bug#36190: 27.0.50; `put-text-property' etc. with buffer argument calls current buffer's `after-change-functions'

[Pip Cet]

On Thu, Jun 13, 2019 at 4:36 PM Eli Zaretskii <eliz@gnu.org> wrote:
>
> > From: Pip Cet <pipcet@gmail.com>
> > Date: Thu, 13 Jun 2019 13:48:40 +0000
> >
> > I've looked at the code, and it appears no particular provisions are
> > being made to make sure we switch to the modified buffer before
> > calling signal_after_change().
> >
> > As far as I can tell, this makes `put-text-property' with a buffer
> > argument pretty useless.
>
> Only if you have a buffer-local value of after-change-functions.

I'm not sure what you're saying. I'm seeing weird behavior in these cases:
 - buffer-local value of after-change-functions
 - global value of after-change functions (current-buffer is wrong!)
 - overlay property modification-hooks

(let ((buffer1 (generate-new-buffer "A"))
      (buffer2 (generate-new-buffer "B")))
  (with-current-buffer buffer2
    (insert "BBB"))
  (with-current-buffer buffer1
    (insert "AAA")
    (overlay-put (make-overlay 1 4 buffer1) 'modification-hooks
         (list (lambda (&rest args)
           (message "%S"
                args))))
    (goto-char 2)
    (insert "AAA")
    (put-text-property 1 4 'read-only t buffer2)))

produces three calls to the lambda, but should produce only two. (The
last_overlay_modification_hooks logic is a bit weird, thus the second
insertion).

That seems pretty wrong to me. In which cases do you think we're
seeing the right behavior?

Here's a first patch, which adds a "buffer" argument to
signal_after_change, to be explicit about where the change happens. It
should be pretty cheap in the case where we don't switch buffers.

0001-Switch-to-modified-buffer-in-signal_after_change.patch
Description: Text Data