David Hanley <·····@netright.com> writes:
> It is interesting to see copy-on-read used to increase heap locality,
> but, unless I am mistaken, isn't this what a copying gc will generally
> do? I.E. the nodes of a list, even if allocated distinctly, should be
> copied into the same page, generally speaking.
Yes, but they generally do it with a GC-directed graph traversal, whereas the
TI Explorer would mostly let the mutator do it so that it would not be layed
out in a breadth-first or depth-first manner but in a "actual access pattern"
manner. In other words, if your graph traversal is fairly constant, one GC of
your datastructure will make sure that you will later step through memory
sequentially.
Stefan