Re: [Gnu-arch-users] undo for single files?

[Graham Hughes]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Dec 27, 2003, at 17:26, Miles Bader wrote:

> On Sat, Dec 27, 2003 at 08:10:07PM -0500, James Blackwell wrote:
> > If we added -f <filename> to "tla undo", then we'd end up making 
> > little
> > one-file changesets that the user would have to rm if he didn't want 
> > them.
>
> None-the-less, I think the proper solution is still to make `tla undo' 
> accept
> limits (just like commit, e.g, `tla undo -- file1.c file2.c').  Not 
> just
> because it's easier for users to remember/understand (though it is 
> that too),
> but because it's going to have to be done anyway: there are many cases 
> where
> I _want_ to undo the changes to a single file, do some tla commands, 
> and then
> redo the changes, and a limited undo + redo is the natural way to 
> provid this
> functionality.

I'd like to second this; the bad habit I have that shows up all the 
revision control systems I've ever used is that my patchsets tend to do 
way too much, and I don't realize until after I'm halfway through them 
that `oh, I could factor this bit out into a changeset of its own' etc.

Aegis had the least bad way of dealing with this that I've run across 
(you'd clone the changeset and then revert anything you really weren't 
interested in), but undo with limits would be sufficient.

I recall some discussion about the difficulty of doing this precisely 
(with renames etc.) but I personally would settle at the moment for 
something along commit's line that simply refuses to do anything not 
obviously correct.

Graham
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (Darwin)

iD8DBQE/7jWf2kJk3FmxWkQRAmUlAJ9H+h/rtgYNI1ddBOTIgdZiwf4EPwCghMOq
e8P3/SY2YJ13ZQ99KzevT6U=
=b8bv
-----END PGP SIGNATURE-----