Boost logo

Boost-Build :

Subject: Re: [Boost-build] It's time for my per release beg for patch on ticket #2552 (Extending bjam to accept parameters 10-19)
From: Vladimir Prus (ghost_at_[hidden])
Date: 2009-11-10 02:22:47


On Wednesday 04 November 2009 Jon Biggar wrote:

> Vladimir Prus wrote:
> >>>> 1.41 is coming soon. Can the patch in this ticket please be integrated
> >>>> this time? It's been pending since 1.36.
> >>> What is the change in performance and memory consumption of bjam with
> >>> this patch, when building Boost C++ Libraries?
> >> It increases the storage requirements for bjam rule parameters from
> >> 9*sizeof(pointer) to 19*sizeof(pointer) for the memory allocation
> >> associated with each bjam rule invocation stack frame.
> >>
> >> It adds slight overhead when parsing rule text to find parameters $(10)
> >> through $(19).
> >>
> >> I expect the effects of either to be quite small.
> >
> > Given that you surely have both the patched version and the original version,
> > can you try? The gut feeling is that the change should not be huge, but gut
> > feelings are often wrong.
>
> Unfortunately my main home linux machine went down, so I had to do this
> on a VM which is somewhat memory constrained and accesses its storage
> over NFS. I ran 'time bjam stage' on the entire 1.40.0 distro and got
> these results:
>
> original bjam: 166.328u 167.922s 18:33.62 30.0%
> bjam with patch: 144.215u 157.788s 19:24.57 25.9%
> second run: 158.794u 171.369s 18:32.89 29.6%
>
>
> So with my setup, I can't detect any significant time difference.

Jon,

thanks for testing. I have now checked in the patch to trunk. It should
be available in 1.42 (we've missed 1.41 in a number of ways)

- Volodya

>
>
>


Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk