On Neko targets, SignalBase.remove() fails to find and remove the listener.

[scriptorum]

My game which has a significant number of entities slows to a crawl on Neko. It happens whenever an entity is removed from a node list while the Ash engine is mid-update. ComponentMatchingFamily removes the entity and node, puts the node on-deck for the node pool, then hooks into Engine.updateComplete. When the engine finishes updating, the callback releases the on-deck node into the pool. It does this part fine, but then the callback tries to remove itself from the updateComplete signal. This fails. The signal remains, and over a time the signals accumulate, dragging the app to a standstill.

The failure appears to be in SignalBase.remove. In two locations, nodes in the linked list are compared to the listener, but the comparison returns false on Neko.

       node = head;
        while (node != null)
        {
            if (node.listener == listener)
            {
                foundNode = true;
                break;
            }
            node = node.next;
        }

CPP makes the compare successfully. I changed the comparison to this:

            if (Reflect.compareMethods(node.listener, listener))

And that seems to fix the problem. There are other comparisons in that method which may suffer from the same comparison problem, I don't know, but changing these two fixed an egregious problem for me.

I'm on Neko 1.x and Haxe 2.1, but I've had this problem for a while and I'm pretty sure it was present when I was messing around with Neko 2.

[nadako]

Thanks, right now neko is not supported (mainly because i was too lazy), but I'm gonna look into this. Looks like one can't just compare functions using == and must use Reflect.compareMethods. Gotta research on this thing some more.

BTW, why (and how) do you use neko? I thought it's not very popular for games, and only used for scripts and server-side web.

[scriptorum]

Yeah, I was surprised that function reference comparison is so fickle on Neko. I really love Haxe, but there's still a lot of cross-target gotchas like this.

I use Neko as a target for development testing only because it compiles so quickly. Flash also compiles quickly, but Neko dumps console output to stdout, which makes me a happier logger. Flash just dumps traces straight to the screen, which drives me nutso. HaxePunk has a debug console that improves that situation, but it's not nearly as useful as stdout console output. If I could figure out a clever way to write Flash traces to a file and tail -f that file, I'd switch to Flash, which is often the final target anyhow.

[nadako]

I kinda fixed it the same way but with more testing, so I'm closing this issue.

[scriptorum]

Thanks Dan!