bug#14061: 24.3.50; Globbing in completion not working correctly

bug#14061: 24.3.50; Globbing in completion not working correctly

[Michael Heerdegen]

Hi,

In emacs -Q, eval

  (require 'bytecomp) ; just for the defvar

Now 

   C-h v *compil*warn* TAB TAB

"byte-compile-warnings" should be among the matches, but that's not
the case.


Thanks,

Michael.




In GNU Emacs 24.3.50.1 (x86_64-pc-linux-gnu, GTK+ Version 3.4.2)
 of 2013-03-17 on dex, modified by Debian
 (emacs-snapshot package, version 2:20130317-1)
Windowing system distributor `The X.Org Foundation', version 11.0.10707000
System Description:	Debian GNU/Linux 7.0 (wheezy)

bug#14061: 24.3.50; Globbing in completion not working correctly

[Michael Heerdegen]

Michael Heerdegen <michael_heerdegen@web.de> writes:

>    C-h v *compil*warn* TAB TAB
>
> "byte-compile-warnings" should be among the matches, but that's not
> the case.

Looks like `completion-pcm--merge-completions' is introducing the error:

(completion-pcm--merge-completions
 '("byte-compile-error-on-warn"
   "byte-compile-warning-types"
   "byte-compile-warnings"
   "byte-compile-last-warned-form")
 '(star "compil" star "warn"))

==>

(any "warn" "-" star "compile-" "byte-")
             ^               ^
Note the two hyphens.


Regards,

Michael

bug#14061: 24.3.50; Globbing in completion not working correctly

[Michael Heerdegen]

Michael Heerdegen <michael_heerdegen@web.de> writes:

> >    C-h v *compil*warn* TAB TAB
> >
> > "byte-compile-warnings" should be among the matches, but that's not
> > the case.
>
> Looks like `completion-pcm--merge-completions' is introducing the error:
>
> (completion-pcm--merge-completions
>  '("byte-compile-error-on-warn"
>    "byte-compile-warning-types"
>    "byte-compile-warnings"
>    "byte-compile-last-warned-form")
>  '(star "compil" star "warn"))
>
> ==>
>
> (any "warn" "-" star "compile-" "byte-")

Some more thoughts:

In the definition of `completion-pcm--merge-completions', we have

  (when (memq elem '(star point prefix))
    ;; Extract common suffix additionally to common prefix.
    ;; Only do it for `point', `star', and `prefix' since for
    ;; `any' it could lead to a merged completion that
    ;; doesn't itself match the candidates.
    (let ((suffix (completion--common-suffix comps)))
      (cl-assert (stringp suffix))
      (unless (equal suffix "")
        (push suffix res))))

In the pathological cases, the `suffix' includes a hyphen that is
already included in the prefix:

  (completion-pcm--merge-completions
   '("byte-compile-error-on-warn"
     "byte-compile-warning-types"
     "byte-compile-warnings"
     "byte-compile-last-warned-form")
   '(star "compil" star "warn"))       ;; suffix == "-"

  (completion-pcm--merge-completions
   '("byte-compile-a-warning-types"
     "byte-compile-a-test-warning-types")
   '(star "compile" star "types"))     ;; suffix == "-warning-"

but not all cases where `suffix' starts with "-" are problematic:

  (completion-pcm--merge-completions
   '("byte-compile-a-c-b-warning-types"
     "byte-compile-a-test-b-warning-types")
   '(star "compile" star "types"))

so it would be wrong to remove a leading "-" from `suffix'
unconditionally.

Dunno what a good fix would look like.  It tried

  (let* ((pre-length (length prefix))
         (suffix (completion--common-suffix
                  (mapcar (lambda (comp) (substring comp pre-length))
                          comps))))
    (cl-assert (stringp suffix))
    (unless (equal suffix "")
      (push suffix res)))

which seems to work, but my insight in this code is limited.


HTH,

Michael.

bug#14061: 24.3.50; Globbing in completion not working correctly

[Stefan Monnier]

> Dunno what a good fix would look like.  It tried

>   (let* ((pre-length (length prefix))
>          (suffix (completion--common-suffix
>                   (mapcar (lambda (comp) (substring comp pre-length))
>                           comps))))
>     (cl-assert (stringp suffix))
>     (unless (equal suffix "")
>       (push suffix res)))

> which seems to work, but my insight in this code is limited.

You were right on the money.  I installed a slightly refined fix, which
tries to optimize away a few cases.
Thank you,


        Stefan