Re: [jamboost] Performance degradation vs number of files

From: Pedro Ferreira (pedro.ferreira_at_[hidden])
Date: 2003-10-15 09:18:24

> > I can spend some time on it because we'd really like to be able to use
jam
> > on the regular development cycle and stop using Visual Studio projects
> > altogether. Currently, we're only using jam to create releases because
it's
> > impossible to wait this long every time one wants to compile.
>
> Could you give a bit more details. The profile you've sent to me off-list
> tells bjam takes 13 seconds to start up, which, of course, it not good.
What
> time would be acceptable?

Less than 5 s would be nice.

> Also, is your stree-test script representative of your project.

We've got a couple of main targets with more than 200 source files and a few
others with less than 50.

> I'm asking
> because I've found O(n^2) algorithm in V2 which might be the reason why
it's
> does not scale well with the number of targets.

That's what I thought.

> However, this algorithm does
> not scale when you add many sources to a main target. E.g one target with
> 1000 sources is worse for it than 20 targets with 50 sources. And 1000
> targets case is not very typical.

Right. The problem is that we must create a couple of DLLs based on
hundereds of source files.
If it were possible, I could create smaller static libs and then link them
together to create the large DLLs. However, I don't think this is possible,
at least in Windows, due to having to explicitly export the symbols.

Thanks for your help,

Pedro