bug#44500: 26.3; `prefix-command-echo-keystrokes-functions' causes regression

bug#44500: 26.3; `prefix-command-echo-keystrokes-functions' causes regression

[Drew Adams]

This variable was introduced in Emacs 25, apparently.

I'm not happy that it's used/implemented the way it is, by default.  I
have a function that calls `read-char' in a loop, and it took me a while
to figure out why this code broke starting with Emacs 25.  That function
is called by a command bound to `P B', where `P' is a prefix key.

I'm not sure why, based on the doc of the variable etc., but even though
`P B' is not itself a prefix key, the default value of the variable
interfered with the normal behavior in the echo area.  After finding out
the cause, I now bind the var to nil in that part of my code.  But this
variable shouldn't have the effect it has, IMO.

Here's the function I use:

(defun bmkp-bmenu-read-filter-input ()
  "Read input and add it to `bmkp-bmenu-filter-pattern'."
  (setq bmkp-bmenu-filter-pattern "")
  (let ((curr-bmk (bookmark-bmenu-bookmark)))
    (when (eq 'QUIT
              (let ((tmp-list                   ())
                    (bmkp-bmenu-title           bmkp-bmenu-title)
                    (bmkp-bmenu-filter-function bmkp-bmenu-filter-function)
                    (inhibit-quit               t)

                    ;; Had to add this binding to counteract the regression.
                    ;; (prefix-command-echo-keystrokes-functions ())

                    char)
                (catch 'bmkp-bmenu-read-filter-input
                  (while (condition-case nil
                             (setq char (read-char
                                         (concat "Pattern: "
                                                 bmkp-bmenu-filter-pattern)))
                           ;; `read-char' raises an error for non-char event.
                           (error (throw 'bmkp-bmenu-read-filter-input nil)))
                    (unless (or (not (fboundp 'characterp))
                                (characterp char)) ; E.g. `M-x', `M-:'
                      (throw 'bmkp-bmenu-read-filter-input nil))
                    (case char
                      ((?\e ?\r) (throw 'bmkp-bmenu-read-filter-input nil))
                      (?\C-g     (setq inhibit-quit nil)
                                 (throw 'bmkp-bmenu-read-filter-input 'QUIT))
                      (?\d       (or (null tmp-list)
                                     (pop tmp-list)
                                     t))
                      (t         (push (text-char-description char)
                                       tmp-list)))
                    (setq bmkp-bmenu-filter-pattern
                          (mapconcat #'identity (reverse tmp-list) ""))))))
      (message "Restoring display prior to incremental filtering...")
      (bookmark-bmenu-list 'FILTEREDP)
      (bmkp-bmenu-goto-bookmark-named curr-bmk)
      (message "Restoring display prior to incremental filtering...done"))))

This code worked fine prior to Emacs 25.  Starting with that release I
had to bind `prefix-command-echo-keystrokes-functions' to nil.  I
shouldn't have had to do that.

If I don't add that binding, then:

You don't always even see the prompt "Pattern: <etc.>".  Dunno why.

Instead, you see `P-', then `P B-', as if `P B' were a prefix key (only
`P' is a prefix key), and each char you type is also echoed as if it
were part of a prefix key, i.e., it's followed by a hyphen.

So if you type "red" after the (invisible) prompt, you see `P B r-',
then `P B r e-', then `P B r e d-'.  You should instead see `Pattern: r',
then `Pattern: re', then `Pattern: red'.

Binding the variable to nil fixes the problem.  But this behavior is a
bug, IMO, and the introduction of the variable shouldn't have changed
the default echo-area behavior of Emacs.

Hope the problem description is clear enough.

