Crystal: Local variable considered defined even if it's only defined in a noreturn branch

[straight-shoota]

if 1 < 0
  local_var = true
  raise "foo"
end

local_var # => nil

In the last line, local_var can't actually be defined because it's only defined in a branch that raises. The assignment in line 2 could never have happened before the variable is accessed in line 6.

Outside the if branch, it should be treated like it was never assigned in the first place:

if 1 < 0
  raise "foo"
end

local_var # Error: undefined local variable or method 'local_var' for top-level

It's probably not a super important feature, but it would be useful if the compiler could detect that.

This has surfaced in #9376: After extracting a noreturn branch to a separate method, the compiler complained about an undefined local variable when it wasn't defined for the main branch in the first place.

[jhass]

I guess this is because local variables are a parser level feature, while branch analysis only happens in semantic.

[asterite]

Yes, I was going to reply what jhass said.

[RX14]

I thought all local-var definitions are method-scoped, the type was just nil if it was impossible to be assigned in that branch of the code.

[RX14]

should this issue be closed?

[straight-shoota]

Why? I think this is a valid feature enhancement. It's not going to be implemented anytime soon. But it does not seem to be impossible.

[RX14]

It's a really really wide-ranging language change if we change the scoping rules for local variables... I don't see this happening but maybe others want to weigh in.

[asterite]

Outside the if branch, it should be treated like it was never assigned in the first place

That's the current behavior: not assigned means nil.

I don't think there's anything to do here.

[straight-shoota]

Maybe this is something a linter like ameba could help with.

[Sija]

@straight-shoota IIRC it already does that :)