> That's what I really don't understand, why do we need to generate so > many complex data items in order to invoke a single subroutine? The > whole operation should be possible with a fraction of the data > shuffling that we do currently. It's not just PMCs that we're > creating, either. There are lots of little structures created here and > there that aren't GCable. We need a serious overhaul of these > mechanisms to get this right, sooner rather then later.
I was thinking about that some days ago. We have lots of little structures, to support pmc's ATTR, for example, that have fixed size and never are realloc'ed. A special purpose allocator can avoid lots of calls to the mem_sys_alloc family. -- Salu2 _______________________________________________ http://lists.parrot.org/mailman/listinfo/parrot-dev
