Infinite loops on type solver issue affecting elm/compiler

[rupertlssmith]

SSCCE

module Try exposing (..)

f x =
    let
        g : (a -> ()) -> ()
        g h =
            h x
    in
    x


main : Platform.Program () () ()
main =
    Platform.worker
        { init = \_ -> ( (), Cmd.none )
        , update = \_ _ -> ( (), Cmd.none )
        , subscriptions = \_ -> Sub.none
        }

When compiling this with guida make, the compiler appears to infinite loop. The main is just there so you can compile this as an application.

Removing the type annotation on g it will run fine.
Add the type annotation f : a -> a it will also run fine.

The problem is potentially around introducing a new type variable in a let..in that affects the overall type of the outer function?

On the elm compiler, the compiler crashes with rank = 2: elm#2301

[rupertlssmith]

When guida infinite loops it is actually because it called Utils.Crash.crash with the String:

"You ran into a compiler bug. Here are some details for the developers:\n\n a [rank = 2]\n\nPlease create an http://sscce.org/ and then report it\nat https://github.com/elm/compiler/issues\n\n"

But crash does not print the string out. So it seems to be the same bug as in the elm compiler.

[decioferreira]

These might be related/relevant to this:

• Fix compiler bug with rank 2 types in mapAccuml elm/compiler#2342
• certain un-annotated declarations or incorrect let annotation → compiler thread blocked indefinitely in an MVar operation gren-lang/compiler#105

[decioferreira]

Also relevant: https://cs3110.github.io/textbook/chapters/interp/inference.html#let-polymorphism

[decioferreira]

This might be a simpler version of the same issue (ref: elm#1839):

-- one : a -> Int
one two =
    let
        three : a
        three =
            two
    in
    1