• bearophile (56/56) Dec 27 2012 In some situations I need to pass to a function a "nullable"

"bearophile" <bearophileHUGS lycos.com> writes:

In some situations I need to pass to a function a "nullable" 
fixed sized array. This happens if you use fixed-sized arrays a 
lot. So:
- You can't use ref, because it can't be null.
- Passing to the function a raw pointer and then using it as  
(*arr)[x]  doesn't look nice.
- NullableRef is a solutions, but it contain enforces that kill 
inlining and make the code slower, so it's not usable for 
performance-sensitive code (unless you turn the nullable into a 
regular pointer inside the function, but this makes using 
NullableRef much less useful).

This solution is derived by NullableRef:


struct Ptr(T) {
     private T* ptr;

     this(T* ptr_) pure nothrow {
         this.ptr = ptr_;
     }

     bool opCast(T)() const pure nothrow if (is(T == bool)) {
         return ptr !is null;
     }

      property ref inout(T) get()() inout pure nothrow
     in {
         assert(ptr);
     } body {
         return *ptr;
     }

     alias get this;
}

Ptr!T ptr(T)(ref T x) {
     return typeof(return)(&x);
}

// Example usage ---------------------------------------

alias TA = immutable(int[5]);

bool foo(Ptr!TA arr=Ptr!TA.init) nothrow {
     if (arr)
         return arr[1] == 20;
     else
         return false;
}

bool foo(typeof(null) _) nothrow {
     return false;
}

void main() {
     assert(!foo());
     assert(!foo(null));
     TA items = [10, 20, 30, 40, 50];
     assert(foo(ptr(items)));
}


The usage in foo() is clean.

I've created a kind of benchmark and unfortunately I've seen that 
with DMD (compiling with -O -release -inline -noboundscheck) this 
very light wrapping is not as efficient as a pointer to 
fixed-sized array :-(
(But it's better than NullableRef).

Bye,
bearophile