backporting features to D1

[Frank Benoit]

Christopher Wright schrieb:
> Frank Benoit wrote:
>> How can you guess anything here?
>> 
>> Old behavior == just the direct call, no extra cost New behavior ==
>>  same direct call plus a heap allocation with unbound cost
> 
> There are plenty of randomized algorithms with potentially unbounded 
> cost that are still used. In practice, potentially unbounded things 
> work in a reasonable amount of time or get replaced.
> 

You said it is just a thing of optimization.
That would be true if the runtime would have been increased by a defined
value, like double runtime. But it has changed from deterministic
behavior to undeterministic behavior.
=> that means you can no longer use that in code that is made for high
data rate or low latency. And optimizations will not change that.

There is also no sense in doing benchmarks, because you are
measuring the GC performance, nothing more. And I already know, that I
need the guarantee for no heap allocation.

My point is, the delegate are a D1 tool that can be used in high
performance code. Shortly after the full closure feature was announced
(Nov 07), it was shown that this brakes D1 runtime behavior. Still this
is not fixed.

> While this does violate Tango's contracts about heap allocation, once
>  scope delegates are available, it won't take much work to replace 
> "delegate" with "scope delegate" everywhere in Tango.

I understand this as a confirmation, that with actual D2 you cannot port
tango and expect it to work as it does with D1.
Even an optimization cannot help. IMO the only two possible solutions are
1. Redesign the code -or-
2. Wait for scoped delegate