Comments on: How slow is static access in as3/avm2 (exactly)

By: hristo

hristo — Mon, 12 Oct 2009 12:38:26 +0000

If you are looking for a way to make static access quicker, the answer is no; but listen, depending on what you are doing with static constants, you most probably don’t have to bother. (Event handling and type enums is a good example, you really don’t care about efficiency there.)

But when there is a lot of static stuff going on, I usually resort to Borg designs – I wrap the static functionality in a class with its access points on the instances, but where all data is shared internally as statics. Then I can just instantiate this class and keep a reference to its instances here and there where I need it and that way can access the same functionality in a much more flexible way. If you add in some abstraction, you might end up with a very sound code architecture actually.

My tweening engine works a bit like this if I remember correctly.

By: Dan

Dan — Sun, 11 Oct 2009 15:55:59 +0000

Is there a good workaround for keeping static constants fast? I suppose each class could make a copy of the static variables inside their own class, that way access is fast. But that is messy. What is a good way?

By: hristo

hristo — Sat, 12 Sep 2009 17:16:13 +0000

Nothing – the compiler leaves static constants as ‘variables’ on the class – and the code retrieves the values at runtime.
It would have been so easy to make the compiler inline what are references to compile-time available constant literals in the code (selectively maybe, but at least what’s short strings, integers, booleans and numbers). I am truly amazed by how a team of developers can undermine their own work – in this and many other cases the compiler openly offends the performance potential of the virtual machine.

By: Jethro Grassie

Jethro Grassie — Tue, 08 Sep 2009 10:32:25 +0000

While there are reasons to understand the ’static’ overhead, I wonder what happens when you make use of ’static const’, as this is used a lot in as3 classes to mimic enums.

By: hristo

hristo — Fri, 22 May 2009 10:04:59 +0000

Ha, true. I never use them, but I suppose it could be interesting.

By: Alan Shaw

Alan Shaw — Fri, 22 May 2009 03:36:36 +0000

Now I want to see how speed of access to package-level functions compares.

By: hristo

hristo — Tue, 21 Apr 2009 20:06:42 +0000

My pleasure, you raised a good point in the first place.

By: dVyper

dVyper — Tue, 21 Apr 2009 12:37:14 +0000

An excellent article! I personally haven’t come across any tests determining the speed of static access and this demonstrates the subject very well – thanks for spending time on this and sharing your findings! :)