Futures (and other runtime structures) get corrupted during runtime

[EliasC]

See #47 for an example of this bug in action. During debugging, I have seen the address being used for a future reused for a stream link (struct scons) before they were used, leading me to suspect the garbage collector. Maybe adding tracing functionality to futures and stream links (and any other runtime structures) is enough (although we also need a proper root set).

[albertnetymk]

The below code snippet contains cyclic dependency on purpose. The mem allocated for the stream would be reclaimed when start is finished. Hopefully, it provides some insight.

#!/usr/bin/env encorec -run

-- a Sequence of Integers
class IntSeq

  stream start(a: Main, fr : int, to : int) : int {
    print "intseq";
    while fr <= to {
      yield fr;
      fr = fr+1
    };
    get a.dummy()
  }

class Main
  def dummy() : void
    print "Over"

  def main() : void
    let
      lst = 0
      str = (new IntSeq).start(this, 1,100000)
    in {
      while not eos str {
        lst = get str;
        str = getNext str;
        print lst;
      };
      print lst -- never reach here due to cyclic dep
    }

[TobiasWrigstad]

Can you be more specific? What other structures?

What happens if you replace pony_alloc with malloc for the structures?

[EliasC]

I've seen it for sure with the structures for streams (i.e. struct scons), and there the problem goes away if malloc is used everywhere.

[EliasC]

I think this is no longer a problem (similar reasoning as #91, plus largestream.enc is now passing). Closing.