DataProcessor processing nested (encoded) attributes

[Olek Nowodziński]

1. Open replacebycode sample.
2. Enable allowedContent: true

   CKEDITOR.replace( 'editor1', {
   	allowedContent: true
   } );
   

3. Call

   CKEDITOR.instances.editor1.dataProcessor.toHtml( '<p><img data-foo="&lt;a href=&quot;XXX&quot;&gt;YYY&lt;/a&gt;" /></p>' )
   

   data-foo attribute contains encoded HTML:

   <a href="XXX">YYY</a>
   

Expected:

<p><img data-foo="&lt;a href=&quot;XXX&quot;&gt;YYY&lt;/a&gt;" /></p>

Actual:

<p><img data-foo="&lt;a data-cke-saved-href=&quot;XXX&quot;&gt;YYY&lt;/a&gt; href=&quot;XXX&quot;&gt;YYY&lt;/a&gt;" /></p>

What happened? Dataprocessor considered href="XXX" as an attribute of <img> and tried to protect it, making a lot of mess. The RegExp responsible for that is not perfect

protectAttributeRegex = /\s(on\w+|href|src|name)\s*=\s*(?:(?:"[^"]*")|(?:'[^']*')|(?:[^ "'>]+))/gi;

|(?:[^ "'>]+) matches href="XXX" as an unquoted attribute of <img> (​this kind of attributes is allowed in HTML)

---

Another TC:

Paste

<p><img data-foo="&lt;a href=&quot;XXX&quot;&gt;YYY&lt;/a&gt;" /></p>

into source view and switch modes twice.

[Olek Nowodziński]

Pushed a fix to ​t/11508 (+tests).

Note: Although I didn't notice performance loss (quite contrary in fact, which is funny) it may still be an issue because changes in RegExp made it more general. Watch out.

[Frederico Caldeira Knabben]

I was able to notice a small decrease on performance on setDate with this code (between 5-10%), but it is hard to measure it for sure. Still it has little impact on the overall editor experience. If we really want to talk about performance, then better to try to not have that code at all, which is a much bigger task.

I don't see better ways to solve this problem though. I just pushed a minor simplification to the regex and I think the proposed solution is fine.

[Olek Nowodziński]

​git:abb150f landed in master (eb840d1 tests).