicking?  This is all you, man.  ;)

Comment 2

[Fabian Guisset]

OS->All, Hardware->All, P4, see also bug 195350
Marking NEW.

Comment 3

[Boris Zbarsky [:bzbarsky]]

Actually, it looks like .specified on an Attr node should be "true" if it has no
element, regardless of whether the attribute has any sort of value (see
http://www.w3.org/TR/2000/REC-DOM-Level-2-Core-20001113/core.html#ID-637646024).

At least if I read this spec correctly (it really doesn't seem to do a good job
of addressing attributes that are not attached to elements).

Comment 4

[David "liorean" Andersson]

Yeah, .specified should be set to true when creating the Attr node, as this
passage IMO rakther clearly specifies:

    "If the ownerElement attribute is null (i.e. because it was just created or
was set to null by the various removal and cloning operations) specified is true."


Appending to an element SHOULD make no difference, but I tested it just to check
that the behavior was the same.


In moz it currently, wrongly, alerts false. (As a note, iem and iew alerts
false. Op7 and saf alerts true.)

Comment 5

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Mozilla currently doesn't do anything proper when it comes to .specified. We
don't at all keep track of which attributes comes from default values and which
were explcitly set.

There is some code that is executed when getting .specified, but it looks like
the person who wrote that totally missunderstood what it was supposed to be. The
code looks if the attribute is set on the element or not, which of course is
compleatly wrong.

Anyhow, my recommendation is to not use .specified at all in mozilla, there is
no support for it currently and I don't think it's very high up on anybodys
priority-list.

The problem is implementing .specified without adding bloat or reducing speed of
other operations. Does anyone have a usecase where .specified is really needed,
or is this strictly a correctness request?

(not saying that correctness isn't important, just saying that there might be
more important things out there)

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

Attachment: possible patch

I think we could do this for now, since DTDs and Schemas aren't really supported.
Gives one point in ACID3.

Comment 7

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 320744 [details] [diff] [review]
possible patch

Wow, ACID3 really tests a lot of junk

Comment 8

[Olli Pettay [:smaug][[e-mail]]]

Comment on attachment 320744 [details] [diff] [review]
possible patch

You could have sr'd too ;)
Thanks.

Comment 9

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 320744 [details] [diff] [review]
possible patch

Meh.

Comment 10

[Boris Zbarsky [:bzbarsky]]

Worth adding a unit test?

Comment 11

[Olli Pettay [:smaug][[e-mail]]]

Well, ACID3 has a test and dom1 testsuite has other tests

Comment 12

[Boris Zbarsky [:bzbarsky]]

Ok, as long as we have an automated test for this in our tests we run on tinderbox, please flag in-testsuite+.  If we don't, we need one.

Comment 13

[Olli Pettay [:smaug][[e-mail]]]

Attachment: mochitest

I'll commit this once the tree is less orange

Comment 14

[Dietrich Ayala (:dietrich)]

backed out due to orange, per sheriff myk.

Comment 15

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

The patch has been cleared of culpability for the test failure and can land again when the tree reopens.

Comment 16

[Dão Gottwald [::dao]]

http://hg.mozilla.org/index.cgi/mozilla-central/rev/f08f6a21deb7