Exploratory execution with the @amb operator.


@amb(state, options...)

x = @amb state 1 2 3


This piece of code says that x is either 1, 2, or 3. f can decide to admit a value or not using require. In effect you can think of this as calling f with a value appropriate for f as decided by f.

require(state, condition)

Require a certain condition to be true, if not, backtrack and try the next available set of assignments to @amb variables.

require is implemented as: require(state, cond) = !cond ? @amb(state) : nothing

Example: Pythagorean triples

julia> function int_between(state, lo, hi)
           require(state, lo <= hi)
           @amb lo int_between(state, lo+1, hi)

julia> function a_pythagoras_triple(state, lo, hi)
           i = int_between(state, lo, hi)
           j = int_between(state, i, hi)
           k = int_between(state, j, hi)

           require(state, i*i + j*j == k*k)
           return (i, j, k)


Runs a zero-argument function f in the amb-world, where the language knows that it needs to search for a set of assignments to @amb variables that runs the code till the end without rejection. If no such path is found, ambrun returns nothing.

In other words, ambrun computes one admissible return value of the function passed to it.

julia> ambrun((state)->a_pythagoras_triple(state, 1, 20))
(3, 4, 5)


return a Channel which gives back f() for all admissible assignments. collect on gives an array of all possible answers.

julia> collect(ambiter((state)->a_pythagoras_triple(state, 1, 20)))
6-element Array{Any,1}:
 (3, 4, 5)
 (5, 12, 13)
 (6, 8, 10)
 (8, 15, 17)
 (9, 12, 15)
 (12, 16, 20)

Thanks to Prof. Jerry Sussman for introducing us to amb in 6.945. Thanks to Jarrett Revels for writing Cassette!