| I'm working on a library/framework implementing vaguely dataflow-like abilities. It is extremely unfinished at this point, but can be reached as a darcs repository at http://www.rntz.net/darcs/arc/flow/. It expects to be in its own subdirectory of the libs/ directory in arc, and it will only work in anarki (it requires a lot of hacking with plt scheme under the hood, and also makes heavy use of the utility functions in anarki's lib/util.arc). The basic idea is that you define "dataflow-aware" functions using the special form 'flowdef, and these functions will cache their values (as with 'defmemo) until any of the computations they depend on change - these computational dependencies being automatically determined by what other dataflow-aware functions are called by the function. Then, you can specifically force a given computation to be updated in various ways, and when you call any function which would depend on it (transitively), it will be recomputed. A simple example: arc> (flowdef test (x) (prn "test: " x) x)
#<procedure: test>
arc> (flowdef incby () 1)
#<procedure: incby>
arc> (flowdef inctest (x) (prn "inctest: " x) (inc (test x) (incby)))
#<procedure: inctest>
arc> (test 0)
test: 0
0
arc> (test 0)
0 ; the value is cached
arc> (inctest 1)
inctest: 1
test: 1
2
arc> (inctest 1)
2 ; again, cached
arc> (flowdef incby () 2)
*** redefining incby
#<procedure: incby>
arc> (inctest 1)
inctest: 1 ; no longer cached, because 'incby changed
3
There is also a simple and barely functional example of using this framework to implement a make-like system in build.arc in the repository. |
