Expression: CommaExpression
An expression is a sequence of operators and operands that specifies an evaluation. The syntax, order of evaluation, and semantics of expressions are as follows.
Expressions are used to compute values with a resulting type. These values can then be assigned, tested, or ignored. Expressions can also have side effects.
For any expression expr, the full expression of expr is defined as follows. If expr parses as a subexpression of another expression expr1, then the full expression of expr is the full expression of expr1. Otherwise, expr is its own full expression.
Each expression has a unique full expression. Example:
return f() + g() * 2; The full expression of g() * 2 above is f() + g() * 2, but not the full expression of f() + g() because the latter is not parsed as a subexpression.
Note: Although the definition is straightforward, a few subtleties exist related to function literals:
return (() => x + f())() * g(); The full expression of f() above is x + f(), not the expression passed to return. This is because the parent of x + f() has function literal type, not expression type.
The following expressions, and no others, are called lvalue expressions or lvalues:
Expressions that are not lvalues are rvalues. Rvalues include all literals, special value keywords such as __FILE__ and __LINE__, enum values, and the result of expressions not defined as lvalues above.
The built-in address-of operator (unary &) may only be applied to lvalues.
Given an expression expr that is a subexpression of a full expression fullexpr, the smallest short-circuit expression, if any, is the shortest subexpression scexpr of fullexpr that is an AndAndExpression (&&) or an OrOrExpression (||), such that expr is a subexpression of scexpr. Example:
((f() * 2 && g()) + 1) || h()
The smallest short-circuit expression of the subexpression f() * 2 above is f() * 2 && g(). Example:
(f() && g()) + h()
The subexpression h() above has no smallest short-circuit expression.
Built-in prefix unary expressions ++ and -- are evaluated as if lowered (rewritten) to assignments as follows:
| Expression | Equivalent |
|---|---|
| ++expr | ((expr) += 1) |
| --expr | ((expr) -= 1) |
Therefore, the result of prefix ++ and -- is the lvalue after the side effect has been effected.
Built-in postfix unary expressions ++ and -- are evaluated as if lowered (rewritten) to lambda invocations as follows:
| Expression | Equivalent |
|---|---|
| expr++ | (ref x){auto t = x; ++x; return t;}(expr) |
| expr-- | (ref x){auto t = x; --x; return t;}(expr) |
Therefore, the result of postfix ++ and -- is an rvalue just before the side effect has been effected.
int i = 0; assert(++i == 1); assert(i++ == 1); assert(i == 2); int* p = [1, 2].ptr; assert(*p++ == 1); assert(*p == 2);
Binary expressions except for AssignExpression, OrOrExpression, and AndAndExpression are evaluated in lexical order (left-to-right). Example:
int i = 2; i = ++i * i++ + i; assert(i == 3 * 3 + 4);
OrOrExpression and AndAndExpression evaluate their left-hand side argument first. Then, OrOrExpression evaluates its right-hand side if and only if its left-hand side does not evaluate to nonzero. AndAndExpression evaluates its right-hand side if and only if its left-hand side evaluates to nonzero.
ConditionalExpression evaluates its left-hand side argument first. Then, if the result is nonzero, the second operand is evaluated. Otherwise, the third operand is evaluated.
Calls to functions with extern(D)linkage (which is the default linkage) are evaluated in the following order:
Example calling a function pointer:
voidfunction(int a, int b, int c) fun() { writeln("fun() called"); staticvoid r(int a, int b, int c) { writeln("callee called"); } return &r; } int f1() { writeln("f1() called"); return 1; } int f2() { writeln("f2() called"); return 2; } int f3(int x) { writeln("f3() called"); return x + 3; } int f4() { writeln("f4() called"); return 4; } // evaluates fun() then f1() then f2() then f3() then f4() // after which control is transferred to the callee fun()(f1(), f3(f2()), f4());
Expressions and statements may create and/or consume rvalues. Such values are called temporaries and do not have a name or a visible scope. Their lifetime is managed automatically as defined in this section.
For each evaluation that yields a temporary value, the lifetime of that temporary begins at the evaluation point, similarly to creation of a usual named value initialized with an expression.
Termination of lifetime of temporaries does not obey the customary scoping rules and is defined as follows:
If a subexpression of an expression throws an exception, all temporaries created up to the evaluation of that subexpression will be destroyed per the rules above. No destructor calls will be issued for temporaries not yet constructed.
Note: An intuition behind these rules is that destructors of temporaries are deferred to the end of full expression and in reverse order of construction, with the exception that the right-hand side of && and || are considered their own full expressions even when part of larger expressions.
Note: The ConditionalExpressione1 ? e2 : e3 is not a special case although it evaluates expressions conditionally: e1 and one of e2 and e3 may create temporaries. Their destructors are inserted to the end of the full expression in the reverse order of creation.
