bzr-usertest now supports benchmarking on large history sets

[Ian Clatworthy]

Code has just been pushed for this much needed enhancement to the
usertest (benchmarking) plugin. See
http://people.ubuntu.com/~ianc/plugins/usertest/doc/usertest-tutorial.html#benchmarking-on-large-history-sets
for an introduction. In summary, passing an archive containing a
standalone tree now works for benchmarking bzr using usertest. (It ought
to work for hg, git and a svn checkout as well though I haven't tested
any of those.)

I've chosen a standalone tree as the format because it's really easy to
make one, particularly for any project on Launchpad. Simply go to a
directory not in a shared repo and do this on Linux:

  bzr branch lp:my-project my-project
  tar czf my-project-repo-x.y.tar.gz my-project/

If you have a local project, you either already have a standalone tree
or you can make one using a minor variation of the above.

You can then run usertest on it like this:

  bzr usertest my-project-repo-x.y.tar.gz

It would be sweet to support benchmarking shared repos with history as
well soon, as that's our recommended way of working. That requires a bit
more setup for the user (init-repo, etc.) and more intelligence inside
usertest to detect it and automagically change into the branch directory
(not the top level one) before running the tests. I wonder if we'd be
better accepting a standalone tree regardless and enhancing reconfigure
to turn a standalone tree into a shared repo with a nested branch inside it?

FWIW, a quick test on running the default suite (common) on build 3213
on the archive bzr-repo-1.1.tar.gz shows 'bzr log' as a standout slow
operation: 5 seconds vs less than a second for most others. That
suggests that some of the other tests, to be more meaningful, need to
look at some project-specific interesting files (like NEWS), rather than
the boring ones I create during the benchmark. I'll look into that soon.
(It's increased configuration so I've avoided that to date.)

As always, any questions, please ask.

Ian C.