In duncantl/RTypeInference: Infer Types of Inputs and Outputs for R Expressions

Literal Propagation

Consider the simple system

x#1 <=> IntegerType
y#1 <=> x#1

Solving this system proceeds as follows:

1. Unify x#1 and IntegerType; the substitution x#1 => IntegerType is added to the solution set.
2. Apply the solution set to the remaining constraints, which yields the system Soln: x#1 => IntegerType Cons: y#1 <=> IntegerType
3. Repeat steps 1-2.

What if a constraint is encountered that can't be solved? For example:

y#1 <=> x#1
x#1 <=> IntegerType

This case shows the need to apply the new solution to the existing solution set at each step. That is,

1. Unify x#1 and y#1. Apply the new solution to both the constraint set and the solution set, so: Soln: y#1 => x#1 Cons: x#1 <=> IntegerType
2. Unify x#1 and IntegerType. Apply the new solution to both the contraint set and the solution set, so: Soln: y#1 => IntegerType x#1 => IntegerType

Unresolved Calls

Consider this simple example:

y#1 <=> UnresolvedCall(length, x = x#1)

The length function always returns an integer, so the type of x#1 is not even needed. Resolution should produce this solution:

y#1 => IntegerType

x = 3i
y = 4 + x

The constraint set is:

x#1 <=> ComplexType
y#1 <=> UnresolvedCall(+, RealType, x#1)

The solution set is:

x#1 => ComplexType
y#1 => ComplexType

What should happen if one of the constraints has an as yet unknown type? And how can this happen in examples?

# Assume x is a parameter.
y = 4 + x

duncantl/RTypeInference documentation built on Jan. 16, 2021, 12:30 a.m.