Example:
import std.stdio; struct S { int x; this(int n) { x = n; writefln("S(%s)", x); } ~this() { writefln("~S(%s)", x); } } void main() { bool b = (S(1) == S(2) || S(3) != S(4)) && S(5) == S(6); }
S(1) S(2) S(3) S(4) ~S(4) ~S(3) S(5) S(6) ~S(6) ~S(5) ~S(2) ~S(1)First, S(1) and S(2) are evaluated in lexical order. Per the rules, they will be destroyed at the end of the full expression and in reverse order. The comparison S(1) == S(2) yields false, so the right-hand side of the || is evaluated causing S(3) and S(4) to be evaluated, also in lexical order. However, their destruction is not deferred to the end of the full expression. Instead, S(4) and then S(3) are destroyed at the end of the || expression. Following their destruction, S(5) and S(6) are constructed in lexical order. Again they are not destroyed at the end of the full expression, but right at the end of the && expression. Consequently, the destruction of S(6) and S(5) is carried before that of S(2) and S(1).
CommaExpression: AssignExpressionCommaExpression,AssignExpression
The left operand of the , is evaluated, then the right operand is evaluated. In C, the result of a comma expression is the result of the right operand. In D, using the result of a comma expression isn't allowed. Consequently a comma expression is only useful when each operand has a side effect.
int x, y; // expression statement x = 1, y = 1; // evaluate a comma expression at the end of each loop iteration for (; y < 10; x++, y *= 2) writefln("%s, %s", x, y);
AssignExpression: ConditionalExpressionConditionalExpression=AssignExpressionConditionalExpression+=AssignExpressionConditionalExpression-=AssignExpressionConditionalExpression*=AssignExpressionConditionalExpression/=AssignExpressionConditionalExpression%=AssignExpressionConditionalExpression&=AssignExpressionConditionalExpression|=AssignExpressionConditionalExpression^=AssignExpressionConditionalExpression~=AssignExpressionConditionalExpression<<=AssignExpressionConditionalExpression>>=AssignExpressionConditionalExpression>>>=AssignExpressionConditionalExpression^^=AssignExpression
For all assign expressions, the left operand must be a modifiable lvalue. The type of the assign expression is the type of the left operand, and the result is the value of the left operand after assignment occurs. The resulting expression is a modifiable lvalue.
If the operator is = then it is simple assignment.
Otherwise, the right operand is implicitly converted to the type of the left operand, and assigned to it.
For arguments of built-in types, assignment operator expressions such as
a op= bare semantically equivalent to:
a = cast(typeof(a))(a op b)except that:
For user-defined types, assignment operator expressions are overloaded separately from the binary operators. Still the left operand must be an lvalue.
ConditionalExpression: OrOrExpressionOrOrExpression?Expression:ConditionalExpression
The first expression is converted to bool, and is evaluated.
If it is true, then the second expression is evaluated, and its result is the result of the conditional expression.
If it is false, then the third expression is evaluated, and its result is the result of the conditional expression.
If either the second or third expressions are of type void, then the resulting type is void. Otherwise, the second and third expressions are implicitly converted to a common type which becomes the result type of the conditional expression.
bool test; int a, b, c; ... test ? a = b : c = 2; // error (test ? a = b : c) = 2; // OK
This makes the intent clearer, because the first statement can easily be misread as the following code:
test ? a = b : (c = 2);
OrOrExpression: AndAndExpressionOrOrExpression||AndAndExpression
The result type of an OrOrExpression is bool, unless the right operand has type void, when the result is type void.
The OrOrExpression evaluates its left operand.
If the left operand, converted to type bool, evaluates to true, then the right operand is not evaluated. If the result type of the OrOrExpression is bool then the result of the expression is true.
If the left operand is false, then the right operand is evaluated. If the result type of the OrOrExpression is bool then the result of the expression is the right operand converted to type bool.
AndAndExpression: OrExpressionAndAndExpression&&OrExpression
The result type of an AndAndExpression is bool, unless the right operand has type void, when the result is type void.
The AndAndExpression evaluates its left operand.
If the left operand, converted to type bool, evaluates to false, then the right operand is not evaluated. If the result type of the AndAndExpression is bool then the result of the expression is false.
If the left operand is true, then the right operand is evaluated. If the result type of the AndAndExpression is bool then the result of the expression is the right operand converted to type bool.
Bit wise expressions perform a bitwise operation on their operands. Their operands must be integral types. First, the Usual Arithmetic Conversions are done. Then, the bitwise operation is done.
int x, a, b; x = a & 5 == b; // error x = a & 5 is b; // error x = a & 5 <= b; // error x = (a & 5) == b; // OK x = a & (5 == b); // OK
OrExpression: XorExpressionOrExpression|XorExpression
The operands are OR'd together.
XorExpression: AndExpressionXorExpression^AndExpression
The operands are XOR'd together.
AndExpression: CmpExpressionAndExpression&CmpExpression
The operands are AND'd together.
CmpExpression: EqualExpressionIdentityExpressionRelExpressionInExpressionShiftExpression
EqualExpression: ShiftExpression==ShiftExpressionShiftExpression!=ShiftExpression
Equality expressions compare the two operands for equality (==) or inequality (!=). The type of the result is bool.
Inequality is defined as the logical negation of equality.
assert(5 == 5L); assert(byte(4) == 4F); int i = 1, j = 1; assert(&i != &j); assert(&i != null); // elements of different types are comparable, even when different sizes int[] ia = ['A', 'B', 'C']; assert(ia == "ABC"); byte[] ba = [1, 2]; assert(ba == [1F, 2F]);
x.re == y.re && x.im == y.im
For class references, a == b is rewritten to .object.opEquals(a, b), which handles null. This is intended to compare the contents of two objects, however an appropriate opEquals method override must be defined for this to work. The default opEquals provided by the root Object class is equivalent to the is operator.
