Thu Feb 13 06:45:39 EST 2020
Making a redo evaluator in C
Maybe the problem domain is not specified well: in practical
situations, I think that dynamic dependencies might not be such a big
deal. I.e. we will know the dependency network up front.
So a better way to do this is to us a pure dataflow structure with an
explicit memoized push/pull walker: start at changed input(s), and
recursively pull any of the affected outputs. It will be necessary to
have both o->i and i->o lists available.
It doesn't really look like rocket science. Question is more: do I
have a use for it?
I think atm a more explicit CSP style approach is appropriate on
It would be nice to reframe reactive code as CSP. Basically it is
just adding memoization: each process is responsible for its own
output node. Once it has evaluated, it will keep sending out the same
value until a global signal resets the "phase" of the algorithm. It
does seem rather inefficient to do it like that.