Subject: RE: Perens on arch
From: "Jonathan S. Shapiro" <>
Date: Wed, 6 Feb 2002 10:57:55 -0500

As is so often the case, good ideas come in pairs (or threes, or...).
We've been delaying a release of CPCMS to work out some last issues in
really large scaling (>100,000 projects). Sigh.

While we're not looking for business help, the Hopkins SRL lab also has
a new CM system working, currently called CPCMS but that is soon to
change. We have been debugging some last bits. It shares some things in
common with arch, and provides integrity and provenance tracking as
well. There will be a paper on it in the next FREENIX conference. I've
got to get an NSF proposal out today, but I'll try to get more info on
our system up on the web site and supply a URL later this week. We're
about to cut the EROS project over to it.

The relevance to FSB, I suppose, is as follows:

	1. There are about to be some new and more flexible options in
	   These might make decentralized working styles much easier to
	   support from a technical point of view. Watch this space.
	2. If you live with enough pain long enough, somebody will write
	   a free solution.
	3. Code talks, bullshit...

Hardly profound, but there it is. On point (2), I'll add that the word
from behind the scenes from the FREENIX committee was that people were
really excited about our stuff. Partly, I think we did a good job, but
I'm clearly biased. More importantly, I think that's a strong indicator
of how much pain people feel in the CM area. To a lesser degree it's an
indicator that BitKeeper didn't hit the sweet spot in the eyes of the
user base and people are nervous about the size and complexity of
subversion. The subversion guys have put in a lot of work, but I
definitely share the sense of too many pieces trying to converge in a
mission critical place. I also think they have failed to acknowledge the
importance of replication and disconnection in spite of being told

Arch looks like a really good piece of work. I'm glad Tom has done it,
and I am delighted that the concept sets of arch and CPCMS are close
enough for these systems to learn usefully from each other.

For those who want more info on CPCMS, have a look at the FREENIX draft
paper at