For struct objects, the expression (a == b) is rewritten as a.opEquals(b), or failing that, b.opEquals(a).
For both class references and struct objects, (a != b) is rewritten as !(a == b).
See opEquals for details.
For struct objects, equality means the result of the opEquals() member function. If an opEquals() is not provided, one will be generated. Equality is defined as the logical product of all equality results of the corresponding object fields.
struct S { int i = 4; string s = "four"; } S s; assert(s == S()); s.s = "foul"; assert(s != S());
If there are overlapping fields, which happens with unions, the default equality will compare each of the overlapping fields.
IdentityExpression: ShiftExpressionisShiftExpressionShiftExpression! isShiftExpression
The is operator compares for identity of expression values. To compare for nonidentity, use e1 !is e2. The type of the result is bool. The operands undergo the Usual Arithmetic Conversions to bring them to a common type before comparison.
For class / interface objects, identity is defined as the object references being identical. Class references can be efficiently compared against null using is. Note that interface objects need not have the same reference of the class they were cast from. To test whether an interface shares a class instance with another interface / class value, cast both operands to Object before comparing with is.
interface I { void g(); } interface I1 : I { void g1(); } interface I2 : I { void g2(); } interface J : I1, I2 { void h(); } class C : J { overridevoid g() { } overridevoid g1() { } overridevoid g2() { } overridevoid h() { } } void main() @safe { C c = new C; I i1 = cast(I1) c; I i2 = cast(I2) c; assert(i1 !is i2); // not identical assert(c !is i2); // not identical assert(cast(Object) i1 iscast(Object) i2); // identical }
For struct objects and floating point values, identity is defined as the bits in the operands being identical.
For static and dynamic arrays, identity of two arrays is given when both arrays refer to the same memory location and contain the same number of elements.
Object o; assert(o isnull); auto a = [1, 2]; assert(a is a[0..$]); assert(a !is a[0..1]); auto b = [1, 2]; assert(a !is b);
For other operand types, identity is defined as being the same as equality.
The identity operator is cannot be overloaded.
RelExpression: ShiftExpression<ShiftExpressionShiftExpression<=ShiftExpressionShiftExpression>ShiftExpressionShiftExpression>=ShiftExpression
First, the Usual Arithmetic Conversions are done on the operands. The result type of a relational expression is bool.
For static and dynamic arrays, the result of a CmpExpression is the result of the operator applied to the first non-equal element of the array. If two arrays compare equal, but are of different lengths, the shorter array compares as "less" than the longer array.
Integer comparisons happen when both operands are integral types.
| Operator | Relation |
|---|---|
| < | less |
| > | greater |
| <= | less or equal |
| >= | greater or equal |
| == | equal |
| != | not equal |
It is an error to have one operand be signed and the other unsigned for a <, <=, > or >= expression. Use casts to make both operands signed or both operands unsigned.
If one or both operands are floating point, then a floating point comparison is performed.
A CmpExpression can have NaN operands. If either or both operands is NaN, the floating point comparison operation returns as follows:
| Operator | Relation | Returns |
|---|---|---|
| < | less | false |
| > | greater | false |
| <= | less or equal | false |
| >= | greater or equal | false |
| == | equal | false |
| != | unordered, less, or greater | true |
For struct objects, a RelExpression performs a comparison which first evaluates a matching opCmp method call.
For class references, a RelExpression performs a comparison which first evaluates to an int which is either:
class C { overrideint opCmp(Object o) { assert(0); } } void main() { C c; //if (c < null) {} // compile-time error assert(c isnull); assert(c < new C); // C.opCmp is not called }
Secondly, for class and struct objects, the evaluated int is compared against zero using the given operator, which forms the result of the RelExpression. For more information, see opCmp.
InExpression: ShiftExpressioninShiftExpressionShiftExpression! inShiftExpression
A container such as an associative array can be tested to see if it contains a certain key:
int[string] foo; ... if ("hello"in foo) { // the string was found }
The result of an InExpression is a pointer for associative arrays. The pointer is null if the container has no matching key. If there is a match, the pointer points to a value associated with the key.
The !in expression is the logical negation of the in operation.
The in expression has the same precedence as the relational expressions <, <=, etc.
ShiftExpression: AddExpressionShiftExpression<<AddExpressionShiftExpression>>AddExpressionShiftExpression>>>AddExpression
The operands must be integral types, and undergo the Integer Promotions. The result type is the type of the left operand after the promotions. The result value is the result of shifting the bits by the right operand's value.
int c; int s = -3; auto y = c << s; // implementation defined value auto x = c << 33; // error, max shift count allowed is 31
AddExpression: MulExpressionAddExpression+MulExpressionAddExpression-MulExpressionAddExpression~MulExpression
In the cases of the Additive operations + and -:
If the operands are of integral types, they undergo the Usual Arithmetic Conversions, and then are brought to a common type using the Usual Arithmetic Conversions.
If both operands are of integral types and an overflow or underflow occurs in the computation, wrapping will happen. For example:
If either operand is a floating point type, the other is implicitly converted to floating point and they are brought to a common type via the Usual Arithmetic Conversions.
