On 21.07.2021 04:34, Stefan Monnier via Bug reports for GNU Emacs, the 
Swiss army knife of text editors wrote:

>>> I haven't followed the discussion, so I might be off, but I think what
>>> is described above is something you can get simply by passing the
>>> appropriate value for `mustmatch` to `completing-read` and friends
>>> (including `read-file-name`).
>> Probably not: most third-party completion functions out there don't support
>> any non-nil values of that argument other than t (or treat them like t).
> 
> Their loss.

Actually, ivy-completing-read treats 'confirm' the same as nil.

Oh well.

>> And suppose we chose to use 'confirm', would a neutral prompt "Confirm"
>>   without clarification, as opposed to something like "File does not exist;
>> Create?", be our best choice?
> 
> The same question comes up for `C-x C-f` (and `C-x C-b`).
> 
> So far the answer we have chosen is "yes".  We can revisit it,
> of course.  Just like we may want to revisit the way `mustmatch` works,
> but at least there is an existing "standard protocol" to get that kind
> of behavior and I can't think of a good reason why `project-find-file`
> should behave very differently from `find-file` in this respect.

Fair enough.

Trying this approach, I seem to recall the main problem we tried to 
solve originally: the string passed as DEFAULT was returned as the value 
entered by the user if they simply pressed RET, with no additional check 
for whether it actually matches any of the file names (or the "Confirm" 
prompt).

The attached patch seems to solve that. Does the behavior look good to 
everyone?