On 2019-01-08 03:44, Stuart Marks wrote:
This is one of Stuart's designs
Trying to blame it all on me, aren't you? :-)
Yes: you are being too nice! :-)
Overall the changes look good.
Thanks!
I can't really speak to how to improve the loops in certain
circumstances (the "tricky circular array traversal code from
ArrayDeque" as Martin puts it). But I observe that moving the
randomization to iteration has started to increase the complexity of the
code. I understand why we did this, but it's starting to look like we
might want to refactor the various loops into a single internal
"iterator" to avoid duplicating it all over the place. Perhaps for a
future enhancement.
Right, one generic loop method which accepts one of a few statically
defined Functions might cut it, but it would need to be implemented with
care not to regress anything and might not significantly reduce
complexity until we add a few more specializations.
I observe that the serialized forms aren't randomized. When the storage
order was randomized, the serial form was also randomized. But now that
randomization is done during iteration, the serialized forms are fixed.
Hmmm. Not sure what I think about that.
Doesn't it mean we're now actually better at randomizing iteration order
on a run-to-run basis, whereas if we just picked up and went with a
serialized storage we'd retain the iteration order you'd get in the old
VM instance?
Thanks!
/Claes
