digitalmars.D.learn - Getting NotNull Right
- =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (187/187) Feb 05 2014 Hi!
- Adam D. Ruppe (17/17) Feb 05 2014 The alias this could be changed to allow conversion to the base
- =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (4/23) Feb 05 2014 Members of a derived class T become inaccessible in NotNull!T
- Adam D. Ruppe (6/8) Feb 05 2014 aaah, of course, now alias this returns the base class instead of
- =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (5/13) Feb 05 2014 Thanks anyway.
Hi! I've changed Adam D Ruppes module notnull.d a bit to allow assignment of a NotNull inherited class instance to a NotNull base class using /** Assignment from $(D NotNull) Inherited Class $(D rhs) to $(D NotNull) Base Class $(D this). */ typeof(this) opAssign(U)(NotNull!U rhs) safe pure nothrow if (isAssignable!(T, U)) { this._value = rhs._value; return this; } This allows unittest { class A {} class B : A {} NotNull!B b = assumeNotNull(new B); NotNull!A a = assumeNotNull(new A); a = b; assert(a is b); } However I haven't figure out how to allow the follow code to compile unittest { class A {} class B : A {} void f(NotNull!A a) {} NotNull!B b = assumeNotNull(new B); f(b); } which I believe should work automatically. How do I make that happen? Complete source of module notnull.d follows: module notnull; import std.traits: isAssignable; /** Note that NotNull!T is not NotNullable :) */ alias NotNullable(T) = isAssignable!(T, typeof(null)); /** NotNull ensures a null value can never be stored. * You must initialize it when declared * You must never assign the null literal to it (this is a compile time error) * If you assign a null value at runtime to it, it will immediately throw an Error at the point of assignment. NotNull!T can be substituted for T at any time, but T cannot become NotNull without some attention: either declaring NotNull!T, or using the convenience function, notNull. Condition: T must be a reference type. Instead of: __traits(compiles, { T t; assert(t is null); }. TODO: Merge with http://arsdnet.net/dcode/notnullsimplified.d Examples: --- int myInt; NotNull!(int *) not_null = &myInt; // you can now use variable not_null anywhere you would // have used a regular int*, but with the assurance that // it never stored null. --- */ struct NotNull(T) if (NotNullable!T) { disable this(); // Disallow default initialized (to null) /** Assignment from $(D NotNull) Inherited Class $(D rhs) to $(D NotNull) Base Class $(D this). */ typeof(this) opAssign(U)(NotNull!U rhs) safe pure nothrow if (isAssignable!(T, U)) { this._value = rhs._value; return this; } NotNull!U opCast(U)() safe pure nothrow if (isAssignable!(U, T)) { return NotNull!_value; } // this could arguably break the static type check because // you can assign it from a variable that is null.. but I // think it is important that NotNull!Object = new Object(); // works, without having to say assumeNotNull(new Object()) // for convenience of using with local variables. /// Constructs with a runtime not null check (via assert()). this(T value) safe pure nothrow { assert(value !is null); _value = value; } /** Disable null construction. */ disable this(typeof(null)); /** Disable null assignment. */ disable typeof(this) opAssign(typeof(null)); private T _value; property inout(T) _valueHelper() inout { assert(_value !is null); // sanity check of invariant return _value; } // Apparently a compiler bug - the invariant being uncommented breaks all kinds of stuff. // invariant() { assert(_value !is null); } alias _valueHelper this; /// this is substitutable for the regular (nullable) type /* void toMsgpack (Packer) (ref Packer packer) const { packer.pack(_value); } */ /* void fromMsgpack(Unpacker)(auto ref Unpacker unpacker) { unpacker.unpack(_value); } */ } /** A convenience function to construct a NotNull value from something $(D t) you know isn't null. */ NotNull!T assumeNotNull(T)(T t) if (NotNullable!T) { return NotNull!T(t); // note the constructor asserts it is not null } /** A convenience function to check for null $(D t). If you pass null to $(D t), it will throw an exception. Otherwise, return NotNull!T. */ NotNull!T enforceNotNull(T, string file = __FILE__, size_t line = __LINE__)(T t) if (NotNullable!T) { import std.exception: enforce; enforce(t !is null, "t is null!", file, line); return NotNull!T(t); } unittest { import core.exception; import std.exception; void NotNullCompiliationTest1()() // I'm making these templates to defer compiling them { NotNull!