Re: Undo framework

[Johan Dahlin <johan gnome org>]

Jody Goldberg wrote:
> On Fri, Sep 21, 2007 at 05:51:26PM +0100, Iain * wrote:
> > Hi,
> >
> > I've had an undo framework in Marlin for years now, but recently
> > people have been using it in other things (notably Ross in Tasks - ok,
> > actually, he's the only one) and we discussed suggesting this for
> > inclusion in GTK at some point in the future. QT4[1] and Cocoa[2] both
> > have very similar frameworks. Attached are the header files.
>
> The concept sounds good, and while your implementation looks clean,
> I'd rather not see it go into gtk in it's current form.
>
> 1) As you point out we've all had undo/redo implementations for
>    several years.  If we're going to do this, let's skip a
>    generation and support transaction logging rather than just
>    picking a best of breed version of the existing code.
>    It's something I've been meaning to add to gnumeric [1] for
>    several years.  The goal is to support a transaction file akin to
>    VIM's .swp.  Doing this properly with permissioning for the
>    transact file isn't code that should get replicated in each app.

I tend to agree, with the limited experience I had dealing with undo/redo in 
Gazpacho I found out that I strongly disliked the way of using commands.

However I cannot heartily recommend a transaction based scheme since I only 
did a very simple prototype for applying GObject properties using 
transactions. It might end up way to complicated, perhaps someone with 
experience of implementing a transactional database could add something to 
this discussion.

> 3) Gtk seems like the right place for widgets to display the
>    undo/redo stack.  However, the core of the functionality seems
>    non-gui.  Something that would belong wherever we chose to put
>    GDocument.  To date that type of code has been migrating out of
>    gnumeric and into goffice.  If there is interest, we could
>    collaborate on the next generation of this in there and rename to
>    'libgdocument' or somesuch.

While I agree that an undo stack should be in Gtk+ or lower I'm not sure a 
transaction based should go there, but it should be possible to implement 
one top of the interfaces/manager objects there.

I'm not sure we should talk about a 'document' there, there are many 
operations outside of documents which are undoable.

Johan