Hello,

here is my take to fix the bug: there are three different versions of
`TeX-completing-read-multiple' as suggested by Tassilo.  I'm just not
sure it works with all 24.* versions of Emacs.  A serious problem is
that there are a couple of internal functions (crm--*), and we had
problems  in the past with internal functions,[1] for which we still
get bug reports nowadays.

Actually, the change by Roland Winkler which caused this bug is the
same which forced us to make `TeX-completing-read-multiple' always
return nil,[2] so perhaps we could also revert that change, but in
this way we would break backward compatibility in turn.

Bye,
Mosè

Note:

[1]  http://comments.gmane.org/gmane.emacs.aquamacs.devel/240
[2]  http://comments.gmane.org/gmane.emacs.auctex.devel/3375

2015-03-12 23:05 GMT+01:00 Mosè Giordano <mose@gnu.org>:
> Hi Tassilo,
>
> 2015-03-12 9:51 GMT+01:00 Tassilo Horn <tsdh@gnu.org>:
>> Mosè Giordano <mose@gnu.org> writes:
>>
>>>> PS: The plan is actually to move the minibuffer completion in that
>>>> direction for normal minibuffer completion, so that TAB is bound to
>>>> completion-at-point in the minibuffer as well.
>>>
>>> Uh, thanks for the suggestion, but according to git blame
>>> `completion-at-point' exists since 2009, so Emacs <23 is left out.
>>
>> Sorry, I didn't follow this issue closely.  But isn't it the case that
>> some quite recent change in Emacs is the culprit?
>
> Yes, with this commit
> http://git.savannah.gnu.org/gitweb/?p=emacs.git;a=commitdiff;h=9c44569ea2a18099307e0571d523d8637000a153
> `completing-read-multiple' ignores empty strings, which was essential
> for completion of `siunitx' units.
>
>> In that case,
>> `TeX-completing-read-multiple' could have 3 versions instead of the
>> current two (one for Emacs, one for XEmacs), i.e., one for Emacs <= 22,
>> one for Emacs 23+, and one for XEmacs.
>
> As I said some messages ago, the best I can think is to copy the last
> Emacs implementation before the mentioned commit (making sure return
> value for empty input is always nil) for all Emacs versions, so we
> should be able to keep the number of versions of
> `TeX-completing-read-multiple' as low as two.  The problem will be of
> course some more work on our side to maintain the function, if needed.
>
> Bye,
> Mosè