Should types of match and if expression arms match?

[JSAbrahams]

We can either:

• Make sure they always match, making the language more strict.
• Allow them to diverge. Where we use them we then make not that this is a type union.
  This falls more in line with the language philosophy I think of being strict while also allowing a certain level of flexibility more familiar to Python users.

[JSAbrahams]

See

mamba/src/check/constrain/mod.rs

Line 64 in 63cc218

fn it_stmt_as_expression_none() {

mamba/src/check/constrain/mod.rs

Line 65 in 63cc218

let src = "def a := if True then 10 else None";

[JSAbrahams]

And

mamba/tests/system/valid/control_flow.rs

Line 58 in 63cc218

test_directory(true, &["control_flow"], &["control_flow", "target"], "if_two_types")

Which also refers to source:

def a := if True then 10 else "asdf"

[JSAbrahams]

I'd say that they shouldn't match perse, as we have a union system which is built to deal with such cases.
But we need to update the type system so that it can deal with such unions.

Might get a bit complicated, not sure, since we might start to have circular dependencies here.
Current issue is that we check for an access on a certain type, but if at a later date we discover an expression may be another type, we do not check for an access on that type (or us) since we marked the previous one as resolved.
So we need a unification system where we can revisit old constraints and re-evaluate them.

Until we properly implement such a system, this issue must remain open.