BDFL decision of Python's DVCS
andrew.bennetts at canonical.com
Wed Apr 1 23:57:53 BST 2009
Matthieu Moy wrote:
> Andrew Bennetts <andrew.bennetts at canonical.com> writes:
> > I'm not sure we ever claimed that?
> I quoted the wiki above in the thread:
> Revision 120 as of 2007-01-25 16:00:21
> "We expect 1.0 to release by March 2007 with performance equivalent to
> the best in the field."
Ah, I see. FWIW, I seem to recall that in many respects, we were pretty close.
Between dirstate and packs we did have a backend that was pretty similar in
performance characteristics to the “best in the field”, so the expectation
In hindsight that claim was unfortunate, though.
> You can add this one:
> "Comparing the average of the ratios above, Bazaar is 22% more
> efficient than Mercurial and 15% more efficient than Git."
> This one is still there, has been so for quite some time. Now, look at
> how the benchmark was done (project without history to benchmark a
> tool whose _point_ is to manipulate history). Look at another
> independant benchmark like
> then take your conclusions.
I guess a contributing factor here was that in the effort to produce simple,
easily describable and reproducible benchmarks that they got over-simplified.
Certainly at the time of March 2007 it was pretty hard to use the same
real-world history in different tools for a fair comparison, so it was easier to
leave that part out. Now that the fast-import format is somewhat ubiquitous
(and the tools to read it are increasingly reliable) it's easier to do better
here. (Not that we weren't testing on real-world histories when developing, but
not in a way that was amenable to publishing as an apples-to-apples benchmark.)
That said, the numbers on that post do seem out of whack with what I'd expect...
but I'm fairly sure there was a thread about that months ago, and I'd rather not
have that discussion all over again!
More information about the bazaar