Re: MLton's GC

George Neuner <gneuner2/@[EMAIL PROTECTED]
> writes:
>>Try http://www.ravenbrook.com/project/mps.
>
> Great reference.  Thanks!  
>
> Another neat thing to come out of the depths of Harlequin.  I guess
> the question is how well it works and how easily it can be integrated
> into a project that isn't Lisp or ML.  

It was not, as far as I'm aware, used in LispWorks or MLWorks (both
had well tuned GCs before MPS was created) so the question could be
turned around to ask how easily MPS could be used in a project that is
Lisp or ML :-)  Within Harleqin MPS saw most use in Dylan and PostScript.

The API might look a little complicated initially but that's because
it uses the same general API to cover quite different allocation
schemes e.g. manual fixed size, manual variable size, automatic mostly
copying, automatic mark&sweep, ... etc.  All it really boils down to
is initializing it, creating a pool of the desired type (say automatic
mostly copying) and then calling mps_{reserve,commit} to allocate an
object in that pool.

> I see it uses a 5 phase collection to support coordinating pools,
> rather than the 2 or 3 phase collection typical for an integrated
> heap.  I have to wonder what kind of overhead that imposes.

I'm fairly sure Richard and Nick would be happy to talk about the pros
and cons of the design choices.  There is mailing list for just such
topics: mps-questions@[EMAIL PROTECTED]