Expanding the horizons of D purity

[H. S. Teoh]

On Tue, Nov 12, 2013 at 12:06:25AM +0400, Denis Shelomovskij wrote:
> 07.11.2013 21:41, H. S. Teoh пишет:
> >On Thu, Nov 07, 2013 at 09:22:47PM +0400, Denis Shelomovskij wrote:
> >>01.11.2013 0:05, H. S. Teoh пишет:
> >>>I'd like to propose extending the scope of weak purity one step
> >>>further: allow weakly-pure functions to call (not necessarily pure)
> >>>delegates passed as a parameter.
> >>
> >>Unacceptable. It will break strongly purity. Your mistake is
> >>strongly pure function can easily get an impure delegate:
> >[...]
> >
> >It will not. My proposal only applies to weakly pure functions.
> >Strongly pure functions cannot call impure delegates because you
> >cannot guarantee anything about what the delegate will do.
> >
> >So basically, once a pure function takes a non-pure delegate
> >parameter, it is no longer strongly pure, only weakly pure. (I stated
> >this at the end of my post where strongly pure functions are not
> >allowed to call impure delegates.)
> >
> >
> >T
> >
> 
> Let me repeat my code again:
> ---
> int i;
> 
> struct S
> { void g() { ++i; } }
> 
> void impureCaller(scope void delegate() del) pure
> { del(); }
> 
> void f() pure // strongly pure, will modify `i`
> {
>     S s;
>     impureCaller(&s.g);
> }
> ---
> 
> Currently the only error is in `impureCaller` body. You propose to
> make `impureCaller`'s body valid thus making the code compilable. Do
> you also propose some breaking changes to make `f` body invalid?
[...]

It should be illegal to take the address of S.g in f() if f is pure.


T

-- 
You have to expect the unexpected. -- RL