Running multiple commands on commandserver interface

[Matt Mackall]

On Thu, 2012-04-12 at 19:50 +0300, Idan Kamara wrote:
> On Thu, Apr 12, 2012 at 6:55 PM, Matt Mackall <mpm at selenic.com> wrote:
> 
> > On Thu, 2012-04-12 at 17:33 +0200, Ton Plomp wrote:
> > > Yes Idan,
> > >
> > > removing the null byte on the client side did fix the issue!
> > >
> > > For timing info I ran 100 hg version, versus 100 times version on the hg
> > > commandserver:
> > > Pipe: 4,1 seconds
> > > CLI: 15 seconds
> >
> > Hmm, that's a lot slower than I would have hoped. It runs about 10x
> > faster here (100 loops .5 seconds with python-hglib), but it seems to
> > still be doing a bunch of unnecessary work on each call.
> 
> 
> Did you call rawcommand or version on the repo object? Because the latter
> is cached.
> 
> This is what I'm getting:
> 
> $ python -m timeit -n 1000 -s 'import hglib; r = hglib.open()' -c
> 'r.rawcommand(["version"])'
> 1000 loops, best of 3: 30 usec per loop

30 usec is suspiciously short. I got:

1000 loops, best of 3: 0 usec per loop

..which rather strongly suggests that something's confusing timeit here.

A loop of 100 calls to rawcommand takes ~.5 seconds here.

> But version is probably not the best candidate for a benchmark.

It's actually a great candidate because it does so little work: almost
all the time is the overhead we're trying to measure. Granted, some of
that overhead is not overhead for "real" commands.

-- 
Mathematics is the supreme nostalgia of our time.