Add expressions for floating point operands are not associative.
If the first operand is a pointer, and the second is an integral type, the resulting type is the type of the first operand, and the resulting value is the pointer plus (or minus) the second operand multiplied by the size of the type pointed to by the first operand.
int[] a = [1,2,3]; int* p = a.ptr; assert(*p == 1); *(p + 2) = 4; // same as `p[2] = 4` assert(a[2] == 4);
IndexOperation can also be used with a pointer and has the same behaviour as adding an integer, then dereferencing the result.
If the second operand is a pointer, and the first is an integral type, and the operator is +, the operands are reversed and the pointer arithmetic just described is applied.
Producing a pointer through pointer arithmetic is not allowed in @safe code.
If both operands are pointers, and the operator is +, then it is illegal.
If both operands are pointers, and the operator is -, the pointers are subtracted and the result is divided by the size of the type pointed to by the operands. In this calculation the assumed size of void is one byte. It is an error if the pointers point to different types. The type of the result is ptrdiff_t.
int[] a = [1,2,3]; ptrdiff_t d = &a[2] - a.ptr; assert(d == 2);
In the case of the Additive operation ~:
A CatExpression concatenates a container's data with other data, producing a new container.
For a dynamic array, the other operand must either be another array or a single value that implicitly converts to the element type of the array. See Array Concatenation.
MulExpression: UnaryExpressionMulExpression*UnaryExpressionMulExpression/UnaryExpressionMulExpression%UnaryExpression
The operands must be arithmetic types. They undergo the Usual Arithmetic Conversions.
For integral operands, the *, /, and % correspond to multiply, divide, and modulus operations. For multiply, overflows are ignored and simply chopped to fit into the integral type.
For integral operands of the / and % operators, the quotient rounds towards zero and the remainder has the same sign as the dividend.
The following divide or modulus integral operands:
are illegal if encountered during Compile Time Execution.
For floating point operands, the * and / operations correspond to the IEEE 754 floating point equivalents. % is not the same as the IEEE 754 remainder. For example, 15.0 % 10.0 == 5.0, whereas for IEEE 754, remainder(15.0,10.0) == -5.0.
Mul expressions for floating point operands are not associative.
UnaryExpression: &UnaryExpression++UnaryExpression--UnaryExpression*UnaryExpression-UnaryExpression+UnaryExpression!UnaryExpressionComplementExpressionDeleteExpressionCastExpressionThrowExpressionPowExpression
| Operator | Description |
|---|---|
| & | Take memory address of an lvalue - see pointers |
| ++ | Increment before use - see order of evaluation |
| -- | Decrement before use |
| * | Dereference/indirection - typically for pointers |
| - | Negative |
| + | Positive |
| ! | Logical NOT |
The usual Integer Promotions are performed prior to unary - and + operations.
ComplementExpression: ~UnaryExpression
ComplementExpressions work on integral types (except bool). All the bits in the value are complemented. The usual Integer Promotions are performed prior to the complement operation.
DeleteExpression: deleteUnaryExpression
If the UnaryExpression is a class object reference, and there is a destructor for that class, the destructor is called for that object instance.
Next, if the UnaryExpression is a class object reference, or a pointer to a struct instance, and the class or struct has overloaded operator delete, then that operator delete is called for that class object instance or struct instance.
Otherwise, the garbage collector is called to immediately free the memory allocated for the class instance or struct instance.
If the UnaryExpression is a pointer or a dynamic array, the garbage collector is called to immediately release the memory.
The pointer, dynamic array, or reference is set to null after the delete is performed. Any attempt to reference the data after the deletion via another reference to it will result in undefined behavior.
If UnaryExpression is a variable allocated on the stack, the class destructor (if any) is called for that instance. The garbage collector is not called.
CastExpression: cast (Type)UnaryExpressionCastQual
A CastExpression converts the UnaryExpression to Type.
cast(foo) -p; // cast (-p) to type foo (foo) - p; // subtract p from foo
For situations where implicit conversions on basic types cannot be performed, the type system may be forced to accept the reinterpretation of a memory region by using a cast.
An example of such a scenario is represented by trying to store a wider type into a narrower one:
int a; byte b = a; // cannot implicitly convert expression a of type int to byte
When casting a source type that is wider than the destination type, the value is truncated to the destination size.
int a = 64389; // 00000000 00000000 11111011 10000101 byte b = cast(byte) a; // 10000101 ubyte c = cast(ubyte) a; // 10000101 short d = cast(short) a; // 11111011 10000101 ushort e = cast(ushort) a; // 11111011 10000101 writeln(b); writeln(c); writeln(d); writeln(e);
For integral types casting from a narrower type to a wider type is done by performing sign extension.
ubyte a = 133; // 10000101 byte b = a; // 10000101 writeln(a); writeln(b); ushort c = a; // 00000000 10000101 short d = b; // 11111111 10000101 writeln(c); writeln(d);
See also: Casting Integers.
Any casting of a class reference to a derived class reference is done with a runtime check to make sure it really is a downcast. null is the result if it isn't.
class A {} class B : A {} void main() { A a = new A; //B b = a; // error, need cast B b = cast(B) a; // b is null if a is not a B assert(b isnull); a = b; // no cast needed a = cast(A) b; // no runtime check needed for upcast assert(a is b); }
In order to determine if an object o is an instance of a class B use a cast:
if (cast(B) o) { // o is an instance of B } else { // o is not an instance of B }
Casting a pointer type to and from a class type is done as a type paint (i.e. a reinterpret cast).
