Why runtime ...wouldnt region analysis show this ? eg you have a tree in the analyis , on the condition when a branch gets sheared insert code to create the seperate regions for each root of the branch unless another part the tree references it . Especially in the mixed region / GC or ref couting shared world ( where unknown / regions cant be formed it would be either GCs/ or ref counted).
On Tue, Sep 3, 2013 at 3:27 PM, David Jeske <[email protected]> wrote: > On Mon, Sep 2, 2013 at 11:47 PM, Sandro Magi <[email protected]>wrote: > >> On 31/08/2013 12:59 PM, David Jeske wrote: >> >>> Rather than splitting into strictly two subsets (reachable and >>> non-reachable) they do a union-find to split into N disjoint object graphs >>> - which become the new post-split regions. (good luck doing that without >>> stop the world) >>> >> >> The lazy techniques Bacon and Petrank pioneered for the Recycler would >> work here to virtually eliminate stop-the-world, >> > > Just a minor point of clarity... It isn't the cycle find which I am > suggesting would create pause problems, it's the difficulty of subdividing > the live portions of a region into N separated regions (because an > aggregate object tying them together was removed) while keeping concurrent > runtime region inference coherent. > > _______________________________________________ > bitc-dev mailing list > [email protected] > http://www.coyotos.org/mailman/listinfo/bitc-dev > >
_______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
