fix($compile): use createMap() for $$observe listeners when initialized ...

[jbedard]

...from attr interpolation

This alternate initialization of $$observers was missed in a27d827.

[googlebot]

CLAs look good, thanks!

[lgalfaso]

It is my understanding that we are going to use objects without prototype
only if they are not accessible by developers (even for private properties)

[jbedard]

@lgalfaso $$observers already uses createMap, its just this alternate attr-interpolation case that still initializes it as a plain object.

The test failure in IE is interesting though...

[lgalfaso]

mmm... it looks like this was relax some time ago. Then it LGTM

[lgalfaso]

landed as 8e28bb4

[pkozlowski-opensource]

@jbedard @lgalfaso seems like this change upsets IE: https://travis-ci.org/angular/angular.js/jobs/43993200

It looks to me as a legitimate failure so we need to either roll-back or fix this one.

[lgalfaso]

@pkozlowski-opensource you are right, reverting this now

[lgalfaso]

reverted with f389f86

[pkozlowski-opensource]

Cool - just a minor nit - I think that the commit message for revert should be revert(..): ...

[lgalfaso]

@pkozlowski-opensource done

[lgalfaso]

this is not ready to be merged (at least, until IE is happy with it), moving it to 1.3.x

[caitp]

@lgalfaso you're shadowing properties from the prototype chain --- you probably need to change attrs.hasOwnProperty(key) to hasOwnProperty.call(attrs, key) --- but this won't help performance

Also, valueOf and toString are really stupid properties to shadow as non-functions for any object, because they will cause '' + object to throw

[jbedard]

I don't understand why only IE failed. But I think it makes sense to not support hasOwnProperty being overridden so I've removed it from the test. This commit wasn't trying to add support for anything but just fix the case where $$observers was initialized in this alternate path.

[caitp]

@jbedard --- I think the reason is that String/Number/Boolean host objects in IE probably don't support hasOwnProperty(), but do in other browsers. Edit: nahhh that's not it. I dunno =]

[jbedard]

The test was causing attrs.hasOwnProperty(key) to basically be "2"(key) which I'd think would fail in every browser. I dunno either :|

[caitp]

Yeah exactly, it's weird.

[jbedard]

@lgalfaso I updated the test to not break in IE. I guess attributes named 'hasOwnProperty' is not supported which makes sense. I still don't understand how non-IE passed though :|

[jbedard]

Is this ok to merge without the breaking test?

[lgalfaso]

LGTM

[caitp]

thanks @jbedard --- @lgalfaso any particular reason why this was not re-merged? I'll take a look tomorrow