Casting a pointer variable to another pointer type modifies the value that will be obtained as a result of dereferencing, along with the number of bytes on which pointer arithmetic is performed.
int val = 25185; // 00000000 00000000 01100010 01100001 char *ch = cast(char*)(&val); writeln(*ch); // a writeln(cast(int)(*ch)); // 97 writeln(*(ch + 1)); // b writeln(cast(int)(*(ch + 1))); // 98
Similarly, when casting a dynamically allocated array to a type of smaller size, the bytes of the initial array will be divided and regrouped according to the new dimension.
import core.stdc.stdlib; int *p = cast(int*) malloc(5 * int.sizeof); for (int i = 0; i < 5; i++) { p[i] = i + 'a'; } // p = [97, 98, 99, 100, 101] char* c = cast(char*) p; // c = [97, 0, 0, 0, 98, 0, 0, 0, 99 ...] for (int i = 0; i < 5 * int.sizeof; i++) { writeln(c[i]); }
When casting a pointer of type A to a pointer of type B and type B is wider than type A, attempts at accessing the memory exceeding the size of A will result in undefined behaviour.
char c = 'a'; int *p = cast(int*) (&c); writeln(*p);
It is also possible to cast pointers to basic data types. A common practice could be to cast the pointer to an int value and then print its address:
import core.stdc.stdlib; int *p = cast(int*) malloc(int.sizeof); int a = cast(int) p; writeln(a);
T[] a; ... cast(U[]) a Casting a non-literal dynamic array a to another dynamic array type U[] is allowed only when the result will contain every byte of data that was referenced by a. This is enforced with a runtime check that the byte length of a's elements is divisible by U.sizeof. If there is a remainder, a runtime error is generated. The cast is done as a type paint, and the resulting array's length is set to (a.length * T.sizeof) / U.sizeof.
byte[] a = [1,2,3]; //auto b = cast(int[])a; // runtime error: array cast misalignment int[] c = [1, 2, 3]; auto d = cast(byte[])c; // ok // prints: // [1, 0, 0, 0, 2, 0, 0, 0, 3, 0, 0, 0] writeln(d);
See also:Casting array literals.
A slice of statically known length can be cast to a static array type when the byte counts of their respective data match.
void f(int[] b) { char[4] a; staticassert(!__traits(compiles, a = cast(char[4]) b)); // unknown length staticassert(!__traits(compiles, a = cast(char[4]) b[0..2])); // too many bytes a = cast(char[4]) b[0..1]; // OK const i = 1; a = cast(char[4]) b[i..2]; // OK }
See also:Slice conversion to static array.
Casting a static array to another static array is done only if the array lengths multiplied by the element sizes match; a mismatch is illegal. The cast is done as a type paint (aka a reinterpret cast). The contents of the array are not changed.
byte[16] b = 3; // set each element to 3 assert(b[0] == 0x03); int[4] ia = cast(int[4]) b; // print elements as hex foreach (i; ia) writefln("%x", i); /* prints: 3030303 3030303 3030303 3030303 */
Casting an integer to a smaller integral will truncate the value towards the least significant bits. If the target type is signed and the most significant bit is set after truncation, that bit will be lost from the value and the sign bit will be set.
uint a = 260; auto b = cast(ubyte) a; assert(b == 4); // truncated like 260 & 0xff int c = 128; assert(cast(byte)c == -128); // reinterpreted
Converting between signed and unsigned types will reinterpret the value if the destination type cannot represent the source value.
short c = -1; ushort d = c; assert(d == ushort.max); assert(uint(c) == uint.max); ubyte e = 255; byte f = e; assert(f == -1); // reinterpreted assert(short(e) == 255); // no change
Casting a floating point literal from one type to another changes its type, but internally it is retained at full precision for the purposes of constant folding.
void test() { real a = 3.40483L; real b; b = 3.40483; // literal is not truncated to double precision assert(a == b); assert(a == 3.40483); assert(a == 3.40483L); assert(a == 3.40483F); double d = 3.40483; // truncate literal when assigned to variable assert(d != a); // so it is no longer the same constdouble x = 3.40483; // assignment to const is not assert(x == a); // truncated if the initializer is visible }
Casting a floating point value to an integral type is the equivalent of converting to an integer using truncation. If the floating point value is outside the range of the integral type, the cast will produce an invalid result (this is also the case in C, C++).
void main() { int a = cast(int) 0.8f; assert(a == 0); long b = cast(long) 1.5; assert(b == 1L); long c = cast(long) -1.5; assert(c == -1); // if the float overflows, the cast returns the integer value of // 80000000_00000000H (64-bit operand) or 80000000H (32-bit operand) long d = cast(long) float.max; assert(d == long.min); int e = cast(int) (1234.5 + int.max); assert(e == int.min); // for types represented on 16 or 8 bits, the result is the same as // 32-bit types, but the most significant bits are ignored short f = cast(short) float.max; assert(f == 0); }
An expression e can be cast to a struct type S:
struct S { int i; } struct R { short[2] a; } S s = cast(S) 5; // same as S(5) assert(s.i == 5); staticassert(!__traits(compiles, cast(S) long.max)); // S(long.max) is invalid R r = R([1, 2]); s = cast(S) r; // reinterpret r assert(s.i == 0x00020001); byte[4] a = [1, 0, 2, 0]; assert(r == cast(R) a); // reinterpret a
A struct instance can be cast to a static array type when their .sizeof properties each give the same result.
