Repository collection practices

[Mark White]

Bryan O'Sullivan writes:
> On Mon, 2006-06-26 at 23:58 +0100, Mark White wrote:
> > I guess for when you want to track (atomic) versions across
> > a wider tree than is convenient for branching.  The obvious
> > reason is size: for something like pkgsrc or FreeBSD ports
> > (as Sanjoy mentioned), it's important to be able to get a
> > coherent version of everything at once, but you probably
> > don't want to branch the whole tree just to experiment with
> > one package.
> 
> Benoit has a patch that lets you associate multiple working directories
> with a single repository, which partly addresses this problem.  Couple
> that with the much-desired support for partial checkouts ("I only want
> to check out *this* hunk of the tree"), and you've got something that
> will, I think, work quite well for small experiments on large trees.

Sure: sounds like a useful approach.  Though looking at the
patch I guess this would mean several [maybe partial]
working trees committing to the same repository, rather than
a Mercurial branch in the normal sense?

I suppose it'd be fairly easy to share common revlog files
directly (rather than with hardlinks), then make real copies
for local commits as needed, if the speed of making the
linkfarm turned out to be limiting with a large tree.

Mark <><