I've just merged the branch and the fixes from not_gerd into master. Improvements like the binary search of arenas is a good idea, but is only incremental. It's not going to blow the benchmarks out of the water to replace it with a binary search at this point.
As plobsing pointed out on IRC (several times), the real fix is to move away from stack walking and make a proper, precise GC. I'm not sure if such a thing will be possible before M0, but there are real wins to be had if we can do it. If anybody has any ideas regarding that journey, I would be very interested to hear them. Thanks, --Andrew Whitworth On Mon, Sep 5, 2011 at 3:00 PM, Nicholas Clark <[email protected]> wrote: > On Mon, Sep 05, 2011 at 06:19:19AM -0700, Jimmy Zhuo wrote: >> I don't know how to binary search it, it's not an ordered array. > > I was assuming that it could be kept ordered. > > (And that the cost of keeping it ordered by inserting each new arena at the > right point in the middle is less than the benefits of being able to use a > binary search rather than a linear scan.) > > Nicholas Clark > _______________________________________________ > http://lists.parrot.org/mailman/listinfo/parrot-dev > _______________________________________________ http://lists.parrot.org/mailman/listinfo/parrot-dev
