[plug] Source Code 'Repository'

[Craig Ringer]

On Thu, 2005-08-18 at 17:12 +0800, Cameron Patrick wrote:
> Craig Ringer wrote:
> 
> > CVS doesn't as far as I know offer anything like that, and I have no
> > experience with the other alternatives.
> 
> It does.  You can do e.g.
> 
>     export CVSROOT=somewhere
>     cvs co myproject/some/subdir blah
>     cvs co myproject blah2
> 
> CVS allows a lot of flexibility in doing this, more than SVN or any of
> the distributive systems, but at the cost of utterly unintuitive
> semantics for commands like 'update'.

News to me, but good to know. I'm not too sure I can take any more
unintiutive behaviour from `cvs up' than the defaults though (all I need
to say is "revert in a branch").

> > It'd probably what I'd pick. The only issue is that it's a pain if
> > you're disconnected for long periods, since you can't `svn diff', `svn
> > ci', revert a file, etc.
> 
> You can 'svn diff' when offline (unlike CVS) since every check-out
> contains a pristine copy of the tree as well your working copy.

Again, that's news to me. Very good news. I'm just getting the hang of
svn now, and that'll make life considerably easier.

> That said, I also find it hard to recommend CVS (because although it's
> so widespread¸ pretty much all of the alternatives are better), or
> Subversion (ditto, except for the widespread bit, and it _is_ better
> than CVS).  If you do use Subversion, I'd recommend avoiding the
> berkely DB backend at all costs and using fsfs instead.

Seconded with considerable enthusiasm. I steered clear of it until I
could use fsfs, becuase my experience with Berkley DB with NetATalk and
other projects has shown that it's either (a) just not robust enough or
(b) way too hard to tune for reliable operation.

I managed to get a corrupt bdb in my first test of svn some time ago. It
was a large project, but on a stable machine in a building with
excellent power (ISP colo).

--
Craig Ringer