[darcs-users] keeping unusable stuff around

[Ganesh Sittampalam]

On Wed, 26 Oct 2005, Albert Reiner wrote:

> Now I have spent (far too much!) time modifying C1 so as to
> investigate the effects of some change to the underlying theory, and
> this has turned out not to be usable.  It has resulted in numerous
> changes all over C1, and now I want to go back to the old code without
> permanently loosing all the modifications I made.

I just keep around the old repository when this happens. Is this a problem 
for you? There are various measures you can take to reduce the disk space 
it'll use:

  - 'darcs optimize --relink --sibling=foo' will hard link the patch files 
representing common history with foo to those in foo

  - deleting the entire working directory except _darcs, and the contents 
of _darcs/current. Make sure you have no local changes first. You can 
restore them using 'darcs repair' followed by 'darcs revert -a' (Someone 
correct me if this is "unsupported" and likely to lead to tears later..)

> - Record and rollback.  This way the useless code is in the repo and I
>  can go back to it later.  But will this negatively affect patch
>  commutation?  The changes are all over the place, so there might
>  easily arise additional conflicts when patches are commuted past the
>  patch and its inverse.  Or is darcs smart enough to know that a
>  sequence of patch and its inverse are just an identity operation?

darcs isn't yet smart enough for this, although there are some hopes of 
making it so in future. I think even if it was, having these dead changes 
cluttering up history isn't such a good idea, though that's mainly a 
matter of aesthetics.

> - Tag the old version of C1, record the changes, use |send| to make a
>  patchbundle that is then simply stored in the D branch, and unrecord
>  the changes in C1.  This seems nicer to me, but I am not sure it is
>  a good idea.  In particular, is the patchbundle produced by `darcs
>  send` suitable for storing random patches in the long run, or is it
>  liable to change?

Recording a patch bundle in version control itself seems a little ugly to 
me when there are alternatives, but I can't see any fundamental problem 
with this.

Cheers,

Ganesh