[darcs-users] darcs host binary (x86/Solaris)

David Roundy droundy at abridgegame.org
Wed Feb 2 11:49:54 UTC 2005


On Wed, Feb 02, 2005 at 12:34:52PM +0100, Juliusz Chroboczek wrote:
> > I don't know, I really like the idea of pushing via sftp.  It is inherently
> > less efficient in terms of bandwidth usage, since you have to write
> > complete files.  Certainly allowing pushes to a repository that has both
> > pristine cache and working directory sounds like a very bad idea -- 
> 
> (Speaking about that: what about an option to have repositories
> without a working directory?)

Yeah, this has been suggested before, and is a reasonable idea.

> > you'd have to send over the network the entire contents of your code
> > tree two times at the minimum,
> 
> I'm not following here.  Why do you need to copy files that you're not
> touching?  I don't see why going over FTP or sftp[1] should be worse than
> over NFS.

It's because you can't check the timestamps to see which files have
changed.  You could indeed make it not so expensive if you changed the
algorithms around so that we didn't need to "smart_diff" to see what's
changed in the working directory.  Currently we diff everything that's
changed, and then merge this with the new patch.  If we had something smart
like the patch domains idea, we could perhaps diff only the files that will
affect the merge.

But of course, the fundamental issue is that it's stupid having a working
directory on a computer that doesn't even have darcs installed.
-- 
David Roundy
http://www.darcs.net




More information about the darcs-users mailing list