Created attachment 1463 [details]
Test cases: A file with 3 sets of tests.

Problem: 

1 For percent-encoded bytes in an activated link, some browsers try a
  literal match for the raw, percent-encoded string before they try a
  ”semantic” match for the string of decoded characters.
2 When this happens, the URL can target two different destinations - 
  e.g. an idref that matches the raw percent-encoded string plus an 
  idref that matches the decoded characters. If the raw string exists 
  in an idref, it will ”win” over the an idref with the decoded string.
3 The URL spec clarify whether testing for a literal match before 
  before testing for a decoded match is permitted.

Affected browsers:

          (*Not* affected: Firefox and IE11.)

* Blink:  problem manifests itself only if author uses percent-encoded
          strings as URL value (e.g. href='#%C3%A5' or href="#%61").
* Webkit: problem manifests itself even for directly typed URLs when the
          code points are higher than U+009F (in which case the coded 
          points are converted to percent-encoded bytes by the URL
          parser). Affects href="å" + also href="å" href="å"

Example - how to replicate:

1) To a document with this fragment, <div  id="å">Fragment 1.</div>
2) add the following two links,
    a) <a href='#å'     >Link A.</a> - directly typed;
    b) <a href='#%C3%A5'>Link B.</a> - percentage encoded;
3) and activate Link A + Link B to verify that each targets Fragment 1
4) Now, add yet another fragment: <div  id='%C3%A5'>Fragment 2.</div>
4) Activate Link A and Link B, and check which fragment is targeted:

   Expected results: 
       * Both links should target Fragment 1. 
       * No link should target Fragment 2.

   Actual results:
       * IE11/Firefox: Both links targets Fragment 1.
       * Blink:        Link B targets Fragment 2.
       * Webkit:       Both links targets Fragment 2.