Full API docs available here
@nogc
and@safe
- Shows the intent of your code that may or may not return a value
Optiona!int fun() {} // Might return an int, or might not
- Use pattern matching
fun.match!( (int value) => writeln("it returns an int"), () => writeln("did not return anything"), );
- Safely call functions on classes that are null or structs that don't exist
class C { int fun() { return 3; } } Optional!C a = null; a.dispatch.fun; // no crash, returns no!int
- Forwards any operator calls to the wrapped typed only if it exists, else just returns a
none
Optional!int a = 3; Optional!int b = none; a + a; // evaluates to some(6); a + b; // evaluates to no!int;
- Copmatible with
std.algorithm
andstd.range
fun.each!(value => writeln("I got the value")); fun.filter!"a % 2 == 0".each!(value => writeln("got even value"));
The purpose of this library is two fold, to provide types that:
- Eliminate null dereferences - Aka the Billion Dollar Mistake.
- Show an explicit intent of the absence of a value
This is done with the following:
Optional!T
: Represents an optional data type that may or may not contain a value that acts like a range.NotNull!T
: Represents a type that can never be null.dispatch
: A null-safe dispatching utility that allows you to call methods on possibly null values (including optionals, andstd.typecons.Nullable
)
An Optional!T
signifies the intent of your code, works as a range and is therefor useable with Phobos algorithms, and allows you to call methods and operators on your types even if they are null references - i.e. safe dispatching.
You can use this library:
- When you need a type that may have a value or may not (
Optional!Type
) - When you want to safely dispatch on types (
possibleNullClass.dispatch.someFuncion // safe
) - When you want a guaranteed non null object (
NotNull!Type
) - When you want to not crash with array access (
some([1, 2])[7] == none // no out of bounds exception
)
Lets take a very contrived example, and say you have a function that may return a value (that should be some integer) or not (config file, server, find operation, whatever), and then you have functions add1 and add2, that have the requirements that they may or may not produce a valid value. (maybe they do some crazy division, or they contact a server themselves to fetch a value, whatevs).
How can you go about this?
int* add1(int *v) {
// Gotta remember to protect against null
if (!v) {
return v;
}
*v += 1;
return v;
}
int* add2(int *v); // might forget to check for null
void f() {
int* v = maybeGet();
if (v)
v = v.add1;
if (v)
v = v.add2;
if (v)
writeln(*v);
}
You can also replace int* with Nullable!int and then instead of if (v)
you'd have to do if (!v.isNull)
and instead of *v
you'd do v.get
.
There's std.range.only:
auto add2(Range)(Range r)
if (isInputRange!Range && is(ElementType!Range == int))
// constrain to range type only and int element type?
// I need to ensure it has a length of one.
// And there's no way to ensure that in compile time without severly constraigning the type
{
// do we have one element or more now?
// what do we do if there's more than one?
// do we restrain it at run time to being there?
enforce(r.walkLength <= 1); // ??
// Should we map all of it?
return v.map!(a => a + 1);
// Or just the first?
return v.take(1).map!(a => a + 1);
// But what do I do with the rest then?
}
auto add2(Range)(Range r) if (isInputRange!Range) {
// same headache as above
}
void f() {
auto v = maybeGet();
// can we assign it to itself?
v = v.add1.add2;
// No, no idea what it returns, not really the same type
// so this...
refRange(&v).add1.add2; // ??
// no that won't work (can it?), lets create a new var
auto v2 = v.add1.add2 // and let type inference do its thing
writeln(v2); // now ok.
}
auto add1(Optional!int v) {
v += 1;
return v;
}
auto add2(Optional!int v); // same as above
void f() {
auto v = maybeGet().add1.add2;
writeln(v);
}
Well yes, you can, but you can also stick a pencil up your nostril. It's a bad idea for the following reasons:
- In order to achieve stability, you have to enforce checking for null. Which you cannot do
- Null is part of the value domain of pointers. This means you can't an optional of null
- The caller doesn't know who owns the pointer returned. Is it garbage collected? If not should you deallocate it?
- It says nothing about intent.
