[gclist] optimal heap expansion factor?

[Perry E. Metzger]

Fergus Henderson <fjh@cs.mu.OZ.AU> writes:
> I'm in the process of rewriting it so that it instead initially uses
> just a small fraction of the space before collecting.  Then after each
> collection, it will recompute an appropriate size at which to do the
> next collection, based on how much live data remained after collection.
> This approach will hopefully improve locality and reduce working set
> size for programs that don't have much live data.

One question. On a system like Unix, you can easily use mmap to create
and destroy semispaces for collection at will. Why not simply start
with small semispaces and grow them as you need them, instead of just
using portions of larger semispaces?

Perry