In GNU Emacs 26.3 (build 1, x86_64-w64-mingw32)
 of 2019-08-29
Repository revision: 96dd0196c28bc36779584e47fffcca433c9309cd
Windowing system distributor `Microsoft Corp.', version 10.0.18362
Configured using:
 `configure --without-dbus --host=x86_64-w64-mingw32
 --without-compress-install 'CFLAGS=-O2 -static -g3''

bug#44500: 26.3; `prefix-command-echo-keystrokes-functions' causes regression

[Stefan Kangas]

tags 44500 + moreinfo
thanks

Drew Adams <drew.adams@oracle.com> writes:

> This variable was introduced in Emacs 25, apparently.
>
> I'm not happy that it's used/implemented the way it is, by default.  I
> have a function that calls `read-char' in a loop, and it took me a while
> to figure out why this code broke starting with Emacs 25.  That function
> is called by a command bound to `P B', where `P' is a prefix key.
>
> I'm not sure why, based on the doc of the variable etc., but even though
> `P B' is not itself a prefix key, the default value of the variable
> interfered with the normal behavior in the echo area.  After finding out
> the cause, I now bind the var to nil in that part of my code.  But this
> variable shouldn't have the effect it has, IMO.
>
> Here's the function I use:
>
> (defun bmkp-bmenu-read-filter-input ()
[...]
>
> This code worked fine prior to Emacs 25.  Starting with that release I
> had to bind `prefix-command-echo-keystrokes-functions' to nil.  I
> shouldn't have had to do that.

Could you provide a minimal example to reproduce the issue?

> If I don't add that binding, then:
>
> You don't always even see the prompt "Pattern: <etc.>".  Dunno why.
>
> Instead, you see `P-', then `P B-', as if `P B' were a prefix key (only
> `P' is a prefix key), and each char you type is also echoed as if it
> were part of a prefix key, i.e., it's followed by a hyphen.
>
> So if you type "red" after the (invisible) prompt, you see `P B r-',
> then `P B r e-', then `P B r e d-'.  You should instead see `Pattern: r',
> then `Pattern: re', then `Pattern: red'.
>
> Binding the variable to nil fixes the problem.  But this behavior is a
> bug, IMO, and the introduction of the variable shouldn't have changed
> the default echo-area behavior of Emacs.
>
> Hope the problem description is clear enough.
>
> In GNU Emacs 26.3 (build 1, x86_64-w64-mingw32)
>  of 2019-08-29
> Repository revision: 96dd0196c28bc36779584e47fffcca433c9309cd
> Windowing system distributor `Microsoft Corp.', version 10.0.18362
> Configured using:
>  `configure --without-dbus --host=x86_64-w64-mingw32
>  --without-compress-install 'CFLAGS=-O2 -static -g3''

bug#44500: [External] : Re: bug#44500: 26.3; `prefix-command-echo-keystrokes-functions' causes regression

[Drew Adams]

> tags 44500 + moreinfo

> > Here's the function I use:
> >
> > (defun bmkp-bmenu-read-filter-input ()
> [...]
> >
> > This code worked fine prior to Emacs 25.  Starting with that release
> > I had to bind `prefix-command-echo-keystrokes-functions' to nil.  I
> > shouldn't have had to do that.
> 
> Could you provide a minimal example to reproduce the issue?

I think I've provided the info I have, and all
that should be needed to understand the problem.

> > If I don't add that binding, then:
> > You don't always even see the prompt "Pattern: <etc.>".  Dunno why.
> >
> > Instead, you see `P-', then `P B-', as if `P B' were a prefix key
> > (only `P' is a prefix key), and each char you type is also echoed
> > as if it were part of a prefix key, i.e., it's followed by a hyphen.
> >
> > So if you type "red" after the (invisible) prompt, you see `P B r-',
> > then `P B r e-', then `P B r e d-'.  You should instead see `Pattern:
> > r', then `Pattern: re', then `Pattern: red'.
> >
> > Binding the variable to nil fixes the problem.  But this behavior is
> > a bug, IMO, and the introduction of the variable shouldn't have
> > changed the default echo-area behavior of Emacs.

bug#44500: 26.3; `prefix-command-echo-keystrokes-functions' causes regression

[Lars Ingebrigtsen]

Drew Adams <drew.adams@oracle.com> writes:

>> > This code worked fine prior to Emacs 25.  Starting with that release
>> > I had to bind `prefix-command-echo-keystrokes-functions' to nil.  I
>> > shouldn't have had to do that.
>> 
>> Could you provide a minimal example to reproduce the issue?
>
> I think I've provided the info I have, and all
> that should be needed to understand the problem.

More information was requested, but no response was given within a
month, so I'm closing this bug report.  If progress can be made,
please respond to this email and we'll reopen the bug report.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no