actionscript 3 - What's happening when I use for(i in object) in AS3?

Question

To iterate over the properties of an Object in AS3 you can use for(var i:String in object) like this:

Object:

var object:Object = {

    thing: 1,
    stuff: "hats",
    another: new Sprite()

};

Loop:

for(var i:String in object)
{
    trace(i + ": " + object[i]);
}

Result:

stuff: hats
thing: 1
another: [object Sprite]

The order in which the properties are selected however seems to vary and never matches anything that I can think of such as alphabetical property name, the order in which they were created, etc. In fact if I try it a few different times in different places, the order is completely different.

Is it possible to access the properties in a given order? What is happening here?

Answer 1

I'm posting this as an answer just to compliment BoltClock's answer with some extra insight by looking directly at the flash player source code. We can actually see the AVM code that specifically provides this functionality and it's written in C++. We can see inside ArrayObject.cpp the following code:

// Iterator support - for in, for each
Atom ArrayObject::nextName(int index)
{
    AvmAssert(index > 0);

    int denseLength = (int)getDenseLength();
    if (index <= denseLength)
    {
        AvmCore *core = this->core();
        return core->intToAtom(index-1);
    }
    else
    {
        return ScriptObject::nextName (index - denseLength);
    }
}

As you can see when there is a legitimate property (object) to return, it is looked up from the ScriptObject class, specifically the nextName() method. If we look at those methods within ScriptObject.cpp:

Atom ScriptObject::nextName(int index)
{
    AvmAssert(traits()->needsHashtable());
    AvmAssert(index > 0);

    InlineHashtable *ht = getTable();
    if (uint32_t(index)-1 >= ht->getCapacity()/2)
        return nullStringAtom;
    const Atom* atoms = ht->getAtoms();
    Atom m = ht->removeDontEnumMask(atoms[(index-1)<<1]);
    if (AvmCore::isNullOrUndefined(m))
        return nullStringAtom;
    return m;
}

We can see that indeed, as people have pointed out here that the VM is using a hash table. However in these functions there is a specific index supplied, which would suggest, at first glance, that there must then be specific ordering.

If you dig deeper (I won't post all the code here) there are a whole slew of methods from different classes involved in the for in/for each functionality and one of them is the method ScriptObject::nextNameIndex() which basically pulls up the whole hash table and just starts providing indices to valid objects within the table and increments the original index supplied in the argument, so long as the next value points to a valid object. If I'm right in my interpretation, this would be the cause behind your random lookup and I don't believe there would be any way here to force a standardized/ordered map in these operations.

Sources
For those of you who might want to get the source code for the open source portion of the flash player, you can grab it from the following mercurial repositories (you can download a snapshop in zip like github so you don't have to install mercurial):

http://hg.mozilla.org/tamarin-central - This is the "stable" or "release" repository

http://hg.mozilla.org/tamarin-redux - This is the development branch. The most recent changes to the AVM will be found here. This includes the support for Android and such. Adobe is still updating and open sourcing these parts of the flash player, so it's good current and official stuff.

While I'm at it, this might be of interest as well: http://code.google.com/p/redtamarin/. It's a branched off (and rather mature) version of the AVM and can be used to write server-side actionscript. Neat stuff and has a ton of information that gives insight into the workings of the AVM so I thought I'd include it too.