[darcs-users] Issues with darcs
baruch at ev-en.org
Fri Mar 11 12:41:57 UTC 2005
I'm trying to use Darcs for my Linux kernel development efforts, and
it's not all smooth. Let me first describe what I need.
I'm working on multiple patches in succession to be submitted. So I have
a base htcp patch, and on top of it several patches, one by one,
though they are parallel to each other, I just set some ordering for
On top of that I needed some performance improvements so I have another
set of patches as well, all in all about 10 patches to the Kernel.
The current way of work is with quilt and having the quilt patches in a
source control repository.
I thought of working with darcs in the way of a waterfall of patches so
I can have the history of each patch as each patch will be maintained in
its own tree. Merges and stuff will be done aided by darcs and I won't
need to read patches of patches.
The problems I hit:
Importing the Linux Kernel 2.6.11 took 24 hours(!), that's just for
darcs add of the whole kernel source, it was actually done in 10 files
sets using find and xargs at the suggestion of an irc'er. Most of the
time was without any memory pressure, or IO pressure, just pure CPU
time. The last two hours were with a large memory push and thus a large
I then started filling the cascade of branches, darcs get takes about 7
minutes most of which is IO.
The problem I now have is disk space usage, each branch takes 500MB of
disk space. I've seen that there is an option to work without the recent
copy, but it can't be changed after the initial checkout, I'd like to be
able to change this in the middle because I am working now on the
performance patches and don't need the disk space waste on the htcp
patches, but will need to get back to the htcp stuff later on.
I also need an option to hide/unhide the working copy source, so that
there will be no sources used for a specific branch unless I unhide it.
This way I can unhide all the other branches and just work on a specific
branch (or a few branches, as needed).
A tree without current copy and hidden would take 44MB as opposed to
450MB. That would be a saving of 4GB on my machine which will be useful.
p.s. I'm not subscribed a cc would be most welcome.
More information about the darcs-users