It is not like the Nullable
type in Phobos. Nullable
is basically a pointer and applies pointer semantics to value types. It does not give you any safety guarantees and says nothing about the intent of "I might not return a value". It does, however, tell you if something has been assigned a value or not. Albeit a bit counterintuitively, and in some cases nonsensically:
class C {}
Nullable!C a = null;
writeln(a.isNull); // prints false
This may make sense when null is part of the value domain of a type - i.e. for pointers. But with classes you end up having to write code like this:
void f(T)(Nullable!T a) {
if (!a.isNull) {
static if (is(T == class) || (T == interface) || /* what else have I missed? */) {
if (a.get !is null) {
a.callSomeFunction;
}
} else {
a.callSomeFunction;
}
}
}
It is also NOT like std.range.only
. D's only
cannot be used to signify intent of a value being present or not, nor can be used for safe dispatching, nor the result of only(value)
be passed around. It's only (heh) usage is to create a range out of a value so that values can act as ranges and be used seamlessly with std.algorithms
. This Optional
has a type constructor - some
- that can be used for this purpose as well.
In this section we'll see how this Optional is similar to Scala's Option[T]
and Swift's Optional<T>
type (similar to Kotlin's nullable type handling)
Idiomatic usage of optionals in Swift do not involve treating it like a range. They use optional unwrapping to ensure safety and dispatch chaining. Scala on the other hand, treats optionals like a range and provides primitives to get at the values safely.
Like in swift, you can chain functions safely so in case they are null, nothing will happen:
D: Unfortunately the lack of operator overloading makes dispatching a bit verbose.
class Residence {
auto numberOfRooms = 1;
}
class Person {
Optional!Residence residence = new Residence();
}
auto john = some(new Person());
auto n = john.dispatch.residence.numberOfRooms;
writeln(n); // prints [1]
Swift
class Person {
var residence: Residence?
}
class Residence {
var numberOfRooms = 1
}
let john: Person? = Person()
let n = john?.residence?.numberOfRooms;
print(n) // prints "nil"
Like in Scala, a number of range primitives are provided to help (not to mention we have Phobos as well)
D
auto x = toInt("1").orElse(0);
import std.algorithm: each; import std.stdio: writeln;
toInt("1").each!writeln;
toInt("1").match!(
(i) => writeln(i),
() => writeln("😱"),
);
// For completeness, the implementation of toInt:
Optional!int toInt(string str) {
import std.conv: to;
scope(failure) return no!int;
return some(str.to!int);
}
Scala
val x = toInt("1").getOrElse(0)
toInt("1").foreach{ i =>
println(s"Got an int: $i")
}
toInt("1") match {
case Some(i) => println(i)
case None => println("😱")
}
// Implementation of toInt
def toInt(s: String): Option[Int] = {
try {
Some(Integer.parseInt(s.trim))
} catch {
case e: Exception => None
}
}
Also like in Swift, you can unwrap an optional to get at it's value:
D
auto str = "123";
if (auto number = toInt(str).unwrap) {
writeln(*number);
} else {
writeln("could not convert string ", str);
}
Swift
let string = "123"
if let number = Int(str) {
print(number) // was successfully converted
} else {
print("could not convert string \(string)")
}
The following section has example usage of the various types
import optional;
// Create empty optional
auto a = no!int;
assert(a == none);
++a; // safe;
a - 1; // safe;
// Assign and try doing the same stuff
a = 9;
assert(a == some(9));
++a; // some(10);
a - 1; // some(9);
// Acts like a range as well
import std.algorithm : map;
import std.conv : to;
cast(void)some(10).map!(to!double); // [10.0]
cast(void)no!int.map!(to!double); // empty
auto r = some(1).match!((int a) => "yes", () => "no",);
assert(r == "yes");
static class C { void f() {} }
static struct S { void f() {} }
void f0(NotNull!C c) {
c.f();
}
void f1(NotNull!(S*) sp) {
sp.f();
}
auto c = notNull!C;
auto sp = notNull!(S*);
f0(c);
f1(sp);
static assert(!__traits(compiles, { c = null; }));
static assert(!__traits(compiles, { sp = null; }));
static assert(!__traits(compiles, { c = new C; }));
// Safely dispatch to whatever inner type is
struct A {
struct Inner {
int g() { return 7; }
}
Inner inner() { return Inner(); }
int f() { return 4; }
}
auto d = some(A());
// Dispatch to one of its methods
d.dispatch.f(); // calls a.f, returns some(4)
d.dispatch.inner.g(); // calls a.inner.g, returns some(7)
// Use on a pointer or reference type as well
A* e = null;
// If there's no value in the reference type, dispatching works, and produces an optional
assert(e.dispatch.f() == none);
assert(e.dispatch.inner.g() == none);