A WeakMap where values (not keys) are weak references?

# /#!/JoePea (8 years ago)

Hey all,

I'd like to have something like a WeakMap where the keys can be primitives and the values are weakly-held Objects, so if there are no more references to any of the Object values that the entry gets removed from the map.

For example, it might look like this:

{
  ["foo"] => SomeObject,
  ["
​bar​
"] =>
​OtherObject​
,
}

where if there are no more references to OtherObject, then `['bar'] =>

OtherObject` is removed from the map.

Usage would be very similar to WeakMap, like

let m = new ReversedWeakMap

m.add('foo', SomeObject)
m.add('
​bar​
',
​OtherObject​
)
​console.log(m.get('bar')) // OtherObject

... time passes, no more references to OtherObject, OtherObject is
collected ...

console.log(m.get('bar'))​ // undefined

I thought of using WeakMap values as keys, and vice versa, but it seems to be impossible.

I guess maybe it is difficult to add this natively because it would mean that GC needs to be completely deterministic as far as JS programs go. For example:

let m = new ReversedWeakMap

function main() {
  m.set('foo', {})
  console.log(m.get('foo')) // {}
}()

main()

// GC would have to be guaranteed to have happened at this point.

console.log(m.get('foo')) // undefined

# T.J. Crowder (8 years ago)

What are your use cases for it?

Rather than tying it to being a Map, I'd prefer to see something like Java's WeakReference: You could store the WeakReferences in a Map if you wanted strongly-referenced keys with weakly-referenced values.

-- T.J. Crowder

# Denys Seguret (8 years ago)

Having java-like weak references would indeed solve many problems and would let us build our own collections.

# Boris Zbarsky (8 years ago)

On 3/21/17 2:15 AM, /#!/JoePea wrote:

I'd like to have something like a WeakMap where the keys can be primitives and the values are weakly-held Objects

It might be worth searching the archives: this has been proposed multiple times and discussed at quite some length.

If there isn't an existing document that summarizes the discussions (and quick searching doesn't find one in the top several hits, though it does find esdiscuss.org/topic/what-is-the-status-of-weak-references at the very least), there really should be one...

# Mark S. Miller (8 years ago)

What you are asking for is sometimes called a "WeakValueMap". It is a very sensible abstraction and sometimes exactly what is needed. It is easy to build using Maps (or WeakMaps, depending) combined with < tc39/proposal-weakrefs>.

The old wiki.ecmascript.org shows an implementation of a WeakValueMap in terms of these. Unfortunately the wiki still seems to be offline.

Since there are still outstanding links into it, how did the wiki get collected? Were these links weak? ;)

# Guy Bedford (8 years ago)

I hope it wouldn't be too redundant to ask here again if there's been any movement on the WeakRefs proposal? WebAssembly at 1.0 does seem to bring new needs for this.

# Benoit Marchant (8 years ago)

+💯

# Michael Kriegel (8 years ago)

I am also still very interested in Weak references / a weak map, where the values are held weakly. Here is my example: I want to easily build caches, which lose their values, when there are no other references to them. Example pseudo code (assuming server request answers synchronously for simplicity):

var Cache = new WeakValueMap();

function request(Key) {

 var CachedResult = Cache.get(Key);

 if (CachedResult) return CachedResult;

 var UnprocessedResult = doServerRequest(Key);

 var BigProcessedObject = doVeryHeavyComputingTask(UnprocessedResult);

 Cache.set(Key,BigProcessedObject);

 return BigProcessedObject;

}

Thoughts:

  • doHeavyComputingTask takes a long time, so it should be done only once per key (until it is uncached)

  • BigProcessedObject is really big, it shall be in memory at most once at any time

  • When the last reference to a BigProcessedObject is lost, it shall be removed from cache to save memory

  • removal could be delayed by holding a reference to a BigProcessedObject and releasing it using a setTimeout-Timer.

Without WeakValueMap I have to build my own reference counting, so everyone who requests the object for a Key has to "unrequest" it again, else it would provoke a memory leak. This implies, that the life cycle of an object is tracked, so each object, which has a reference to a BigProcessedObject needs something like a destructor. As there is no destructor in JS you have to write a method and manually call it. Thus each object needs to have a single known owner or reference counting/tracking has to be done for that object, too. This propagates through your whole program. Finally as a developer you end up building your own garbage collector, which:

  • is error prone

  • consumes extra memory, as there are already all structures present for garbage collecting in the JS engine, just it is not usable

  • takes up lots of time for development and debugging

So it is a/the classical example case for tc39/proposal

# T.J. Crowder (8 years ago)

Right -- but it's really weak references you're after, right? Because with weak references, a Map can be a WeakValueMap, but other structures are also possible. With that proposal's executor feature, the map can even be proactive removing entries for objects that have become only weakly reachable.

I'd like to see weak references (perhaps not quite that exact proposal, but along those lines), not just a WeakValueMap. I've only needed them rarely (your example is a good one), but when I've needed them, they were pretty much the only option... I don't mind an out-of-the-box WeakValueMap as well, I just don't want it instead.

-- T.J. Crowder

# Michael Kriegel (8 years ago)

Totally agree. WeakRef is what we want and WeakValueMap is what I want to build with it. Sugar would be if the standard provides "an out-of-the-box WeakValueMap as well".

Automatic removal of entries after they became obsolete would be necessary - be it integrated or via a callback mechanism (i guess the executor is sth. like that.)

# Michael Kriegel (8 years ago)

BTW are the security concerns mentioned by Erik Arvidsson in this thread (mozilla.6506.n7.nabble.com/What-is-the-status-of-Weak-References-tp271746p271750.html) still present?

Wouldn't a solution be, to make the weak references of one actor dispose as soon as this actor has no strong references to the object (no matter whether there are other actors having strong references to that same object)?

Also the last commit in tc39/proposal-weakrefs was 9 months ago - is there any way to find information on he status of that? Is there still anyone working on it?