Idea: partially pure functions

[Steven Schveighoffer]

"Bruno Medeiros" wrote
> Yes, if g() is not pure, then the compiler will not check if v points to 
> global data or not. Thus we cannot guarantee that calling "f(v)" will not 
> change global state. But so what?
> That doesn't mean we shouldn't allow it. It just means such call may have 
> side-effects, and the compiler should threat that as a normal function 
> call (make no particular pure optimizations). Thus "partially" pure: pure 
> in some contexts, impure in other contexts.

I re-read this and I see what you are proposing.

I can see how that would work, but you would definitely need another 
keyword.  Any time a developer sees the pure keyword, he will assume he 
doesn't need to worry about thread issues.  However, a partially pure 
function would be subject to thread issues when not called from a pure 
function.

I still think it would be a worthwhile feature to have, as it would save on 
implementing lots of functionality more than once.

BTW, forget my other post, we were arguing about totally different things.

-Steve