From: Vladimir Prus (ghost_at_[hidden])
Date: 2002-12-15 10:27:35
David Abrahams wrote:
> Vladimir Prus <ghost_at_[hidden]> writes:
> >> > Of course, "bjam" run takes more time
> >> It's slower than v1? Or does it take more time than something else?
> > Slower than v1.
> > It's not surprising, given all the features that we want.
> It surprises me a little, since the architecture of v1 lends itself to
> so many inefficiencies.
The V2 architecture itself should be more optimizable. As the matters stand,
there are some aspect which were not optimized at all. Imagine a project where
all main targets are build with the same properties. For each main target
we'll process that properties over, looking for free feature, looking for
dependency feature, looking for propagated features, etc. My previous
profiling show that such manipulation take some time. It's not the only
problem, but you get the idea.
> > When feature set is more or less complete, we'd have to optimize it.
> I can help there.
> >> > but the switch to shared libs seem to save time and space.
> >> > (e.g. 3M vs. 37 M)
> >> Time and space during the build process, or in your project overall?
> > Space In the built project. And time during build.
> That's really great! Well, of course it saves some link time during
> the build. It'd be nice if we could save more time during the bjam
> processing. Well, there's still time to do that.
Yes, of course.
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