Eli Zaretskii <eliz@gnu.org> writes:

Hi Eli,

>> > I think I just fixed it there, please take a look.
>> 
>> That's what I was about to do, too, but that's not right, too.
>
> Well, it fixes the regression.

That's true.

>> Say you have dired buffers for
>> 
>>   ~/
>>   ~/foo/
>>   ~/foo/bar/
>> 
>> and then delete ~/foo from inside the ~/ dired buffer.  You'll be
>> asked if the buffers for foo should be delete, too, and when you
>> confirm, what is deleted are the buffers ~/ and ~/foo but ~/foo/bar
>> persists.  So the buffers are deleted downwards (to the root) instead
>> of upwards which is wrong.
>
> This means your fix for the "ask" part is incomplete, and should be
> improved.  But that is a new feature in Emacs 28, so it is not a
> catastrophe if it is imperfect.

It's still a major annoyance since the feature is not opt-in.

> Regressions in previously correct behavior are much worse.

Of course.

> Of course, if you can come up with a fix for the question-asking part
> that makes it delete all the relevant buffers, and if that fix is safe
> enough (a high bar at this late stage of the pretest), we can install
> that on the release branch.  Failing that, the fix for that will have
> to wait till Emacs 28.2 at the very least.

Here's a patch.  I've created a separate cond-arm for the SUBDIRS case,
so any callers which are not dired-clean-up-after-deletion (the only one
setting SUBDIRS) are not affected.