[darcs-users] darcs patch: Hide the inefficient "darcs put" command.

[Dan Pascu]

On Saturday 06 December 2008, Trent W. Buck wrote:
> Florent Becker <florent.becker at ens-lyon.org> writes:
> > Either we drop put now and forever, or we keep it and a/ try and make
> > it efficient b/ warn our users.
>
> Currently we do (b); I'd like to do (a) but until (a) happens I was
> proposing to hide or remove the current implementation.  (This has been
> rejected by Eric elsewhere in the thread, because there's a lot of
> disagreement.)
>
> > By the way, UI leanness is not measured in terms of number of
> > commands/flags, but in terms of coherence. For this, having symmetric
> > pairs is much preferable to removing the least used direction.
>
> Except that for the most common case, a local branch, get is
> bi-directional, and more efficient that the currently implementation of
> put.  Using darcs' repository as an example:
>
>     $ time darcs get --lazy --repodir $PWD.branch1 $PWD
>     Finished getting.
>
>     real    0m2.206s
>     user    0m1.012s
>     sys     0m0.464s
>
>     $ time darcs put $PWD.branch2
>     Merging them 3414/7011
>     Writing pristine 205/236
>     Optimizing inventory 3575/7011
>     Writing inventory 1/1 :
>     Synchronizing pristine 203/236
>     Finished applying...
>     Put successful.
>
>     real    4m49.517s
>     user    3m23.681s
>     sys     0m35.830s

I think you are comparing apples with oranges here. You issues a lazy get 
which only took the pristine tree, while put took and applied all patches

IMO, it's less useful to have a lazy put (maybe it should be there for 
symmetry), because the use target for put is a bit different from get. 
With get you can afford to retrieve a repository in lazy mode because it 
is a local copy. Put OTOH, is mostly used to setup a central repository 
for syncing between multiple disconnected developers. In this case I do 
not want a central repository that has none or only has a few patches.
I want a complete repository so that people can synchronize their work. 
Otherwise it will try to connect to the original publisher repository to 
fetch patches from, which can fail for many reasons (that computer is not 
online, not authorized to connect to it, ...). With get is more likely 
that you can connect again later, since you already connected once (at 
least if ssh auth doesn't change). But with put you do not know that at 
all since it would require a reverse connection which was never 
established and may never be possible to establish it anyway.

In fact even with get it can be tricky. If a take a repo in lazy mode over 
ssh, and then someone tries to pull a patch from this repo it will try to 
connect to the original repo over ssh, but the getter most likely will 
not have the credentials to do that.

IMO lazy mode is only useful for a personal copy that one will ever 
push/pull from. The only exception is if this lazy copy is just a local 
copy of a repo, so it can always take patches from the original without 
any ssh involvement. But if a lazy copy is taken over ssh then it's 
unusable for others to pull from it.

>
> FWIW the result also includes more files, or does less hard-linking:
>
>     $ cd ..
>     $ du -sch darcs darcs.branch1 darcs.branch2
>     51M         darcs
>     4.5M        darcs.branch1
>     5.9M        darcs.branch2
>     61M         total
>
> _______________________________________________
> darcs-users mailing list
> darcs-users at darcs.net
> http://lists.osuosl.org/mailman/listinfo/darcs-users



-- 
Dan