This is the mail archive of the guile@sources.redhat.com mailing list for the Guile project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: GC stuff


19-Jul-00 20:46 you wrote:
> It's not hard to find the top of the stack. You just need to figure out
> how to do it on your platform, and I guess we could just steal the code
> from the Boehm GC. It's just that nobody did it before... a perpetual
> problem with any project that depends on skilled volunteer labor.

Exactly.

> The benchmarks showed that Boehm was slower,

True.

> and subsequently it was found that there wasn't much tuning to be done.

Quite untrue. There are A LOT OF knobs and and switches in BGC. Just not
where you want to find them. It's all really simple in fact: BGC has TWO
interafaces - external and internal. Extenral one is for someone who what
it to "just work": plug BGC in your C (Pascal, Fortran, Ada, etc) and NEVER
call free (delete, dispose, etc) again. But if you need speed then you must
be ready for hand and dirty work: you can do a lot of tuning with internal
functions but for that you must know how BGC is working (at least some
basic facts) and what your program (guile here) expect from BGC. Of course
it's not clear if BGC will be fast enough even with tuning. Lets wait for
volunteer and see.

> So, there's no reason to dump our current GC (until someone comes up
> with better benchmark results for another GC).

Agree. If we have something what's already works there are no need to
replace it with something that works WORSE. As long as existing solution
really works better :-)

> I don't really understand why you people pay so much attention to
> GC... all the stuff I've done with Guile, the run time is dominated by
> I/O, and the the coding time is dominated by trying to understand or
> work around the module system.

Yes, it's true. Current module system are causing MUCH more problems then GC.




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]