Machine: Fix arity mismatch with effect handlers due to evidence

[marzipankaiser]

For this code (https://github.com/effekt-lang/effekt/blob/master/examples/llvm/capabilities.effekt):

effect Emit(n: Int) : Unit


def loop(n: Int): Unit / Emit = {
  if (n < 0) {
    ()
  } else {
    do Emit(n);
    loop(n - 1)
  }
}

def main() = {
  try {
    loop(10)
  } with Emit { (n: Int) =>
    println(n); resume(())
  }
}

the following lifted core is generated:

module capabilities

import effekt

record Emit(Emit)

def loop(ev581, ev582, n, Emit$capability) =
  if (infixLt(n, 0)) {
    ()
  } else {
    Emit$capability.Emit(<ev582>, n);
    loop(<ev581>, <ev582>, infixSub(n, 1), Emit$capability)
  }

def main(ev583) =
  try { (ev584, Emit$capability) => 
    loop(<ev584, ev583>, <>, 10, Emit$capability)
  } with Emit {
    def Emit(n, resume) = 
      let _ = println(n);
      resume(<>, ())
  }

()

This, with the current machine transformation, leads to an arity mismatch in both the resume and the call of the effect operation.
A hotfix for this part of the issue is to:

• Always ignore the first parameter to resume (which is evidence).
• Ignore the first parameter passed into a handler (which is "its" evidence).

This PR implements this Hotfix and un-ignores the tests fixed by this.

[marzipankaiser]

@b-studios Is there always exactly one leading evidence parameter for calls to effect operations and resume?

[b-studios]

At the moment, yes. :)

[b-studios]

I think the following would be a bit more explicit about what you are doing:

val evidence :: arguments = values;
PushStack(Variable(transform(id), Type.Stack()),
  Return(arguments))

[b-studios]

Looks good to me, one one small nitpick.