Re: [jamboost] Some refactorings and speedups

From: Vladimir Prus (ghost_at_[hidden])
Date: 2003-02-12 01:20:08

Hi Jürgen,

> On Monday 10 February 2003 15:50, Vladimir Prus wrote:
> > At this moment, I've switched most of the code to using the new
> > class, and see about 5x speedup on two use cases. I also see
> > that execution profile changed considerably: the code which
> > does header scanning becomes hot spot.
> > Before going further, it would be really great to look on other
> > profiles. Jurgen, could you please rerun bjam on your project
> > and send me "-d+10"output?
>
> Yes, please find the gzipp'ed file attached.
> I've not run the whole build, but maybe 80%. I commented out some
> small projects while playing around.
> But I think this is still big enough ;-)

Thank you! The profilie is very nice: just like in my experiments, the hottest
spot in header scanning code. I'll see about turning header caching on.

> It looks _very_ fast. It's impressive: I'm seeing a speed improment of
> an order of magnitude. The response time for all libs is
> just about 2 minutes.

Oh.. to tell you truth, that's more of an improvement that I expected.

> Great Work, Volodya !

Thanks.

> One glitch: I'm specifying an external lib with
>
> lib LIBX : : <name>LIBX <search>/usr/local/LIBX/lib/linux-g++ ;
>
> in toplevel-Jamfile, but bjam does not add a -L path option to the
> linker, but only -lLIBX. So linking fails for me at the moment.

This means that we don't have tests for this functionality ;-)
I'll look into this later today -- should be easily fixable.

Thanks,
Volodya