completion-cycle-threshold and a suggestion

completion-cycle-threshold and a suggestion

[Leo]

Hello Stefan,

Thank you for adding more features to completion.

I have started using this completion-cycle-threshold. Incidentally I
have written a pcompletion for git and I have this function:

;; (pcmpl-git-string-lessp "word" "word-")  ; => nil
;; (pcmpl-git-string-lessp "words" "word-") ; => t
(defun pcmpl-git-string-lessp (s1 s2)
  "Compare strings S1 and S2 but treat '-' specially."
  (let ((res (compare-strings s1 0 nil s2 0 nil))
        c1 c2)
    (if (eq res t) (setq res 0))
    (ignore-errors
      ;; the following setq could trigger errors
      (setq c1 (aref s1 (1- (abs res)))
            c2 (aref s2 (1- (abs res))))
      (if (or (= c1 ?-) (= c2 ?-)) (setq res (- res))))
    (if (> res 0) nil t)))

to address the sorting of things like:

remote
remote-ftp
remote-ftps
remote-http
remote-https

i.e. when I 'git rem<TAB>' I want to see remote first before seeing the
rest.

The current completion cycle has this issue, for example, assume there
are two possible completions 'feature' and 'feature-old', feature-old is
seen first. This is counter-intuitive to the behaviour without
completion cycle.

Best wishes,

Leo

Re: completion-cycle-threshold and a suggestion

[Stefan Monnier]

> The current completion cycle has this issue, for example, assume there
> are two possible completions 'feature' and 'feature-old', feature-old is
> seen first. This is counter-intuitive to the behaviour without
> completion cycle.

I don't understand.  AFAIK the completion-cycle code is careful to order
the options using a heuristic based on the length of the completions
(choosing shorter ones first).

So are you saying that in your experience, the heuristic gives bad
results, or that the heuristic somehow was not applied?


        Stefan

Re: completion-cycle-threshold and a suggestion

[Leo]

On 2010-05-22 01:57 +0100, Stefan Monnier wrote:
>> The current completion cycle has this issue, for example, assume there
>> are two possible completions 'feature' and 'feature-old', feature-old is
>> seen first. This is counter-intuitive to the behaviour without
>> completion cycle.
>
> I don't understand.  AFAIK the completion-cycle code is careful to order
> the options using a heuristic based on the length of the completions
> (choosing shorter ones first).
>
> So are you saying that in your experience, the heuristic gives bad
> results, or that the heuristic somehow was not applied?

Seems so. But I am not sure which is at fault.

1. Grab magit.el (http://github.com/philjackson/magit) and load it into emacs
2. C-u M-x magit-status
3. Select a 'git' repo that has two branches: feature-old feature
4. type 'b' and switch to 'feature-old'
5. Now type 'b' again and feature-old is completed first.

Leo

Re: completion-cycle-threshold and a suggestion

[Stefan Monnier]

>> So are you saying that in your experience, the heuristic gives bad
>> results, or that the heuristic somehow was not applied?
> Seems so.

Which is it?
- the heuristic gives bad results?
- the heuristic somehow was not applied?


        Stefan

Re: completion-cycle-threshold and a suggestion

[Leo]

On 2010-05-22 13:59 +0100, Stefan Monnier wrote:
>>> So are you saying that in your experience, the heuristic gives bad
>>> results, or that the heuristic somehow was not applied?
>> Seems so.
>
> Which is it?
> - the heuristic gives bad results?
> - the heuristic somehow was not applied?
>
>
>         Stefan

The completion shows 'feature-old' before 'feature', isn't this what the
heuristic tries to fix? Sorry I haven't looked at the code. Simply
observing the behaviour.

Leo

Re: completion-cycle-threshold and a suggestion

[Stefan Monnier]

>>>> So are you saying that in your experience, the heuristic gives bad
>>>> results, or that the heuristic somehow was not applied?
>>> Seems so.
>> Which is it?
>> - the heuristic gives bad results?
>> - the heuristic somehow was not applied?
> The completion shows 'feature-old' before 'feature',

Ok, thanks.

> isn't this what the heuristic tries to fix?

Exactly, so it's apparently a bug somewhere.


        Stefan

Re: completion-cycle-threshold and a suggestion

[Stefan Monnier]

> 1. Grab magit.el (http://github.com/philjackson/magit) and load it into emacs
> 2. C-u M-x magit-status
> 3. Select a 'git' repo that has two branches: feature-old feature
> 4. type 'b' and switch to 'feature-old'
> 5. Now type 'b' again and feature-old is completed first.

That's because the code gives preference to entries that were used
recently (i.e. appear in the history).

In the above example, I guess what we'd really want is to eliminate
`feature-old' from the list since it's the current setting (like we
remove the current-buffer from the C-x b completion).


        Stefan