Re: Proposal: block-based vector allocator

[Stefan Monnier]

> I've tried a few designs of block-based vector allocation, and this is
> the first one which looks good enough to be proposed here.

I don't have time to look at it, but I like the idea of reducing the
cost of small vector allocation (currently O(log N) where N is the
number of malloc'd objects).

Could you give a quick overview of what you've considered and thrown away?

> This vector allocator uses near-to-PAGE_SIZE blocks to allocate small vectors
> and dedicated list for large vectors.

For large vectors we can probably just keep using what we have now.

> be satisfied from exact-fit free list, best-fit is tried, and unused
> tail bytes forms a free vector which is set up on an appropriate free
> list.

I don't like much splitting&coalescing.  Is it really needed?


        Stefan