For any references x and y, "x == y" immediately (synchronously) either returns true, returns false, or throws an Exception. If x and y are both Settled, then == must return either true or false. Therefore, if it throws an Exception, then at least one of the two operands was Unsettled at the time. If the sameness formulas of x and y are the same, then x == y returns true and we say they are are designationally equivalent. If x and y are both settled (ie, their sameness formulas are ground, ie, they contain no unbound variables, ie, their sameness identity does not depend on any promises), then if their sameness formulas are the same, then they designate the same object. If the sameness formulas of two settled references are different, then they designate different objects, and x == y returns false. The == predicate is monotonic, meaning that once it returns an answer for a pair of operands (rather than throwing an Exception), it will always return that same answer for the same pair of operands. Unsettled references don't normally provide enough information to determine what the answer will be when they're Settled, which is why == needs the option of not answering when the operands are Unsettled. As you might expect by these properties, and as you can see in the state transition diagram below, a reference can transition from being Unsettled to Settled, but once Settled always Settled. E's built-in hashtable-based collection abstraction, EMap, requires Settled keys. If an attempt is made to use an Unsettled reference as a key, they will throw an Exception rather than accept the key. When are two references the same?As with the Expanded Reference Taxonomy on the Reference Mechanics page, the following diagram also illustrates the possible transitions between the states of a Live reference, but now organized for purposes of reasoning about distributed reference sameness. Promises as Logic Variables*** explain PowerKeys*** explain ******* Just stale notes below this line. Please ignore *******
Currently, a == implementation is allowed (but not required) to answer true or false when either or both x and y are Unsettled, provided it is able to prove that they can't settle in such a way as to cause == to provide a different answer at a later time, which would violate monotonicity. Although it would be expensive, at some point we will probably tighten the spec (and the reference implementation) to require == to throw an Exception unless both operands are Settled, just as hashtable do with their keys.
Wraps a possibly unsettled base reference, in order to be able to use it as key in an EMap. This is called a PowerKey because it has a relationship to the base reference-as-key much like the relationship of a powerset to a base set. In E, by definition, any settled reference designates a particular settled object. Also by definition, a settled object, whether Selfish or Selfless, has a particular sameness identity. Given two settled references, x and y x == yreturns true or false according to whether x and y designate objects with the same sameness identity. Once a reference is settled, it is always settled to the same sameness identity, so answers to the above question are stable -- the same question asked at a later time will give the same answer. EMaps (hashtables) require such stability of their keys, and therefore only accept settled references as their keys. Were all references Near, Far, or Disconnected, there would only be settled references and settled objects. The other cases occur because of the other kinds of references, and because of their interaction with Selfless objects. The remaining kinds of references are Promises (or Unresolved references) and Unconnected references.
* * * * (XXX I suspect the following paragraph is stale, and that PowerKey now has better properties, like more stable sameess, as documented after this paragraph.) Used to finitely and successfully walk possibly cyclic & unsettled structures. A PowerKey wrapper always claims to be settled, though a cbx == cby that returns false now can return true later. Nevertheless, one can safely use CycleBreakers as keys in the built in hashtables (EMap, etc), since the hashCode they report to the map is stable. This object itself has stable sameness, which is why it may be used as a key in an EMap. It's sameness is no worse that a snapshot of the sameness-possibility-set of the value it wraps, where this snapshot is made when the PowerKey is constructed. Its sameness is no better than that of the object it wraps. For example, after the sequence def cb1 := makeCycleBreaker(p) # ... def cb2 := makeCycleBreaker(p)cb1 will always be == to cb1, and cb2 will always be == cb2. If p did not settle further during the "# ...", the cb1 will be == cb2. If it did settle some, then they may or may not be ==. If p3 != p4, then makeCycleBreaker(p3) != makeCycleBreaker(p4) |
||||||||||||
Unless stated otherwise, all text on this page which is either unattributed or by Mark S. Miller is hereby placed in the public domain.
|