Re: Updating *Completions* as you type

[Spencer Baugh <sbaugh@catern.com>]

Juri Linkov <juri@linkov.net> writes:

>>> I think preselecting the default value in the middle of completions
>>> would make sense only when completions were sorted by the order of the
>>> list of default values (from M-n M-n ...).  Then the first default value
>>> would be at the top of completions, and it would easier for users
>>> to navigate completions top-down.
>>
>> Yes, that was the case that inspired this specific idea: the case where
>> completions are sorted by the list of defaults, and so the default value
>> is the first completion.
>>
>> But it occured to me that there's no reason to limit to just the case
>> where the default value is the first completion.  We can just always
>> move point to the default wherever it occurs in the completions, as long
>> as it occurs somewhere.  Seems like a nice improvement.
>>
>> Although I guess it does cause M-<down> to no longer go to the first
>> completion, and M-<up> to no longer go to the last one.  Which is a bit
>> annoying.
>
> Indeed.
>
>> Maybe we should have bindings to move to the first and last completions.
>> Whatever they are, they should also work in completion-in-region-mode,
>> though, so M-< and M-> definitely won't work... Maybe M-0 M-<down> to go
>> to the first, and M-0 M-<up> to go to the last?  (Or vice versa?)
>
> M-< and M-> work in the minibuffer as well as in the normal buffer.
> But they don't work with your first patch ;-)

Right right, I was just saying one reason we can't use M-< and M-> for
said bindings.

Anyway, I realized that we already have a nice way to move to the first
and last completions: ? M-<up> and ? M-<down>.

That is, run minibuffer-completion-help to recreate *Completions* and
wipe out the current value of point, then go down or up to access the
first or last.

Seems great so let's just stick with that.

>>> BTW, while looking at this case I found a problem with your first patch:
>>> after navigating in *Completions* and switching back to the minibuffer
>>> point is reset to the beginning of the *Completions* buffer.
>>
>> Good catch!  If you see my other patch which I posted in this thread,
>> "Keep point on the same completion in the completions buffer", that is a
>> nice orthogonal improvement which incidentally also fixes that bug in
>> completions-auto-update.
>
> Your third patch adds a new feature that just hides the bug by forcing
> the previous position over the wrong new one.  I think the proper fix
> would be to add more commands to completions-no-auto-update-commands
> in your first patch.  Or to change it from opt-out to opt-in by renaming
> to completions-auto-update-commands.

I would much rather not list commands as opt-in or opt-out at all, I'd
rather all commands just work without special cases.  So I'm working on
features to allow that.

I don't agree with the characterization of "it just hides the bug".  The
bug is that point in *Completions* gets wiped out by auto-updating.  My
third patch preserves point across auto-updating.  That's very directly
solving the bug.

>> OK, so maybe the thing I want to propose is not "we'll change the
>> default to whatever the current selected completion is" but instead "RET
>> with an empty minibuffer will submit the current selected completion".
>> That's equivalent, if I drop the ideas of changing the prompt and
>> changing M-RET.
>
> Agreed, I think the idea of changing the prompt on the fly won't fly.
>
>>>> I think this has some nice benefits in reducing the number of concepts
>>>> people need to track.  If the minibuffer is empty, they can just use
>>>> minibuffer-next-completion a few times followed by RET to select a
>>>> completion, no need to use M-RET.  Plus, the new M-RET and C-u M-RET
>>>> would be useful even to users who don't use minibuffer-next-completion.
>>>> True, good analysis.
>>>
>>> Specifically though it's about the case when the minibuffer is empty.  I
>>> think it would be nice for RET to submit the highlighted candidate in
>>> that case, if there is one.
>>>
>>> That matches icomplete-mode's behavior, actually, which is nice.
>>
>> Oh, actually it doesn't.  It matches ido-mode and fido-mode and a host
>> of completion mechanisms outside core, though.  I still think it's
>> desirable, at least as a user option.
>
> But then such idiosyncrasy of fido-mode causes a lot of bug reports
> like bug#55800.

But we would not have those kinds of issues because when completion
starts, by default, there is no highlighted candidate.