bug#8447: Undoing M-x revert-buffer

bug#8447: Undoing M-x revert-buffer

[Hrvoje Nikšić]

It would be nice to be able to undo a revert-buffer operation.

I've just had a situation where I used M-x revert-buffer, and
immediately wanted to see the old version again.  (I wanted to briefly
compare the old and new contents, which only differed in several
characters.)  I routinely pressed C-_, and got a "no undo data" message
in the echo area.  For the shortest instant I was surprised, but I
quickly remembered that M-x revert-buffer clears the undo stack, and
that it has always worked that way.

But I would like to question this design choice.  What I wanted to do
was not really unreasonable.  The buffer held several kilobytes of data,
and my computer has gigabytes of memory at its disposal, enough room to
hold the buffer contents literally a million times over.  If you think
about it, reverting an existing buffer is just another operation on
buffer text, not that different than M-x erase-buffer followed by M-x
insert-file, which would happily remember everything.

If the maintainers agree in principle, I'd like to look into patching
revert-buffer so it records the revert as it would any other change to
the buffer text.

Hrvoje

bug#8447: Undoing M-x revert-buffer

[Stefan Monnier]

> It would be nice to be able to undo a revert-buffer operation.

So do I.  I've been using a local patch to do that, actually.  IIRC back
when I wrote it, RMS opposed the change.  I think my change only
affected the "revert while keeping the same major-mode" such as the
revert performed silently by auto-revert-mode and VC operations.  It has
bit-rotted somewhat, but I remember that the current code works hard at
forgetting the undo data (it does it once at the Lisp level and another
time inside insert-file-contents).


        Stefan

bug#8447: Undoing M-x revert-buffer

[Kevin Rodgers]

On 4/8/11 9:54 AM, Hrvoje Nikšić wrote:
> It would be nice to be able to undo a revert-buffer operation.
>
> I've just had a situation where I used M-x revert-buffer, and
> immediately wanted to see the old version again.  (I wanted to briefly
> compare the old and new contents, which only differed in several
> characters.)  I routinely pressed C-_, and got a "no undo data" message
> in the echo area.  For the shortest instant I was surprised, but I
> quickly remembered that M-x revert-buffer clears the undo stack, and
> that it has always worked that way.
>
> But I would like to question this design choice.  What I wanted to do
> was not really unreasonable.  The buffer held several kilobytes of data,
> and my computer has gigabytes of memory at its disposal, enough room to
> hold the buffer contents literally a million times over.  If you think
> about it, reverting an existing buffer is just another operation on
> buffer text, not that different than M-x erase-buffer followed by M-x
> insert-file, which would happily remember everything.
>
> If the maintainers agree in principle, I'd like to look into patching
> revert-buffer so it records the revert as it would any other change to
> the buffer text.

In the short term, or if the maintainers do not agree to Hrvoje's proposal,
how about adding a warning to the "Revert buffer from file FILE? " yes-or-no-p
prompt?

-- 
Kevin Rodgers
Denver, Colorado, USA

bug#8447: Undoing M-x revert-buffer

[Stefan Monnier]

I just installed into the trunk a patch which should make
revert-buffer undoable.


        Stefan

bug#8447: Undoing M-x revert-buffer

[Drew Adams]

> I just installed into the trunk a patch which should make
> revert-buffer undoable.

FWIW, see my comment on this in the help list:
http://lists.gnu.org/archive/html/help-gnu-emacs/2013-05/msg00671.html

In sum, why?  And why no discussion?  Where was the "bug"?

bug#8447: Undoing M-x revert-buffer

[Andreas Röhler]

Am 29.05.2013 05:15, schrieb Drew Adams:
>> I just installed into the trunk a patch which should make
>> revert-buffer undoable.
>
> FWIW, see my comment on this in the help list:
> http://lists.gnu.org/archive/html/help-gnu-emacs/2013-05/msg00671.html
>
> In sum, why?  And why no discussion?  Where was the "bug"?
>
>
>
>

Hi Drew,

as it happened, run sometimes into this:
reverting the buffer, which turned out not the right thing - but undo-history was lost.
Consider it a useful change, thanks.

