>> This patch tries to solves this issue.
>
> Can you tell how?  All I see is a different value for colwidth.  I
> guess I'm missing something.

Yes but this different colwidth would leave more space between a
keybinding and the "selection character" of the next column.  But, I
have to say that it would be hard to prove that it will work for every
possible settings.

>> Here, how I justify the modification of `colwidth':
>> 
>>       - I don't think we need the "(min 30)" part since, if the frame is
>>         wide enough, we always get a colwidth of 30.
>>         
>>       - I don't think "(window-width)" is what we need since, by
>>         default, the *Completions* buffer will use the full frame width.
>
> Martin, is that guaranteed?
>
> And even if it is, what's the harm in keeping window-width?

I don't think that a full frame width *Completions* buffer is
guaranteed: it is only what I see with "emacs -Q".

Keeping window-width in this calculation seems a bit strange because, by
default, it has nothing to do with the *Completions* buffer window
width.

> Finally, does this change some user-facing aspect of the tmm behavior?
> If so, maybe we need a NEWS entry.

I don't think there is behavioral change only a layout change here.
-- 
Manuel Giraud