Object.observe: change array truncation logic to efficiently handle large sparse arrays

Object.observe: change array truncation logic to efficiently handle large sparse arrays

Committed: https://code.google.com/p/v8/source/detail?r=13592

[adamk, 2013-01-24 21:58:06]

From a quick discussion with rafaelw it seems likely that what I actually want
here is some heuristic to decide between the new and the old approach, based on
how sparse the array is.

[adamk, 2013-01-24 23:35:52]

On 2013/01/24 21:58:06, adamk wrote:
> From a quick discussion with rafaelw it seems likely that what I actually want
> here is some heuristic to decide between the new and the old approach, based
on
> how sparse the array is.

One more possibility would be to ignore ordering and simply iterate directly
over the SeededNumberDictionary backing store, emitting deletes for those
indices >= new_length.

And a more refined version of that would be to sort after collecting the old
values.

I'm not sure how much optimization is worthwhile here at the moment; the biggest
win for observed array performance would be to find some way to use fast backing
stores.

[rossberg, 2013-01-28 13:46:46]

LGTM.

I would be fine with using a heuristic for dense arrays, although I'm not sure
it really matters. For the time being, I would oppose going to the backing store
directly, though -- don't break through abstraction layers unless you have a
very convincing case.

[adamk, 2013-01-31 02:03:30]

Updated to handle dense arrays.

[rossberg, 2013-02-04 13:09:55]

lgtm