click constraints [ref: ma]

Daniel Jackson dnj at lcs.mit.edu
Sat Jun 5 22:55:50 EDT 1999


eddie,

>Say you want a faction which models random lossage...
>
>I think maybe the thing to do is to use the flow patterns we've already
>got, but only add fake connections from agnostic ports to OTHER agnostic
>ports...

how about this example? a splitter that sends each input to one of two
outputs, alternating. you might want to have one output and one input pull,
and the other output push, so the system is essentially driven from the
pull output: it causes a packet to be taken off the input, and passed to
it, then it causes a second packet to be taken off the input and passed  by
a push through the other output. 

wouldn't you want to allow either output to be chosen as the push output?
so it's a faction with 3 agnostic ports (in, out0, out1), that can be
instantiated as

pull, pull, push
pull, push, pull
pull, pull, pull

in this case, it seems that the reason why we can have a personality
mismatch between the input port and one of the outputs is that actually the
third port is driving the computation.

/daniel



More information about the click mailing list