Possible bug with coderef highlighting in HTML export

[Thibault Marin <thibault.marin@gmx.com>]

Hi all,

I am using Org mode version 9.1.3 (release_9.1.3-216-g259656 @
/.../org-mode/lisp/) and I am experiencing an issue with the exported
HTML of the following org file:

,----
| this links get highlighted [[(link0)]].
| 
| this link does not get highlighted [[(link1)]].
| 
| #+begin_src emacs-lisp -r
| 0(ref:link0)
| #+end_src
| 
| #+begin_src emacs-lisp -r
| 
| 1(ref:link1)
| #+end_src
`----

The relevant HTML output is:

,----
| <p>
| this links get highlighted <a href="#coderef-link0" class="coderef" onmouseover="CodeHighlightOn(this, 'coderef-link0');" onmouseout="CodeHighlightOff(this, 'coderef-link0');">1</a>.
| </p>
| 
| <p>
| this link does not get highlighted <a href="#coderef-link1" class="coderef" onmouseover="CodeHighlightOn(this, 'coderef-link1');" onmouseout="CodeHighlightOff(this, 'coderef-link1');">1</a>.
| </p>
| 
| <div class="org-src-container">
| <pre class="src src-emacs-lisp"><span id="coderef-link0" class="coderef-off">0</span>
| </pre>
| </div>
| 
| <div class="org-src-container">
| <pre class="src src-emacs-lisp">1
| </pre>
| </div>
`----

The issue is that the link to the line in the second source block is not
highlighted (it does not get the "coderef-off" span markup).

I tried to dig a little and it appears that `org-html-do-format-code' does not
handle empty lines at the beginning of a block.  `(org-split-string "\n1")'
returns '("1") which looses the first empty line.  The line reference received
via the `refs' variable (which has value ((2 . link1))) then becomes out of sync
with the `code' variable (split by lines) used for formatting of the code block.

I am not sure what is the best way to handle this:

1. Should the `refs' variable be built accounting for the top empty lines?
2. Alternatively, should the `org-html-do-format-code' and
   `org-export-format-code' functions count the number of top empty lines and
   adjust the line number accordingly?
3. Should top empty lines be completely deleted, before the `refs' array is
   built?

I can try to propose a patch if the best option can be decided.  Option 2 seems
relatively simple but feels like a hack.

I would appreciate any suggestions on how to best fix this.

Thanks in advance.

thibault