The following code prevents the compiler from terminating

I reported this to fsharp and it was suggested I should report it here as well.

Description: The idea was to make a function format that can return any function which’s type passes a static check. The functions that are valid take parameters (arbitrarily many) of type string or int and produce either a unit or an int. Including the checker function in another function would make sure that the type parameter is a valid function type, which would then be constructed unsafely via reflection or similar and then finally casted to the appropriate target type.

Unfortunately, this code causes the compiler to hang indefinitely (or as far as my patience extended). This might not be a bug if it’s a design decision to allow non-termination, but I’d find that strange considering my code consists only of valid language constructs.

Repro Steps: Compile this code

type Switcher = Switcher

let inline checker< ^s, ^r when (^s or ^r) : (static member pass : ^r -> unit)> (s : ^s) (r : ^r) = ()

let inline format () : ^r =
    checker Switcher Unchecked.defaultof< ^r>
    () :> obj :?> ^r

type Switcher with
    static member inline pass(_ : string -> ^r) =
        checker Switcher Unchecked.defaultof< ^r>
    static member inline pass(_ : int -> ^r) =
        checker Switcher Unchecked.defaultof< ^r>
    static member inline pass(_ : unit) = ()
    static member inline pass(_ : int) = ()

[<EntryPoint>]
let main argv = 
    let res : unit = format () "text" 5 "more text" ()
    printfn "%A" res
    Console.ReadKey()
    0 // return an integer exit code

Severity: It’s pretty bad that the compiler fails to terminate, but it’s a fairly obscure use case.

Version: Visual Studio 2015 Preview, my project targets .NET Framework 4.5.3 and the runtime is F# 4.0 (FSharp.Core, 4.4.0.0)

[forki]

@gmpl one thing is for sure: giving a error message immediately is better than compiler going into quasi infinite loop. So from my point of view your pr is a win for this case here.

[gusty]

After analyzing and working with the Constraint Solver I’m not surprised that some samples (like this) never ends. But after the fix I’m proposing they should. Now, whether they end successfully or with an error is a different thing. I need to study more this repro, but in principle I would say that if the code in the repro makes sense it should compile and if it don’t compile then we’re facing another compiler bug, similar to the ones I reported and they should be fixed in a separated pull request.