Domain relaxation partially broken in FF

[wwalc]

Under some conditions, FF is throwing this error:

Permission denied for <​http://dev.ckeditor.local:8080> (document.domain has not been set) to get property Window.document from <​http://dev.ckeditor.local:8080> (document.domain=<​http://ckeditor.local>).

(confirmed in FF 3.5.7).

It seems that this bug has been introduced when fixing #4514. Reverting [4492] removes that particular error.

Steps to reproduce

• Using release version of CKEditor, open the following URL in FF: ​http://dev.ckeditor.local:8080/ckeditor/_dev/releaser/release/release/_samples/replacebycode.html (note: there is a port number in the url)
• edit replacebycode.html sample and add there:

  <script type="text/javascript">
  document.domain = "ckeditor.local";
  </script>
  

  after

  <script type="text/javascript" src="../ckeditor.js"></script>
  

  (make sure you're editing replacebycode.html, it seems that it does not happen in replacebyclass.html)
• reload ​http://dev.ckeditor.local:8080/ckeditor/_dev/releaser/release/release/_samples/replacebycode.html
• result:

  Permission denied for <http://dev.ckeditor.local:8080> 
  (document.domain has not been set) to get property Window.document 
  from <http://dev.ckeditor.local:8080> 
  (document.domain=<http://ckeditor.local>).
  

[fredck]

Let's see if we're able to have it done for the 3.2.

[garry.yao]

First of all I can't reproduce the bug by following the above steps.

While Wiktor's finding here clue me in another TC which breaks the editor easily: load the editor inside a cross-domain frame, see ticket TC for details:
​run OR ​view source.

[4492] has introduced a pretty nasty piece of window.top where all bugs root from, the problem is that we should never assuming a domain friendly environment.
Back to the original purpose of that codes was to make sure the fix also applied in case editor was placed inside a frame. But by further checking it today, I found the B/F cache doesn't even impact the iframe page at all, so we could safely scope #4514 to host page only.

[garry.yao]

As states above, 5091.patch is assumed to be resolving both the problem in ticket description and the one in my comment.

[garry.yao]

Fixed with [5046].