Segmentation fault: 11 when trying to define a method which returns a new “instance” of a class

[paulyoung]

I’m sure I brought this up before but I ran into it again and couldn’t find anything about it.

Here’s a fairly minimal example that reproduces the issue when run with --check:

module {
  public class Id<A>(x : A) {
    public func map<B>(f : A -> B) : Id<B> {
      Id(f(x));
    };
  };
};

[ggreif]

Heh, if the good old identity functor fails, then all bets are off! ;-)

[ggreif]

I bet this is a mutual recursion problem. I have seen OCaml segfaulting when exhausting the stack several times.

[crusso]

Interesting...

[paulyoung]

If anyone has any suggestions on how to tackle this I can try to take a look.

[ggreif]

@paulyoung is this a curiosity problem ("let's code up some Haskell idioms in Motoko") or appearing in your production code?

[ggreif]

@nomeata you have squashed "not lazy enough reference" bugs in the past. Maybe this rings a bell?

[nomeata]

Not in the typechecker… :-)

[paulyoung]

is this a curiosity problem ("let's code up some Haskell idioms in Motoko") or appearing in your production code?

@ggreif I’m trying to do the former to see if there’s something interesting to write about and perhaps promote Motoko, but bugs like this are preventing me from reaching the latter.

[crusso]

I think this type should actually be rejected as expansive since it involves polymorphic recursion.

c.f.

module {
 /*
  public class Id<A>(x : A) {
    public func map<B>(f : A -> B) : Id<B> = Id(f x)
  }
  */

 type Id<A> = object { map : <B>(A->B) -> Id<B> }; // note polymorphic recursion
}

Hacky branch investigating the bug with sad attempts to alter the expansiveness check here are in branch
#3134

[ggreif]

I think this type should actually be rejected as expansive since it involves polymorphic recursion.

By which definition of polymorphic recursion? It would be polymorphic recursion when the type variable (A or B) were wrapped (nested) in a different type constructor, like Id<?B> or Id<(B, A)>. But Id<> applied to a tyvar isn't polymorphic recursion.

EDIT: Maybe you are right, after re-reading the wikipedia article — "where the type parameter changes with each recursive invocation" I can see that data Foo a b = Bar (Foo b a) in Haskell would also fall under polymorphic recursion. The question now is whether we can disregard the map method in some way to make the definition valid...