bug#20063: 24.4: read-from-minibuffer improperly setting hist parameter

[Stefan Monnier <monnier@iro.umontreal.ca>]

>> That is indeed a problem, but it is more general than the case of
>> HIST=nil, since even if HIST is non-nil the history may contain entries
>> which are not valid according to COLLECTION.
>> So what we need to do is to filter out those entries dynamically.
> Yes, if you mean once, at the time the function is invoked;

No, I was thinking of doing it on-the-fly when navigating in the
history.

> but the benefit of this to the end-user is very limited, and has
> a downside if done simply.

If the benefit is limited, it means the problem you mention is
correspondingly minor.

> Once REQUIRE-MATCH=t, nothing but elements of COLLECTION will ever be
> accepted, so `concat'-ing the filtered elements of HIST would present
> only legitimate options, in the sequence most recently used, but with
> potentially a lot of duplicate entries.

I'm not sure exactly what you mean here, but I suspect you assume
COLLECTION to be finite and small.

> Using `add-to-list', starting with an empty list would avoid
> the duplications and present the elements in sequence
> most-recently-used.

Duplicate elements in the history are yet again orthogonal.
You probably want to set history-delete-duplicates to t.

>> Actually, IIRC a value of t for HIST
> I don't see that in the v24.4 documentation for `completing-read' or

Indeed, it's not documented.  It's basically a side effect of t holding
a non-list value, which we've tried to make work in order for
read-password to behave properly (i.e. not have history since that
would be a security problem).

>> does provide the option of "no history".
> Which brings us full-circle to line 974 of `minibuf.c'

I don't understand this, since this code checks for a nil value, not
a t value.


        Stefan