Can any programming language infer domains of a function from the domains of its input functions?

Question

So I like test-driven development and frequently I have wanted a programming language to assist with the creation of test routines. I have wanted to define the domain of a function and then have some test 'buddy' generate some random test input based on the domain, some test data deliberately inside the domain and some deliberately outside to test the error case, also some deliberately on the boundary which is where more errors can occur.

It seems possible with a custom C# attribute adorned on a method parameters. However, I would like an extra level of cleverness I would like inferred domains, inferred from the domains of feeder functions. I should introduce some example code at this point, the following is VBA and illustrates a case where (given enough language support) the domain of i(x) could be inferred by looking at the domains of g(x) and h(x).

Option Explicit

'* I'd like to define the domain of g(x) to be 2 <= x <= 5 , whilst g(x) maps x->x+1
Function g(x)
    If Not (2 <= x And x <= 5) Then g = CVErr(2011): Exit Function '* VBA domain screened with If

    g = x + 1
End Function

'* I'd like to define the domain of h(x) to be -4 <= x <= 4, whilst h(x) maps x->x^2
Function h(x)    
    If Not (-4 <= x And x <= 4) Then h = CVErr(2011): Exit Function '* VBA domain screened with If

    h = x ^ 2
End Function

Function i(x)
    '* it would be nice if a computer language could understand that the domain of i(x) can 
    '* be inferred from the domains
    '* of g(x) and h(x) and is thus 2 <= x <= 3 
    '* obviating the need for me to write the line explicitly

    i = h(g(x))
End Function

Sorry if this is a contrived example but real world instances do arise in my experience.

I guess in formal terms the language would need 'declarative definition of function domains'.

So could, in theory, a programming language support this?

Could this be related to Constraint programming - Wikipedia?

Other Links:

• Ceballos, Gasca, Borrego : Constraint satisfaction techniques for diagnosing errors in Design by Contract software - ACM SIGSOFT Software Engineering Notes 31(2) 2006

Answer 1

Let D(f) be the domain of a function f, C(f) its codomain and & the intersection operator. Let further g^-1(M) the domain values of g which map into a certain set M. Then one can write the domain of i as

 D(i) =  g^-1 ( C(g) & D(h) )

So as you see, there is more involved than just the domains of g and h, one needs to know the codomain of g and the reverse mapping g^-1 for an arbitrary subset of that codomain.

This boils down to knowing the exact behaviour of the function g, not just those domains. So except for very simple or restricted functions g, the only general way to achieve this is probably to fill some map for every value of D(g), and calculate the reverse mapping for each value of C(g) & D(h) one by one. This might be possible for small, finite sets D(g) and D(h), but it becomes unfeasible when D(g) and D(h) are too huge to fit fully into the available memory space (for example, when x in g(x) is a double value or a 64 bit integer with no further restrictions).

Note further this becomes impossible (in general) if g behaves "nasty" and goes into an endless loop. Then the analyser would need to solve the halting problem.