Coding standards and type inference

With type inference you avoid “finger-typing” on both sides of an assignment:

Foo myFoo = new Foo()
var myFoo = new Foo() //Type-inference (no "win" here, but with parameterized types...)

Type inference also works with functions, allowing you to write:

def bar() : Foo = { ...etc... }
def bat(foo : Foo) = { ...etc... }

var myFoo = bar(); //Type-inference works fine

But I’m wondering if a coding standard that specified a type on the LHS of such assignments would be clearer:

Foo myFoo = bar();

Pro arguments:

  1. If you think type signatures are important for reasoning (which you presumably do if you’re using a type-inferred language), it follows that the explicitness is helpful; and
  2. If you change bar()s signature, the compiler’s going to complain about the call to bat (“Foo expected, found SomeNewType”) and that’s a little misleading.

Con arguments:

  1. Complaining about what’s passed to bat() is only “misleading” in the sense that people aren’t used to it; and
  2. Putting types on the LHS is an insult to referential transparency, since it is making explicit a difference in assignment use-cases

I am tending towards the “Con” arguments, but I think there’s merit to the “Pro” arguments. Anyone care to make the case that the “Pro” coding standard is superior?