struct S { short a, b, c; } S s = S(1, 2, 3); staticassert(!__traits(compiles, cast(short[2]) s)); // size mismatch short[3] x = cast(short[3]) s; assert(x.tupleof == s.tupleof); auto y = cast(byte[6]) s; assert(y == [1, 0, 2, 0, 3, 0]);
CastQual: cast (TypeCtorsopt)UnaryExpression
A CastQual replaces the qualifiers in the type of the UnaryExpression:
sharedint x; staticassert(is(typeof(cast(const)x) == constint));
Casting with no type or qualifiers removes any top level const, immutable, shared or inout type modifiers from the type of the UnaryExpression.
sharedint x; staticassert(is(typeof(cast()x) == int));
Casting an expression to void type is allowed to mark that the result is unused. On ExpressionStatement, it could be used properly to avoid a "has no effect" error.
void foo(lazyvoid exp) {} void main() { foo(10); // NG - expression '10' has no effect foo(cast(void)10); // OK }
ThrowExpression: throwAssignExpression
AssignExpression is evaluated and must yield a reference to a Throwable or a class derived from Throwable. The reference is thrown as an exception, interrupting the current control flow to continue at a suitable catch clause of a try-statement. This process will execute any applicable scope (exit) / scope (failure) passed since entering the corresponding try block.
thrownew Exception("message");
The Throwable must not be a qualified as immutable, const, inout or shared. The runtime may modify a thrown object (e.g. to contain a stack trace) which would violate const or immutable objects.
A ThrowExpression may be nested in another expression:
void foo(intfunction() f) {} void main() { foo(() => thrownew Exception()); }
The type of a ThrowExpression is noreturn.
PowExpression: PostfixExpressionPostfixExpression^^UnaryExpression
PowExpression raises its left operand to the power of its right operand.
PostfixExpression: PrimaryExpressionPostfixExpression.IdentifierPostfixExpression.TemplateInstancePostfixExpression.NewExpressionPostfixExpression++PostfixExpression--PostfixExpression(NamedArgumentListopt)TypeCtorsoptBasicType(NamedArgumentListopt)PostfixExpressionIndexOperationPostfixExpressionSliceOperation
| Operation | Description |
|---|---|
| .Identifier | Either:
|
| .NewExpression | Instantiate a nested class |
| ++ | Increment after use - see order of evaluation |
| -- | Decrement after use |
| (args) | Either:
|
| IndexOperation | Select a single element |
| SliceOperation | Select a series of elements |
ArgumentList: AssignExpressionAssignExpression,AssignExpression,ArgumentListNamedArgumentList: NamedArgumentNamedArgument,NamedArgument,NamedArgumentListNamedArgument: Identifier:AssignExpressionAssignExpression
A callable expression can precede a list of named arguments in parentheses. The following expressions can be called:
void f(int, int); void g() { f(5, 6); (&f)(5, 6); }
Arguments in a NamedArgumentList are matched to function parameters as follows:
A type can precede a list of arguments. See:
IndexOperation: [ArgumentList]
The base PostfixExpression is evaluated. The special variable $ is declared and set to be the number of elements in the base PostfixExpression (when available). A new declaration scope is created for the evaluation of the ArgumentList and $ appears in that scope only.
If the PostfixExpression is an expression of static or dynamic array type, the result of the indexing is an lvalue of the ith element in the array, where i is an integer evaluated from ArgumentList. If PostfixExpression is a pointer p, the result is *(p + i) (see Pointer Arithmetic).
If the base PostfixExpression is a ValueSeq then the ArgumentList must consist of only one argument, and that must be statically evaluatable to an integral constant. That integral constant n then selects the nth expression in the ValueSeq, which is the result of the IndexOperation. It is an error if n is out of bounds of the ValueSeq.
The index operator can be overloaded. Using multiple indices in ArgumentList is only supported for operator overloading.
SliceOperation: [ ][Slice][Slice,]Slice: AssignExpressionAssignExpression,SliceAssignExpression..AssignExpressionAssignExpression..AssignExpression,Slice
The base PostfixExpression is evaluated. The special variable $ is declared and set to be the number of elements in the PostfixExpression (when available). A new declaration scope is created for the evaluation of the AssignExpression..AssignExpression and $ appears in that scope only.
If the base PostfixExpression is a static or dynamic array a, the result of the slice is a dynamic array referencing elements a[i] to a[j-1] inclusive, where i and j are integers evaluated from the first and second AssignExpression respectively.
If the base PostfixExpression is a pointer p, the result will be a dynamic array referencing elements from p[i] to p[j-1] inclusive, where i and j are integers evaluated from the first and second AssignExpression respectively.
If the base PostfixExpression is a ValueSeq, then the result of the slice is a new ValueSeq formed from the upper and lower bounds, which must statically evaluate to integral constants. It is an error if those bounds are out of range.
