[darcs-users] darcs for system administration?

[Fredrik Wikefeldt]

Hello,

I started to write a tool for system administration half a year ago
with the goals:

1. Forcing sysadmins to revision control all changes to config files
(like those in /etc)

2. Ability to have different "change" permissions for different files
in the same directory. For example, I don't want all sysadmins to be
able to change the /etc/sudoers because then they are practically
root.

3. Ability to leave files in a directory like /etc out of revision
control if I think those files will never be changed.  This in fact
should be the standard. Thus, you start with the /etc tree you have
after OS install (in my case the minimum Debian). After that, I can
let apt-get add files and delete or change them too but only if they
are not revision controled.  This might require modifying apt-get.

4. When changing the actual /etc directory (by getting a version the
revision control system), it should preferably be done no more
disruptively than when you normally change configuration files.
Whiping out the whole /etc and replacing it with another one is thus
not desirable.

5. Ability to block files from being revision controled.  Could be
useful for files like /etc/passwd and /etc/group because you never
want to revert to a previous  version where maybe a dismissed employee
still had his account.

6. Ability to cherry-pick changes. For example, if a mail server
setting is changed on one server and I discover later that I want to
do exactly that on another mail server, I can easily transfer the
patch/change between the servers.


I have written a Haskell program that does this except from the
apt-get part and 4).
But I used arch instead of darcs because I didn't know about darcs
when I started. Sounds strange given that I did it in Haskell but I
actually started writing it in bash, then went to python and then
wrote it in Haskell after being convinced about type checking.

Now the question is, could I benefit from using darcs instead of arch
apart from integration benefits stemming from both parts being
writting in Haskell?
But more importantly,  I'd like to know if I'm missing something
important here. Is the program I've written interesting? Am I solving
the right problem? Or maybe darcs can handle this right out of the
box.

Best regards,
Fredrik Wikefeldt
Santiago de Chile