[darcs-users] Rollback on a lazy repo

[Owen Stephens]

Hi Gian,

On 13 December 2012 22:13, Gian Piero Carrubba <gpiero at rm-rf.it> wrote:
> I have a --lazy repo in which I'd like to experiment with two alternative
> patches.  The workflow I had in mind was:
> - record the first patch
> - rebuild
> - test
> - record the second patch (they don't conflict)
> - rollback the first patch
> - rebuild
> - test
>
> The problem is that now I typed `darcs rollback -p issue2238` and darcs
> seems being downloading all the patches missing in the local repo before
> even asking me to select the patches to rollback.

This is strange (I can confrim I see the same behaviour here) - I'd expect the
rollback to be lazy in the patches it inspects. I don't know much about the
internals of the patch selection code, so can't be sure - Florent or Ganesh
would probably know more.

Interestingly:

    darcs cha -p issue2238

only inspects one patch file (according to --debug-verbose), but still inspects
all the inventory files. I would guess that we could/should inspect
patches/inventories on a per-inventory and then per-patch-in-inventory basis.

> BTW, the issue is issue2238 ( darcs wh -ll suggests unadded files are added
> )[0]. Removing duplicated --look-for-adds passed to whatsnew seems to work,
> anyway I thought a more general solution could be applied, so removing every
> duplicated option passed to darcs. I think there's no option that can be
> legitimately passed multiple times but surely I can be wrong. Can someone
> confirm there are no such cases or inversely confirm they exist ?

I can't think of anything, but that's not a confirmation :-) you'd have to be
careful of detecting short/long options i.e. not just equal strings - for
instance darcs wh -l --look-for-adds.

Cheers,
Owen.