Andreas

bug#8447: Undoing M-x revert-buffer

[Drew Adams]

> >> I just installed into the trunk a patch which should make
> >> revert-buffer undoable.
> >
> > FWIW, see my comment on this in the help list:
> > http://lists.gnu.org/archive/html/help-gnu-emacs/2013-05/msg00671.html
> >
> > In sum, why?  And why no discussion?  Where was the "bug"?

> as it happened, run sometimes into this:
> reverting the buffer, which turned out not the right thing - but undo-
> history was lost. Consider it a useful change, thanks.

By "turned out not the right thing" I guess you mean that someone used `revert-buffer' interactively and by mistake.  And by mistake also _confirmed_ reverting.  Yes, it can happen (to anyone).

So define a command `revert-buffer-keeping-undo', and use that interactively instead of `revert-buffer'.  Or define a variable (option or internal) `revert-buffer-keeps-undo'.

What you cite is a user-interface issue - i.e., only for interactive use of the command.  Such a change, to safeguard against mistaken reverting, is akin to removing files to a trash bin instead of deleting them.  But we did not just willy nilly change the behavior of the basic function (command) `delete-file' when we added support for a recycle/trash bin.  That's not the right approach.

With extra interactive protection as the only reason, this is also akin to some users (me, for instance) wanting `C-w' to prompt for confirmation if the region is over a certain size (as in wimpy-del.el).  Certainly such safeguard features can be useful.

But this change goes way beyond offering users optional extra protection.  `revert-buffer' is also used in basic code.

The right way to add such protection against mistaken reversion is to create a separate command or option, letting users choose to use it or not.  And leave the basic `revert-buffer' alone.

FWIW, I agree that such a feature can be useful for reverting interactively.

And why no discussion before making such a change?

bug#8447: Undoing M-x revert-buffer

[Andreas Röhler]

Am 29.05.2013 15:27, schrieb Drew Adams:
>>>> I just installed into the trunk a patch which should make
>>>> revert-buffer undoable.
>>>
>>> FWIW, see my comment on this in the help list:
>>> http://lists.gnu.org/archive/html/help-gnu-emacs/2013-05/msg00671.html
>>>
>>> In sum, why?  And why no discussion?  Where was the "bug"?
>
>> as it happened, run sometimes into this:
>> reverting the buffer, which turned out not the right thing - but undo-
>> history was lost. Consider it a useful change, thanks.
>
> By "turned out not the right thing" I guess you mean that someone used `revert-buffer' interactively and by mistake.  And by mistake also _confirmed_ reverting.  Yes, it can happen (to anyone).
>
> So define a command `revert-buffer-keeping-undo', and use that interactively instead of `revert-buffer'.  Or define a variable (option or internal) `revert-buffer-keeps-undo'.
>
> What you cite is a user-interface issue - i.e., only for interactive use of the command.  Such a change, to safeguard against mistaken reverting, is akin to removing files to a trash bin instead of deleting them.  But we did not just willy nilly change the behavior of the basic function (command) `delete-file' when we added support for a recycle/trash bin.  That's not the right approach.
>
> With extra interactive protection as the only reason, this is also akin to some users (me, for instance) wanting `C-w' to prompt for confirmation if the region is over a certain size (as in wimpy-del.el).  Certainly such safeguard features can be useful.
>
> But this change goes way beyond offering users optional extra protection.  `revert-buffer' is also used in basic code.
>
> The right way to add such protection against mistaken reversion is to create a separate command or option, letting users choose to use it or not.  And leave the basic `revert-buffer' alone.
>
> FWIW, I agree that such a feature can be useful for reverting interactively.
>
> And why no discussion before making such a change?
>

Hi Drew,

thanks, okay, you are right opposing it then. Mistook the matter so far.

Cheers,

Andreas