The first AssignExpression is taken to be the inclusive lower bound of the slice, and the second AssignExpression is the exclusive upper bound. The result of the expression is a slice of the elements in PostfixExpression.
If the [ ] form is used, the slice is of all the elements in the base PostfixExpression. The base expression cannot be a pointer.
The slice operator can be overloaded. Using more than one Slice is only supported for operator overloading.
A SliceOperation is not a modifiable lvalue.
If the slice bounds can be known at compile time, the slice expression may be implicitly convertible to a static array lvalue. For example:
arr[a .. b] // typed T[] If both a and b are integers (which may be constant-folded), the slice expression can be converted to a static array of type T[b - a].
void f(int[2] sa) {} int[] arr = [1, 2, 3]; void test() { //f(arr); // error, can't convert f(arr[1 .. 3]); // OK //f(arr[0 .. 3]); // error int[2] g() { return arr[0 .. 2]; } }
void bar(refint[2] a) { assert(a == [2, 3]); a = [4, 5]; } void main() { int[] arr = [1, 2, 3]; // slicing an lvalue gives an lvalue bar(arr[1 .. 3]); assert(arr == [1, 4, 5]); }
PrimaryExpression: Identifier.IdentifierTemplateInstance.TemplateInstance$LiteralExpressionAssertExpressionMixinExpressionImportExpressionNewExpressionFundamentalType.IdentifierTypeCtoropt(Type) .Identifier(Type) .TemplateInstanceFundamentalType(NamedArgumentListopt)TypeCtoropt(Type)(NamedArgumentListopt)TypeofTypeidExpressionIsExpression(Expression)SpecialKeywordRvalueExpressionTraitsExpressionLiteralExpression: thissupernulltruefalseIntegerLiteralFloatLiteralCharacterLiteralStringLiteralInterpolationExpressionSequenceArrayLiteralAssocArrayLiteralFunctionLiteral
| Expression | Description |
|---|---|
| .Identifier | Module Scope Operator |
| $ | Number of elements in an object being indexed/sliced. |
| (Type).Identifier | Access a type property or a static member of a type. |
| FundamentalType(arg) | Uniform construction of scalar type with optional argument. |
| (Type)(args) | Construct a type with optional arguments. |
| (Expression) | Evaluate an expression - useful as a subexpression. |
Within a constructor or non-static member function, this resolves to a reference to the object for which the function was called.
typeof(this) is valid anywhere inside an aggregate type definition. If a class member function is called with an explicit reference to typeof(this), a non-virtual call is made:
class A { char get() { return 'A'; } char foo() { returntypeof(this).get(); } // calls `A.get` char bar() { returnthis.get(); } // dynamic, same as just `get()` } class B : A { overridechar get() { return 'B'; } } void main() { B b = new B(); assert(b.foo() == 'A'); assert(b.bar() == 'B'); }
Assignment to this is not allowed for classes.
See also:
super is identical to this, except that it is cast to this's base class. It is an error if there is no base class. (The only extern(D) class without a base class is Object, however, note that extern(C++) classes have no base class unless specified.) If a member function is called with an explicit reference to super, a non-virtual call is made.
Assignment to super is not allowed.
See also: Base Class Construction.
null represents the null value for pointers, pointers to functions, delegates, dynamic arrays, associative arrays, and class objects. If it has not already been cast to a type, it is given the singular type typeof(null) and it is an exact conversion to convert it to the null value for pointers, pointers to functions, delegates, etc. After it is cast to a type, such conversions are implicit, but no longer exact.
See StringLiteral grammar.
String literals are read-only. A string literal without a StringPostfix can implicitly convert to any of the following types, which have equal weight:
| immutable(char)* |
| immutable(wchar)* |
| immutable(dchar)* |
| immutable(char)[] |
| immutable(wchar)[] |
| immutable(dchar)[] |
By default, a string literal is typed as a dynamic array, but the element count is known at compile time. So all string literals can be implicitly converted to an immutable static array:
void foo(char[2] a) { assert(a[0] == 'b'); } void bar(refconstchar[2] a) { assert(a == "bc"); } void main() { foo("bc"); foo("b"); // OK //foo("bcd"); // error, too many chars bar("bc"); // OK, same length //bar("b"); // error, lengths must match }
A string literal converts to a static array rvalue of the same or longer length. Any extra elements are padded with zeros. A string literal can also convert to a static array lvalue of the same length.
String literals have a '\0' appended to them, which makes them easy to pass to C or C++ functions expecting a null-terminated const char* string. The '\0' is not included in the .length property of the string literal.
Concatenation of string literals requires the use of the ~ operator, and is resolved at compile time. C style implicit concatenation without an intervening operator is error prone and not supported in D.
Because hex string literals contain binary data not limited to textual data, they allow additional conversions over other string literals.
