Stefan Monnier <monnier@iro.umontreal.ca> writes:

>>> I suspect it might have been when a function is advised but not yet defined.
>> Hmm, how could this happen?
>
> It happens all the time: just call `advice-add` before the function is defined.

I hardly see how one would do this, sorry.

>> The problem is that with Helm completion I have an extra unknown
>> symbol on top of list when I start typing (this is expected when
>> require-match is non-nil),
>
> Could you characterize this "unknown symbol" a bit more?  I'm having
> trouble guessing why/how/where `confirm` would have such an effect.

In Helm, if require-match is 'confirm, when you write in minibuffer something
that doesn't match one of the candidates this string is appended on top
of list and is prefixed (with display prop) with [?]. By contrast when
require-match is nil nothing is appended on top of list and pressing RET
doesn't exit minibuffer (helm-buffer is empty in such case).
IOW the behavior of require-match is the same than with vanilla Emacs.

> Maybe we can fix that without preventing the use of "not quite
> defined" functions.

Sorry, as said above I don't understand what you want to achieve here.

-- 
Thierry