Re: Support for undo-amalgamate in a version of the atomic-change-group macro (with patch)

[Campbell Barton <ideasman42@gmail.com>]

On 11/2/21 04:43, Stefan Monnier wrote:
>> Hi, there have been a handful of times I've needed to treat multiple
>> actions as a single action,
>> others too (see stackoverflow question [0]).
>>
>> While this is supported using change groups, the resulting macro ends
>> up being nearly identical to the existing `atomic-change-group' macro.
>> The only difference is a call to undo-amalgamate-change-group.
> 
> The other difference would be the call to `cancel-change-group` (you
> may want to not undo the changes in case of errors).

Good point (updated my answer)

>> Attached a patch to add `atomic-change-group-undo-amalgamate' so
>> packages don't need to use a modified copy of this macro.
> 
> I suspect that we'd be better off with a separate macro which only does
> the amalgamation without the cancel-change-group-on-error.

Agree, if the cancel group is removed it's no longer atomic and can be a 
separate macro.

> But I wonder how often such a macro would be useful: undo boundaries are
> normally added only by the command-loop (i.e. the loop which repeatedly
> reads a key and calls the corresponding command), so unless your macro's
> body calls `recursive-edit` it's very unlikely that you'll get any undo
> entries that you would then want to remove.

I'm not sure why most other people are running into this (the original 
question relates specifically voice commands), nevertheless the stats on 
the stackexchange question lead me to think it's not so unusual to spawn 
multiple commands that happen to be treated as separate undo steps.

> (Re)reading your stackexchange question, it seems you want to amalgamate
> undo entries from different commands.  So do you run those commands
> within a `recursive-edit`?  If not, how do you manage to run those
> commands from within your (new) macro?

As far as I know I'm not using recursive edit (unless indirectly).

In these situations collapsing undo data was necessary.

- Calling kmacro-exec-ring-item (based on this answer [0])
- Calling some evil functions for e.g.

   (defun evil-paste-and-indent-after ()
     (interactive)
     (with-undo-collapse
       (evil-paste-after 1)
       (evil-indent (evil-get-marker ?\[)
                    (evil-get-marker ?\]))))

[0]: https://emacs.stackexchange.com/questions/70

> 
> 
>          Stefan
>