On Sat, Feb 12 2011, Stefan Monnier wrote:

> For more complex cases, there is also the issue of what to do when some
> parts of the completion are case-sensitive and other parts aren't
> (e.g. completion of case-sensitive envvars in case-insensitive file
> names), although this is less important for  completion-at-point than
> for minibuffer completion since you don't have to return a table that
> covers the completion of the whole field (composed of file names and
> env-vars, for example), and instead you can just limit the completion to
> the particular subfield.

There's still something wrong to me in the solution provided by Tassilo.
It works fine, but it is returning a function to bypass the usual
completion code completion code. Ignoring case, like doing smarter
completion (e.g. where the typed prefix does not match the returned
choices at all) is something that is very useful.

Therefore I wonder if we should either:
- Edit `completion-at-point-functions' docstring to remove the word
  "discouraged" in that sentence:
  "or a function of no argument to perform completion (discouraged),";
- Make completing code allows to replace the region being completed with
  somethig that does not match at all.

-- 
Julien Danjou
❱ http://julien.danjou.info