Brief article on benchmarks of Python repository with leading DVCSen

Stephen J. Turnbull turnbull at
Wed Feb 11 16:44:42 GMT 2009

Nicholas Allen writes:

 > I find it hard to believe that programmers finds one simple command too 
 > complex but I'll take your word for it. I thought they originally used 
 > subversion at some point which requires significantly more setup - 
 > including a server for the repository.

Sure.  It's not the worker bees we're talking about, though.  They've
already moved once from CVS to Subversion fairly recently, and they're
more than willing to do the work to move to a DVCS.  *If* they're
satisfied that that is what the community wants.

 > I mean all they have to do is *one* extra command (not even for each 
 > branch but just once per project):
 > bzr init-repo my-branches
 > Then they just have to branch all there branches below that
 > directory. 

In our context, telling people where to branch is not acceptable.  The
people who want a DVCS badly will do anything you tell them to.  The
people who are pretty much satisfied with Subversion refuse to
countenance any unnecessary change, and will quite likely oppose the
move if they have to reorganize their workspaces.

 > It sounds to me like people are being completely unreasonable here.

No.  We're trying to protect the developers from *any* change in
current habits.  That is the sine qua non: a seamless change.  That's
one great advantage of DVCS: once you get that widget installed, the
people who want to take advantage of it can completely ignore the
stick-in-the-muds, and vice versa.  But you've got to get over that
hump, and a plausible claim that "it will be exactly like before
except better" is the best way to accomplish that.

 > A fair benchmark would have been much more interesting IMHO. I
 > expect Bazaar would have still be the slowest but it would at least
 > be a reasonable and fair comparison.

The Python developers have *no interest whatsoever* in a "fair"
comparison.  They want a comparison in workflows that have the
absolute minimum differences from their current workflows.  (Note the
plural here; that's important.  Workflows differ substantially between
committers and non-committers, and even within each of those classes.)

The Emacs developers ... well, for better or worse Bazaar performance
vis-a-vis other DVCSes had nothing to do with the choice.  For that
very reason I expect that when the switch over to Bazaar actually
happens quite a few people will be displeased.  An equal number will
probably be pleasantly surprised, but that's the stuff that "silent
majorities" are made of.

As for Mr. Fredt, I don't know what his purpose was.  He might very
well be interested in methodological critiques.

More information about the bazaar mailing list