(int*) defaultInitiliation; // should fail because this would be null otherwise } assert(!__traits(compiles, NotNullCompiliationTest1!()())); void NotNullCompiliationTest2()() { NotNull!(int*) defaultInitiliation = null; // should fail here too at compile time } assert(!__traits(compiles, NotNullCompiliationTest2!()())); int dummy; NotNull!(int*) foo = &dummy; assert(!__traits(compiles, foo = null)); // again, literal null is caught at compile time int* test; test = &dummy; foo = assumeNotNull(test); // should be fine void bar(int* a) {} // these should both compile, since NotNull!T is a subtype of T bar(test); bar(foo); void takesNotNull(NotNull!(int*) a) { } assert(!__traits(compiles, takesNotNull(test))); // should not work; plain int might be null takesNotNull(foo); // should be fine takesNotNull(assumeNotNull(test)); // this should work too assert(!__traits(compiles, takesNotNull(assumeNotNull(null)))); // notNull(null) shouldn't compile test = null; // reset our pointer assertThrown!AssertError(takesNotNull(assumeNotNull(test))); // test is null now, so this should throw an assert failure void takesConstNotNull(in NotNull!(int *) a) {} test = &dummy; // make it valid again takesConstNotNull(assumeNotNull(test)); // should Just Work NotNull!(int*) foo2 = foo; // we should be able to assign NotNull to other NotNulls too foo2 = foo; // including init and assignment } unittest { class A {} class B : A {} NotNull!B b = assumeNotNull(new B); NotNull!A a = assumeNotNull(new A); a = b; assert(a is b); }
Feb 05 2014
The alias this could be changed to allow conversion to the base class one step at a time. I'm not happy with it since it'd essentially do a big chain down to Object and it also ignores interfaces, but it's a start: static if(is(T == class) && !is(T == Object)) property NotNull!(BaseClassesTuple!T[0]) _valueHelper() inout { assert(_value !is null); // sanity check of invariant return assumeNotNull(cast(BaseClassesTuple!T[0]) _value); } else property inout(T) _valueHelper() inout { assert(_value !is null); // sanity check of invariant return _value; } (Previously, it just had the second _valueHelper)
Feb 05 2014
On Wednesday, 5 February 2014 at 21:32:36 UTC, Adam D. Ruppe wrote:The alias this could be changed to allow conversion to the base class one step at a time. I'm not happy with it since it'd essentially do a big chain down to Object and it also ignores interfaces, but it's a start: static if(is(T == class) && !is(T == Object)) property NotNull!(BaseClassesTuple!T[0]) _valueHelper() inout { assert(_value !is null); // sanity check of invariant return assumeNotNull(cast(BaseClassesTuple!T[0]) _value); } else property inout(T) _valueHelper() inout { assert(_value !is null); // sanity check of invariant return _value; } (Previously, it just had the second _valueHelper)Members of a derived class T become inaccessible in NotNull!T with this approach. Do you have any clue to why?
Feb 05 2014
On Wednesday, 5 February 2014 at 21:58:08 UTC, Nordlöw wrote:Members of a derived class T become inaccessible in NotNull!T with this approach. Do you have any clue to why?aaah, of course, now alias this returns the base class instead of the derived one. We could possibly work around it with opDispatch forwarding to the methods... or think of a new approach for the implicit conversion. blargh, i don't have a good answer right now
Feb 05 2014
On Wednesday, 5 February 2014 at 22:13:00 UTC, Adam D. Ruppe wrote:On Wednesday, 5 February 2014 at 21:58:08 UTC, Nordlöw wrote:Thanks anyway. If we get it right maybe we could propose it for Phobos :) /PerMembers of a derived class T become inaccessible in NotNull!T with this approach. Do you have any clue to why?aaah, of course, now alias this returns the base class instead of the derived one. We could possibly work around it with opDispatch forwarding to the methods... or think of a new approach for the implicit conversion. blargh, i don't have a good answer right now
Feb 05 2014