Proposed improvements to the separate compilation model

[deadalnix]

On Sunday, 3 March 2013 at 13:24:39 UTC, Dicebot wrote:
> On Sunday, 3 March 2013 at 13:19:35 UTC, deadalnix wrote:
>> ...
>
> Ah, beg my pardon, misunderstood you a bit. Still, difference I 
> am trying to point is that virtual vs final does not rely on 
> actual method implementation at all, while proper CTFE function 
> needs to be written with CTFE in mind as different subset of 
> language can be used. If one function will be found CTFE-able 
> by accident and used in such, any change to its implementation 
> that will add non-CTFE features (preserving public function 
> interface and behavior) will break user code.

Yes both are technically very different. But you argue for 
CTFEability by default or when choosen with almost the same 
arguments.

I explained my oppinion before on that : CTFEability can be 
decoupled from actual source code using some bytecode. Bytecode 
should be opaque enough to make that work.