Regression in function type checking

[vsmenon]

The following code used to work fine (no strong mode errors).

void foo(x) {
  print(x);
}

void bar(void f(int x), int x) {
  f(x);
}

void main() {
  bar(foo, 42);
}

It now gives a static error:

error] Type check failed: foo ((dynamic) → void) is not of type (int) → void because foo cannot be typed as (int) → void

I believe we used to treat foo as top -> void?

@leafpetersen

[vsmenon]

This is with analyzer tip. I don't see an error in 1.14.0-dev.5.0.

[jmesserly]

Let me try that as well after https://codereview.chromium.org/1678313002/ ... we were dropping the fuzzy arrows flag in a few places.

[jmesserly]

Oh, nvm, my CL only affects generic functions.

I thought we interpreted foo as ⊥ -> void when we're doing subtype checks between two function types?

[vsmenon]

I believe @leafpetersen had some logic to handle references to top-level functions - and treat the param type as top.

[jmesserly]

Oh, hmmmm. I haven't seen any code like that, but that would work.

Vijay, could this regression be caused by our change to get rid of UninferredClosure? We were messing with DownCast.create ... specifically making top level functions fall into an "error" path :)

[jmesserly]

But whatever it was, none of our checker tests caught it, so I'm not sure...

[jmesserly]

Actually, isn't this a dupe of #24507?
I'm going to merge it in.

[vsmenon]

Yeah, I think it's the same - thanks for merging. I think the "regression" is really a warning changing to an error - which is exactly what you're saying. :-)

Note, this is causing a break of sorts when I run ddc presubmit tests against analyzer tip, but only because one of the async tests we explicitly run at the end of browser_tests is no longer generated (due to warning promoted to error and no force compile I believe).

[jmesserly]

Note, this is causing a break of sorts when I run ddc presubmit tests against analyzer tip, but only because one of the async tests we explicitly run at the end of browser_tests is no longer generated (due to warning promoted to error and no force compile I believe).

DOH!

In the short term maybe work around it in the test? And I also bumped the priority of #24507. If we do want to special case original function declarations, we can certainly do that in DownCast.create.