bug#71927: 29.4; ibuffer-do-isearch and ibuffer-do-isearch-regexp not prompting for input

[Stephen Berman via "Bug reports for GNU Emacs, the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>]

On Thu, 04 Jul 2024 17:12:43 +0300 Eli Zaretskii <eliz@gnu.org> wrote:

>> From: Stephen Berman <stephen.berman@gmx.net>
>> Cc: me@eshelyaron.com,  kickingvegas@gmail.com,  71927@debbugs.gnu.org,
>>   basil@contovou.net,  jpw@gnu.org,  juri@linkov.net
>> Date: Thu, 04 Jul 2024 15:32:05 +0200
>>
>> On Thu, 04 Jul 2024 15:50:56 +0300 Eli Zaretskii <eliz@gnu.org> wrote:
>>
>> >> From: Stephen Berman <stephen.berman@gmx.net>
>> >> Cc: Eshel Yaron <me@eshelyaron.com>,  kickingvegas@gmail.com,
>> >>   71927@debbugs.gnu.org,  basil@contovou.net,  jpw@gnu.org,
>> >>   juri@linkov.net
>> >> Date: Thu, 04 Jul 2024 12:06:40 +0200
>> >>
>> >> >> FWIW, AFAICT everything is working correctly, it's just that the
>> >> >> "Operation finished" message hides the prompt.  ibuffer-do-isearch
>> >> >> should tell define-ibuffer-op not to display that message, somehow.
>> >> >
>> >> > I don't see how this could be considered "correct": the "Operation
>> >> > finished" message is supposed to be shown only after the Isearch is
>> >> > finished in all the marked buffer, not before.  It looks like we need
>> >> > a function that will not return until all the buffers where searched,
>> >> > because that's what define-ibuffer-op expects.  Don't you agree?
>> >>
>> >> The attached patch appears to DTRT, but I only tested it briefly.
>> >
>> > Thanks, but does it really make sense to change define-ibuffer-op to
>> > fix the functions which (evidently) use it incorrectly?  I think the
>> > fix needs to be in the same place where the bug is.
>>
>> That's what I thought, too, but I don't see a way to do that, because
>> define-ibuffer-op hard-codes displaying a message at the end of the
>> body.  But in this case that message seems superfluous, in addition to
>> hiding the isearch prompt, because when isearch goes through the marked
>> buffers, it then displays its own message that the operation is
>> complete.  So this looks like a case when define-ibuffer-op is currently
>> too rigid.
>
> Then maybe these two commands should be defined using defun, not
> define-ibuffer-op?  There's nothing forcing us to define each ibuffer
> command via define-ibuffer-op, is it?

No, but doing that does seem rather like admitting that
define-ibuffer-op in its current form isn't up to the job in this case,
which is basically what my patch addressed.  But of course that doesn't
mean it's the best fix.

>> However, I acknowledge that I don't really grok the interaction with
>> isearch, i.e., why the "Operation finished" message is shown
>> although isearch hasn't even begun.
>
> Juri will know for sure, but AFAIU all the commands that invoke
> Isearch just enter a special mode (including the conditions/commands
> to exit Isearch).  IOW, when isearch-forward returns, the search did
> not yet begin; instead, Emacs is in a special mode where typing
> characters adds them to search string and triggers another round of
> search.

But if isearch-forward returning is what makes ibuffer-do-isearch (via
define-ibuffer-op) emit the "Operation finished" message, then it seems
like the only alternatives to suppressing that message are either to
somehow postpone isearch-forward's returning or to use some other
mechanism to initiate the actual search.  Maybe one of those is possible
for someone to implement who understands isearch better than I do.

Steve Berman