[GHC] #10780: Weak reference is still alive if key is alive, but weak reference itself not reachable

[GHC]

#10780: Weak reference is still alive if key is alive, but weak reference itself
not reachable
-------------------------------------+-------------------------------------
              Reporter:  atze        |             Owner:  simonmar
                  Type:  bug         |            Status:  new
              Priority:  normal      |         Milestone:
             Component:  Runtime     |           Version:  7.10.2
  System                             |
              Keywords:              |  Operating System:  Unknown/Multiple
          Architecture:              |   Type of failure:  Runtime
  Unknown/Multiple                   |  performance bug
             Test Case:              |        Blocked By:
              Blocking:              |   Related Tickets:
Differential Revisions:              |
-------------------------------------+-------------------------------------
 We've been running into some unexpected behavior with weak references. The
 use case it hackage package HMap, which is a (heterogenous, but that is
 besides the point) map from keys to values. Since there is no way to
 enumerate all elements in the map, there is no way to obtain a value for a
 key if the key is dead.

 As an optimization, I therefore used weak references to the values in the
 map, such that if the key dies, then the value is also dead. I was under
 the impression that a weak reference is alive if:
 (1): The key is still alive
 AND
 (2): The weak reference itself is alive (i.e. reachable from root set)

 However, the actual behavior is that a weak reference is alive if (1), (2)
 does not matter. Hence, the weak references here actually create a space
 leak, because an alive key will keep all values in all maps that it was
 ever used in alive.

 I think it would be desirable, to change the behavior to (1) AND (2),
 instead of just (1). This would prevent space leaks in this use case
 instead of creating them. Or would this cause problems in other use cases?

 See also: https://github.com/atzeus/HMap/issues/5#issuecomment-132538324

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10780>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler