behavior:url(#default#savehistory) causes event error on oldIE

[dwilks]

In IE8

• ​http://jsfiddle.net/ZpgGT/
• update jquery version as appropriate
• Run

Result: Object doesn't support this property or method.

The key conditions:

• body must have a change event
• the dynamically appended element must have an associated style with a behavior:url(#default#savehistory)

Full description at ​http://pdfcast.org/pdf/jquery-calendar-problem

(This is a re-post of the incorrectly closed #12178)

[dmethvin]

Still in -git:

​https://github.com/jquery/jquery/blob/0ee94159023bebe1992c5281d0f4778b4f75ff0d/src/manipulation.js#L843

Seems that when a behavior is attached to an element, typeof elem.removeAttribute === "unknown" and the mere act of asking if the property exists causes it to call the method (incorrectly), similar to the problems that Sizzle works around here:

​https://github.com/jquery/sizzle/blob/5e6ae330a9ccb33fd343a0bdc73ed2b821767f9c/sizzle.js#L599

We'll need a trickier test like !/undefined|unknown/.test( typeof elem.removeAttribute ) perhaps.

[Timmy Willison]

@dmethvin: I don't think the Sizzle code is related. ;P

[dmethvin]

@timmywil, I think it's the same basic problem, a simple falsy-check for elem.getAttribute was calling it and here it seems to be elem.removeAttribute instead.

[Timmy Willison]

@dmethvin: The falsey check is on the return value of getAttribute("class") rather than on the getAttribute method itself. Wouldn't the code just fail if it tried to call a function that IE reports to be type "unknown"? I imagine that Sizzle still doesn't handle that issue.

I think you're thinking of the falsey check we used to use: if ( elem.getAttribute ) { ... } Unfortunately, IE calls the method in XML (and some other edge cases) even without parens. There wasn't a performance loss in using typeof, which never accidentally calls the method, so we switched to that. Your solution of a regex on typeof would work, but I'm not sure we want to add something like that.

I'm inclined to say this is not high priority. The behavior property is proprietary IE and can most likely cause more problems than this.

[gibson042]

For the curious, it's on the order of +15 bytes gzipped.

[dmethvin]

This did seem to be the same problem where .removeAttribute() is being called, I changed to use a typeof and the test case no longer fails. Unfortunately the setup is really complicated and involves the flakey focus events which are async in IE so I am having trouble creating a reliable unit test.

[Dave Methvin]

Fix #12863. Prevent oldIE from calling .removeAttribute

Changeset: 4e0bc169df0f6dee45fe7c19925216453b431ebb