Bad diagnostic when pattern matching nil that cannot be inferred

[hamishknight]

The following gives a bad diagnostic:

func foo<T>() -> T { fatalError() }
if nil ~= foo() {} // error: Expression pattern of type 'String' cannot match values of type 'Substring'

[Rajveer100]

@hamishknight If I correctly understand the issue, the function foo<T>() -> T is a generic function with type T and throws a fatalError().

Next, the if block checks if nil contains the function foo() (which I am unsure of, but am aware of the operator ~= used as range).

Let me know to what I am missing out, and some brief description to get an idea!

[AnthonyLatsis]

Next, the if block checks if nil contains the function foo()

Not quite. See Expression Pattern.

[Rajveer100]

Ok, I changed the snippet code this way, now it does throw fatalError():

import UIKit

func foo<T>() -> T { fatalError() }
if "nil" ~= foo() {} // error: Fatal Error

[hamishknight]

Indeed, but the nil case is invalid because there's no valid type to infer for it, and we should ideally diagnose as such. What's currently happening is that the constraint system is trying each overload of ~=, and coming up with a bunch of bad solutions, and it just so happens that the func ~= (lhs: Substring, rhs: String) -> Bool solution is the least bad one, and we attempt to diagnose based on that.

[Rajveer100]

I see!

[Rajveer100]

After some finding, I managed to find the call point in ConstraintSystem.h under this if condition:

else if (operatorName == ctx.Id_MatchOperator) {
      DE.diagnose(anchor->getLoc(), diag::cannot_match_expr_pattern_with_value,
                  lhsType, rhsType);
    }

So we could probably add an check inside this to first check for the nil case as in the issue and add a more specific message?

[Rajveer100]

@hamishknight What do you suggest?