[PATCH 0 of 3] Fast implemenation of parseindex in C

[Matt Mackall]

On Tue, 2008-10-14 at 00:39 +0200, Martin Geisler wrote:
> "Bernhard Leiner" <mailinglists.bleiner at gmail.com> writes:
> 
> > [*] Just for the record: measuring performance of a function that
> > returns a big data structure like this seems like a bad idea:
> >
> > for x in xrange(100):
> >     start = time.time()
> >     res = functions_that_returns_a_lot_of_data()
> >     stop = time.time()
> >     timing_results.append(stop - start)
> >
> > As soon as 'res' is overwritten a _lot_ of objects are freed and the
> > garbage collector kicks in and delays the following time measurement.
> > This is at least my current theory after staring at strange timing
> > results for too long.
> 
> There is a nice module for this kind of timing experiments:
> 
>   http://docs.python.org/library/timeit.html
> 
> You can even use it directly from the command line:
> 
>   python -m timeit 'functions_that_returns_a_lot_of_data()'

I posted a Mercurial extension that does much the same thing not long
ago (perf.py). I'll probably stick it in contrib/ as I've found it to be
very handy for this sort of tuning.

It's not clear whether disabling GC globally makes sense for performance
testing though.

-- 
Mathematics is the supreme nostalgia of our time.