A little of Partial Compilation

[JAnderson]

Robert Fraser wrote:
> JAnderson wrote:
>> Walter Bright wrote:
>>> Don wrote:
>>>> Note that even if all parameters to a function are known at compile 
>>>> time, doesn't mean it's sensible to run it at compile time. 
>>>> (Example: calculating pi to fifty billion decimal places). The 
>>>> programmer needs to annotate it somehow to tell the compiler that is 
>>>> suitable for compile time.
>>>
>>> It's one of those impossible problems for the compiler to predict if 
>>> it should attempt to execute a function at compile time or not. 
>>> That's why compile time function evaluation is initiated with a 
>>> specific syntax, rather than tried in general.
>>
>> Perhaps that could be solved by automatic profiling.  The worst 
>> offenders could be put into a list and the compiler would spend longer 
>> on those functions to optimize them.  Also it would be something you 
>> could turn on and off with a compiler setting.
>>
>> -Joel
> 
> Optlink already has a feature like this. You can profile your 
> application at runtime and then feed back to optlink a file which will 
> cause it to layout the functions optimally in the binary [1]. Possibly a 
> similar concept (possibly even the same file) could be fed to DMD to 
> prioritize its optimizations.
> 
> [1] http://www.digitalmars.com/ctg/trace.html

I know but I think its more about ordering off functions rather then 
turning them into templates.

-Joel