Fibonacci benchmark fails with Stack Overflow on NET Core, but not on Framework

[ArbilGit]

The following fails with Stack Overflow on NET Core, but runs fine on Framework:

 let rec fib n = job {
      if n < 2L then
        return n
      else
        let! x = fib (n-2L)
        let! y = fib (n-1L)
        return x + y
 }
 fib 30L |> run

The call stack full of four functions repeated over again suggests a tail call problem:

Possibly related to #192 which I've just tested , similarly works fine on Framework but fails on Core. Which seems strange to me: if Core and Framework share the compiler to IL, and the JIT compiler (RyuJIT), shouldn't the tail call optimization succeed on both, or fail on both simultaneously?

Tested on Windows 10 x64, NET Core 3.0 Preview 2

[haf]

@ArbilGit You're welcome to look into it. Otherwise, thanks for reporting it.

[ArbilGit]

@haf I've been trying, but without success thus far. Workaround discovered: compile with CoreRT. It's a JIT issue, then.

[haf]

@ArbilGit Perhaps it would be prudent to file and issue with the .net/coreclr team, then.

[haf]

@ArbilGit Filed a F# lang suggestion, see link above.