question about markers, replace-match, and undo

question about markers, replace-match, and undo

[Drew Adams]

I have some text, with a marker before it and a marker after it. For
instance, the first marker is at position 41, just before `m', and the
second is at position 48, just after `s':

  41     48
    mnopqrs

I match this text, and then call (replace-match "ABC"), with the result that
the first marker is still at position 41 and the second marker is now at
position 44, just after `C', which is what I would expect and what I want:

  41 44
    ABC

I use `undo', and the result is that both markers are now at position 41:

  41
  41
    mnopqrs

I get the same behavior if I use various additional args to `replace-match'.
I've also tried (progn (undo-boundary) (replace-match "ABC")), with no
change.

I'm not saying there is necessarily a bug here; I don't know. I'm asking how
I can get `undo' to restore not only the text but the marker positions. If I
use, for instance, `delete-region' and `insert', there is no problem, but I
need to be able to use `replace-match' here.

What also seems odd is that (buffer-has-markers-at 48) returns t, but
`mark-ring' shows only two markers, both at position 41, and the mark itself
is somewhere else altogether (e.g. 18). I also used a library `mark.el' from
Emacs Wiki that lets you access the markers in a buffer, and it does not
show any marker at 48. So perhaps `buffer-has-markers-at' is reporting
erroneously here?

[BTW, I also wonder why we have `buffer-has-markers-at', but we apparently
have no function that returns the list of markers at a given position.]

Can anyone help with my question of how to get `undo' to act as an inverse
of `replace-match' wrt the markers? Thx.

Re: question about markers, replace-match, and undo

[Johan Bockgård]

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

> I'm not saying there is necessarily a bug here; I don't know. I'm asking how
> I can get `undo' to restore not only the text but the marker positions.

FWIW, the markers are correctly restored in Emacs 21.

-- 
Johan Bockgård

RE: question about markers, replace-match, and undo

[Drew Adams]

> > I'm not saying there is necessarily a bug here; I don't know. 
> > I'm asking how I can get `undo' to restore not only the text
> > but the marker positions.
> 
> FWIW, the markers are correctly restored in Emacs 21.

Ah, thanks Johan. 

So is this an Emacs 22 bug, or is the current behavior intended and a fix for an Emacs 21 bug? 

If this is the intended behavior, can anyone help with how to force the previous behavior (i.e. restore the markers to their original positions)?

Re: question about markers, replace-match, and undo

[Richard Stallman]

    I have some text, with a marker before it and a marker after it. For
    instance, the first marker is at position 41, just before `m', and the
    second is at position 48, just after `s':

      41     48
	mnopqrs

    I match this text, and then call (replace-match "ABC"), with the result that
    the first marker is still at position 41 and the second marker is now at
    position 44, just after `C', which is what I would expect and what I want:

      41 44
	ABC

    I use `undo', and the result is that both markers are now at position 41:

      41
      41
	mnopqrs

That is definitely a bug.  Does this fix it?

*** insdel.c	02 Mar 2007 03:28:13 -0500	1.190
--- insdel.c	27 Apr 2007 23:55:23 -0400	
***************
*** 1603,1610 ****
  
    if (! EQ (current_buffer->undo_list, Qt))
      {
        record_delete (from, deletion);
-       record_insert (from, inschars);
      }
  
    GAP_SIZE -= outgoing_insbytes;
--- 1603,1614 ----
  
    if (! EQ (current_buffer->undo_list, Qt))
      {
+       /* Record the insertion first, so that when we undo,
+ 	 the deletion will be undone first.  Thus, undo
+ 	 will insert before deleting, and thus will keep
+ 	 the markers before and after this text separate.  */
+       record_insert (from + SCHARS (deletion), inschars);
        record_delete (from, deletion);
      }
  
    GAP_SIZE -= outgoing_insbytes;

RE: question about markers, replace-match, and undo

[Drew Adams]

> That is definitely a bug.  Does this fix it?

I cannot tell; I don't build from C sources. If someone else could have a
look, that would be great. Thanks.

> *** insdel.c	02 Mar 2007 03:28:13 -0500	1.190
> --- insdel.c	27 Apr 2007 23:55:23 -0400
> ***************
> *** 1603,1610 ****
>
>     if (! EQ (current_buffer->undo_list, Qt))
>       {
>         record_delete (from, deletion);
> -       record_insert (from, inschars);
>       }
>
>     GAP_SIZE -= outgoing_insbytes;
> --- 1603,1614 ----
>
>     if (! EQ (current_buffer->undo_list, Qt))
>       {
> +       /* Record the insertion first, so that when we undo,
> + 	 the deletion will be undone first.  Thus, undo
> + 	 will insert before deleting, and thus will keep
> + 	 the markers before and after this text separate.  */
> +       record_insert (from + SCHARS (deletion), inschars);
>         record_delete (from, deletion);
>       }
>
>     GAP_SIZE -= outgoing_insbytes;

Re: question about markers, replace-match, and undo

[Richard Stallman]

Can you send a program to construct the test case that fails for you?
Then I will test my patch and make sure it fixes your problem.

Re: question about markers, replace-match, and undo

[Johan Bockgård]

Richard Stallman <rms@gnu.org> writes:

> Can you send a program to construct the test case that fails for you?
> Then I will test my patch and make sure it fixes your problem.

I tested it like this:

$ emacs -Q

---------- Buffer: *scratch* ----------
123456

(defun a ()
  (interactive)
  (setq a (copy-marker 1))
  (setq b (copy-marker 7))
  (goto-char (point-min))
  (re-search-forward "123456")
  (replace-match "000"))

(defun b ()
  (interactive)
  (message "%s, %s" a b))
---------- Buffer: *scratch* ----------

Eval the buffer.

M-x a RET
Undo (C-/)
M-x b RET


Emacs 21
  => #<marker at 1 in *scratch*>, #<marker at 7 in *scratch*>

Emacs 22
  => #<marker at 1 in *scratch*>, #<marker at 1 in *scratch*>

Emacs 22 after your patch
  => #<marker at 1 in *scratch*>, #<marker at 7 in *scratch*>


-- 
Johan Bockgård

RE: question about markers, replace-match, and undo

[Drew Adams]

> > Can you send a program to construct the test case that fails for you?
> > Then I will test my patch and make sure it fixes your problem.
> 
Johan> I tested it like this:
...

What Johan did is essentially what I did, so I think this is fixed. Thanks.

What I did (see my initial report):

emacs -Q
type mnopqrs in a buffer
C-SPC at (before) m
C-SPC after s
C-s mnopqrs RET
(replace-match "ABC")
C-_
Look at mark-ring and (mark).

I also reported a bug (?) about `buffer-has-markers-at'. In this case, even though the bugged result above is two markers before m and no marker after s, `buffer-has-markers-at' returns t for the position after s.

Re: question about markers, replace-match, and undo

[Richard Stallman]

Thanks.  Now I am confident this is fixed.