> On 23 Jan 2016, at 21:31, Andrew Haley <a...@redhat.com> wrote: > > BTW, does anyone here know why we don't have humongous ByteBuffers > with a long index? >
Probably in part because of Java arrays. IMO we need Arrays 2.0 (+ panama i suspect). Here’s some "shoot from the hip" thinking… In principle a direct ByteBuffer could bound a region larger than that which can be expressed with an int limit (i suppose a heap ByteBuffer could do the same if it holds multiple instances if byte[], although that would require more work). The direct BB can be used to access the first 2^31 - 1 values. A VarHandle could be produced that accesses the whole range since it could accept long index values and check bounds against a special long limit field. That might require some additional HotSpot work to optimize bounds checks and any loops to reduce or avoid a safe point check per iteration. Paul.
