Re: [Gnu-arch-users] [PATCH] tla revert

[Robert Anderson]

On Sun, 2003-09-07 at 06:35, Zack Brown wrote:
> On Sat, Sep 06, 2003 at 09:30:26PM -0400, Miles Bader wrote:
> > On Sun, Sep 07, 2003 at 01:54:29AM +0100, Bruce Stephens wrote:
> > > > Regarding this particular change: I wonder if it wouldn't be better
> > > > to add support for limited-scope undo to the `undo' command
> > > > (analogous to the '[-- file ...]' optional arguments to `commit')?
> > > 
> > > Maybe, but that seems overkill for many situations.  I imagine revert
> > > is intended for when you decide that what you've done to the file just
> > > isn't worth saving, and you want to forget all the changes.  undo
> > > seems to have a broader role.
> > 
> > I disagree, undo seems almost perfect for this.  If it could be restricted
> > to a certain file(s), it would have exactly the desired effect -- and has
> > the further advantage of later being redoable if you realize those changes
> > were desirable after all (some people might complain about the ,,undo-N
> > droppings, but they're easy enough to delete; I tend to leave them sitting
> > around until I've finished whatever task I'm working on, Just In Case).
> 
> So how about the case where a user makes changes and realizes they were the
> wrong changes.  They run undo and the changes are gone. Then they make more
> changes that they like, and submit a changeset up to the maintainer. Will
> the maintainer see evidence of the work that was previously undone?

No, if they wanted that they would have to commit those changes. 
Perhaps the original description of what "tla revert" did wasn't very
clear.  Maybe the analogous thing would not be a file-based undo, but a
file-based make-sync-tree?

Bob