Given a list of continuations (with distinct handlers and sizes reflecting the pairing), how does the frame allocator use this information to actually allocate space for the continuation? What is the relationship (if any) between handler inference and frame allocation? -- Suresh