On 2010-08-21, at 15:56, Max Carlson wrote:
> On 8/21/10 12:24 PM, P T Withington wrote:
>> Not approved:
>>
>> 1) I don't understand the point of this (which you have done in any number
>> of places):
>>
>>> (expr&& (expr is LzAttributeDescriptor)
> >
>> that would just seem to add a redundant test. if `expr` is null, it is not
>> going to be an LzAttributeDescriptor, so why add another conditional?
>
> I did this because the inlined is operator (in DHTML at least) is inefficient:
>
> It was expanding to:
> if(LzAttributeDescriptor["$lzsc$isa"]?LzAttributeDescriptor.$lzsc$isa($8):$8
> instanceof LzAttributeDescriptor){
>
> Testing for null saves some work:
> if($8&&(LzAttributeDescriptor["$lzsc$isa"]?LzAttributeDescriptor.$lzsc$isa($8):$8
> instanceof LzAttributeDescriptor)){
>
> I guess I'll change the compiler's inlining instead of hacking the source.
Are there really that many null arguments that this makes a difference? Surely
that must be in the noise level? If you change the compiler, then you are
going to insert this extra test everywhere, which will be a dis-optimization
for every time the LHS of `is` is not null.
If you really want to improve the compiler, then teach it how to read the
meta-information from the LFC and learn at compile time whether a type has an
`isa` method or not (only applies to types with mixins). Then it can emit the
correct code at compile time and not need the conditional expression at all.
>> 2) This can't work in as3, can it? I really don't like cheating like this...
>>
>>> + // play a little fast and loose - if we have a .ready slot,
>>> assume we're an event.
>>> + if (thisevent) {
>>> + if (thisevent.ready) { thisevent.sendEvent( args[ key ] ); }
>
> It seems to be fine, at least with all the sample apps. I don't like
> cheating either, but it does shave off a tiny bit of time...
In my experience, micro-optimizations like this, which are only 'loosely'
correct, will come back to bite you in the long run. It is better to give up a
CPU cycle and get the correct answer.
I am all for optimization, but not over-optimization. Remember the golden
rules of optimization:
1) Don't do it.
2) [For experts only] Don't do it yet.
:P
