[<<][compsci][>>][..]
Sat Mar 1 00:51:13 CET 2014

Data Direction and Control Flow: 2 x 2 = 4

A little arcane, but quite fun.  Do you push or pull data?  It
depends.

Some types, using dataflow parameters (as in Oz, easly
poor-man-modeled usin C++ references or C pointers.)

- sink       : write(from x)
- source     : read(to x)
- operation  : process(to y, from x)

These can be neatly composed:

sink * op = sink
op * source = source
op * op = op

Now, what I often forget is that these have duals.  There's a thing
that "puts something in a sink", and a think that "pulls something
from a source".

In practice, what are these co-objects (anti-objects)?  If sink,
source, and operations are models of data processing (push, pull and
flow), the co-objects correspond somehow to physical ports or the
operating system transferring control flow to a program when an event
occurs.  A co-sink is something that writes into a sink.  Note that an
co-sink is not a source!  The asymmetry is the caller/callee relation.


       |  caller    callee
-------+---------------------
sink   |  sends     receives
source |  receives  sends


process is caller for both send and receive.

then there's a missing 4th case: the buffer, which is callee for both
send and receive.

It seems that "push" programming (sink-oriented programming) is the
most natural, as it has pysical-time coinciding with execution on a
CPU.

So is "pull" programming (function evaluation) then only a model?  Is
the concept of evaluation just upside down?




[Reply][About]
[<<][compsci][>>][..]