Re: generational conservative gc

[Miroslav Silovic <silovic@zesoi.fer.hr>]

mw@ipx2.rz.uni-mannheim.de (Marc Wachowitz) writes:

> Several years ago, I've read a report on a generational variant of
> Joel Bartlett's mostly-copying gc (I've heared it would be patented
> or something like that, though I don't know how much water that would
> hold, given that at least around here one can't patent algorithms).
> I don't have an exact reference ready, but if you search among DEC's
> published research papers, it shouldn't be too hard to find. As far
> as I remember, this gc (not necessarily the generational variant) is
> used in Modula-3 and in Scheme->C, but also available separately as
> collector for C or C++ (look for it on ftp://gatekeeper.dec.com/).

What exactly mostly-copying refers to, here? If it means that it will
avoid copying data referenced by C code, it may hold water. But with
the present Guile design, Scheme structures may be referenced from the
heap, and heap does *not* get scanned in Guile (it'd seriously impact
performance and introduce all the uncertainties associated with
fully-conservative GC if Guile does start scanning the heap). Since
interoperability with C code is one of the design goals for Guile,
copying GC is more trouble than it's worth. IMHO. :)

-- 
I refuse to use .sig