1138368 - Setting span.textContent or span.innerHTML does not fire characterData for MutationObserver on span

Status: RESOLVED INVALID

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

Description

[Serge Lebedev]

Attachment: characterData_test_case.html

User Agent: Mozilla/5.0 (Windows NT 6.1; rv:36.0) Gecko/20100101 Firefox/36.0
Build ID: 20150222232811

Steps to reproduce:

I created two MutationObserver on a span element with characterData option (one with subtree option and other without) and changed text in that element in various ways (see attached test case).


Actual results:

Only cases #3 and #4 fired charactedData event on the MutationObserver with subtree.


Expected results:

I expected cases #1 or #2 to fire characterData event.

The main reason I expected that behavior is because I think that characterData is supposed to be used to track text change of elements, but no one ever changes an element text by changing a text of it's textNode. Everyone just changes textContent or innerHTML of an element itself. I may be wrong though. And if I wrong, then there's no an effective way to track elements text?

Comment 1

[Nickolay_Ponomarev]

Thanks for the testcase!

I can reproduce (current nightly), and no obvious duplicates, but didn't check against the spec.

CCing smaug, who seems to be working on this.

Comment 2

[Boris Zbarsky [:bzbarsky]]

I think per spec you're wrong, and the only way to track text is to track both characterData and insert/remove of textnodes.

Per spec, mutations records with characterData are only queued in the "replace data" algorithm <https://dom.spec.whatwg.org/#concept-cd-replace>.  This algorithm is invoked from the following places:

1)  The "data" setter on the CharacterData interface.
2)  The appendData/insertData/deleteData/replaceData methods on that interface.
3)  The various things that can "split" a text node (splitText, and some range
    operations).
4)  The nodeValue setters on Text, Comment, and ProcessingInstruction nodes.
5)  The textContent setters on Text, Comment, and ProcessingInstruction nodes.
6)  normalize().
7)  Some other range operations (extracting, etc).

Comment 3

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

Yeah, in cases #1 and #2 textContent and innerHTML is used, and those first remove all the child
nodes the span has, and then adds new child node(s). No characterData changes here.