expand

expand is a macro annotation that is kind of like @specialized, but it's more of a templating mechanism. It is pretty... alpha in that the functionality is basically there, but it is now in the least bit battle tested. Don't ask much of it, and it will do fine, ask a lot, and well...

Basically, expand takes a def with type arguments whose types are annotated with breeze.macros.expand.args and generates the cross product of all combinations. For example:

@expand
def foo[@expandArgs(Int, Double) T, @expandArgs(Int, Double) U](x: T, y: U) = x + y

will generate

def foo_T_Int_U_Int(x: Int, y: Int) = x + y
def foo_T_Int_U_Double(x: Int, y: Double) = x + y
def foo_T_Double_U_Int(x: Double, y: Int) = x + y
def foo_T_Double_U_Double(x: Double, y: Double) = x + y

The real power comes from breeze.macros.expand.sequence, which annotates an argument to the method to correlate with a type (the first argument to sequence) and then a sequence of trees which are inlined in place of references to the argument. For example:

@expand
def foo[@expandArgs(Int, Double) T](x: T, y: T)(implicit @sequence(T)({_ + _}, {_ * _}) op: XXX) = op(x,y)
/* The type of op is unimportant, though giving it a "real" type is useful. */

will generate

def foo_T_Int(x: Int, y: Int) = x + y
def foo_T_Double(x: Double, y: Double) = x * y

See breeze.linalg.DenseVectorOps for a more complete example.