Bazaar performance

[Charles Duffy]

Stephen J. Turnbull wrote:
> Charles Duffy writes:
> 
>  > I suspect that you could get significantly different numbers with some 
>  > tuning.
> 
> Yah, but *nobody* in recent memory has managed it.  It would be a very
> great service to the community if you could give some pointers as how
> you did it, since the obvious ways to do things lead to distressingly
> unfavorable comparisons with both git and Mercurial on performance,
> whereas the advantages of bzr are relatively subtle.
> 

Hmm -- I know some people with said former employer, and could try 
getting access to my mailspool.

That said, it may well be that the circumstances associated with my 
results might relate to the ways that my test cases (and use case) 
differed from more typical Free Software workload -- very large trees, 
but with a very small number of historical revisions (each of which 
touched a very large number of files), with access either via the local 
filesystem or over smart server support on a fast local network.

Things like the emacs tree -- with very many revisions, each of which 
touches relatively few files -- could easily have different characteristics.