Closures of different types that do not capture the environment are automatically coerced to function pointers

[shepmaster]

I tried this code:

fn returns_closure(maybe: bool) -> impl Fn(i32) -> i32 {
    if maybe {
        |x| x + 1
    } else {
        |y| y + 2
    }
}

I expected that the compilation would fail because the closures have different types:

error[E0308]: `if` and `else` have incompatible types
 --> src/main.rs:5:9
  |
2 | /     if maybe {
3 | |         |x| x + 1
  | |         --------- expected because of this
4 | |     } else {
5 | |         |y| y + 2
  | |         ^^^^^^^^^ expected closure, found a different closure
6 | |     }
  | |_____- `if` and `else` have incompatible types
  |
  = note: expected type `[closure@src/main.rs:3:9: 3:18]`
          found closure `[closure@src/main.rs:5:9: 5:18]`

Instead, the code compiles.

Meta

Fails to compile (as expected)

rustc +1.44 --version --verbose:

rustc 1.44.1 (c7087fe00 2020-06-17)
binary: rustc
commit-hash: c7087fe00d2ba919df1d813c040a5d47e43b0fe7
commit-date: 2020-06-17
host: x86_64-apple-darwin
release: 1.44.1
LLVM version: 9.0

Compiles (unexpectedly)

rustc +1.45 --version --verbose:

rustc 1.45.2 (d3fb005a3 2020-07-31)
binary: rustc
commit-hash: d3fb005a39e62501b8b0b356166e515ae24e2e54
commit-date: 2020-07-31
host: x86_64-apple-darwin
release: 1.45.2
LLVM version: 10.0

rustc +stable --version --verbose

rustc 1.53.0 (53cb7b09b 2021-06-17)
binary: rustc
commit-hash: 53cb7b09b00cbea8754ffb78e7e3cb521cb8af4b
commit-date: 2021-06-17
host: x86_64-apple-darwin
release: 1.53.0
LLVM version: 12.0.1

rustc +nightly --version --verbose

rustc 1.55.0-nightly (7c3872e6b 2021-06-24)
binary: rustc
commit-hash: 7c3872e6bfd555d2ad753ac1f871db3bd7f2a547
commit-date: 2021-06-24
host: x86_64-apple-darwin
release: 1.55.0-nightly
LLVM version: 12.0.1

[shepmaster]

Looking at the MIR, I see this:

fn returns_closure(_1: bool) -> fn(i32) -> i32 {

I am surprised that the compiler automatically converted my code into dynamic dispatch.

[shepmaster]

Friends have pointed me towards #71599 as the causal root.

[danielhenrymantilla]

What about designing an allow-by-default lint triggering on such coercions? implicit_dyn_dispatch_coercion or something along those lines.