[<<][libprim][>>][..]
Mon Aug 24 11:03:53 CEST 2009

Different continuation types

Continuing the ideas in [1].  Starting with the following context:

  * A partial application is a binding of data context to code.
    Suppose you have an object referenced by `ob' and an operation
    `op' that expects such an object as the first parameter on the
    stack, the following code would be a closure:

          (ob op)

  * In Joy, code == data.  Its VM wouldn't need to distinguish pairs
    for code, partial applications and partial continuations.  Their
    difference is only visible through usage: they are all represented
    as lists.

  * In PF, these 3 need a different representation because code is
    nonlinear, while partial applications and continuations are
    linear.
    




What about this:

and stack = Empty | Next of datum * stack

and cont  = Done  | Next of sub   * cont
                  | Data of datum * cont

Wait.  There is already a mechanism to construct partial applications
as continuations: `dip' followed by `run'.

The question is then: what is best?  Explicit tagging (using different
continuation frame types) or ``prefix parsing''.

It seems to me that tagging is better behaved: in that case each frame
has its meaning isolated, and doesn't depend on the meaning of the
previous frame.  So be it: tagged frames.  This also makes `dip'
simpler.

[1] entry://20090823-093540




[Reply][About]
[<<][libprim][>>][..]