A hex string literal implicitly converts to a constant byte[] or ubyte[].
immutableubyte[] b = x"3F 80 00 00"; constbyte[] c = x"3F 80 00 00";
A hex string literal can be explicitly cast to an array of integers with a larger size than 1. A big endian byte order in the hex string will be assumed.
staticimmutableuint[] data = cast(immutableuint[]) x"AABBCCDD"; staticassert(data[0] == 0xAABBCCDD);
This requires the length of the hex string to be a multiple of the array element's size in bytes.
static e = cast(immutableushort[]) x"AA BB CC"; // Error, length of 3 bytes is not a multiple of 2, the size of a `ushort`
When a hex string literal gets constant folded, the result is no longer considered a hex string literal
staticimmutablebyte[] b = x"AA" ~ "G"; // Error: cannot convert `string` to `immutable byte[]`
ArrayLiteral: [ArgumentListopt]
An array literal is a comma-separated list of expressions between square brackets [ and ]. The expressions form the elements of a dynamic array. The length of the array is the number of elements.
The element type of the array is inferred as the common type of all the elements, and each expression is implicitly converted to that type. When there is an expected array type, the elements of the literal will be implicitly converted to the expected element type.
auto a1 = [1, 2, 3]; // type is int[], with elements 1, 2 and 3 auto a2 = [1u, 2, 3]; // type is uint[], with elements 1u, 2u, and 3u byte[] a3 = [1, 2, 3]; // OK byte[] a4 = [128]; // error
int[2] sa = [1, 2]; // OK int[2] sb = [1]; // error
If any ArrayMemberInitialization is a ValueSeq, then the elements of the ValueSeq are inserted as expressions in place of the sequence.
Escaping array literals are always allocated on the memory managed heap. Thus, they can be returned safely from functions:
int[] foo() { return [1, 2, 3]; }
An array literal is not GC allocated if:
void f(scopeint[] a, int[2] sa) @nogc { sa = [7, 8]; } void g(int[] b) @nogc; // `b` is not scope, so may escape void main() @nogc { int[3] sa = [1, 2, 3]; f([1, 2], [3, 4]); //scope int[] a = [5, 6]; // requires `-preview=dip1000` //g([1, 2]); // error, array literal heap allocated assert([1, 2] < [3, 2]); assert([1, 2][1] == 2); foreach (e; [4, 2, 9]) assert(e > 0); }
When array literals are cast to another array type, each element of the array is cast to the new element type. When arrays that are not literals are cast, the array is reinterpreted as the new type, and the length is recomputed:
// cast array literal constubyte[] ct = cast(ubyte[]) [257, 257]; // this is equivalent to: // const ubyte[] ct = [cast(ubyte) 257, cast(ubyte) 257]; writeln(ct); // writes [1, 1] // cast other array expression // --> normal behavior of CastExpression byte[] arr = [1, 1]; short[] rt = cast(short[]) arr; writeln(rt); // writes [257]
AssocArrayLiteral: [KeyValuePairs]KeyValuePairs: KeyValuePairKeyValuePair,KeyValuePairsKeyValuePair: KeyExpression:ValueExpressionKeyExpression: AssignExpressionValueExpression: AssignExpression
Associative array literals are a comma-separated list of key:value pairs between square brackets [ and ]. The list cannot be empty. The common type of the all keys is taken to be the key type of the associative array, and all keys are implicitly converted to that type. The common type of the all values is taken to be the value type of the associative array, and all values are implicitly converted to that type. An AssocArrayLiteral cannot be used to statically initialize anything.
[21u: "he", 38: "ho", 2: "hi"]; // type is string[uint], // with keys 21u, 38u and 2u // and values "he", "ho", and "hi"
If any of the keys or values in the KeyValuePairs are a ValueSeq, then the elements of the ValueSeq are inserted as arguments in place of the sequence.
Associative array initializers may contain duplicate keys, however, in that case, the last KeyValuePair lexicographically encountered is stored.
auto aa = [21: "he", 38: "ho", 2: "hi", 2:"bye"]; assert(aa[2] == "bye")
FunctionLiteral: functionRefOrAutoRefoptBasicTypeWithSuffixesoptParameterWithAttributesoptFunctionLiteralBodydelegateRefOrAutoRefoptBasicTypeWithSuffixesoptParameterWithMemberAttributesoptFunctionLiteralBodyRefOrAutoRefoptParameterWithMemberAttributesFunctionLiteralBodyBlockStatementIdentifier=>AssignExpressionBasicTypeWithSuffixes: BasicTypeTypeSuffixesoptParameterWithAttributes: ParametersFunctionAttributesoptParameterWithMemberAttributes: ParametersMemberFunctionAttributesoptFunctionLiteralBody: =>AssignExpressionSpecifiedFunctionBodyRefOrAutoRef: refauto ref
FunctionLiterals enable embedding anonymous functions and anonymous delegates directly into expressions. Short function literals are known as lambdas.
For example:
intfunction(char c) fp; // declare pointer to a function void test() { staticint foo(char c) { return 6; } fp = &foo; }is exactly equivalent to:
intfunction(char c) fp; void test() { fp = functionint(char c) { return 6; }; }
A delegate is necessary if the FunctionLiteralBody accesses any non-static local variables in enclosing functions.
int abc(intdelegate(int i)); void test() { int b = 3; int foo(int c) { return 6 + b; } abc(&foo); }is exactly equivalent to:
int abc(intdelegate(int i)); void test() { int b = 3; abc( delegateint(int c) { return 6 + b; } ); }
The use of ref declares that the return value is returned by reference:
void main() { int x; auto dg = delegaterefint() { return x; }; dg() = 3; assert(x == 3); }
If a literal omits function or delegate and there's no expected type from the context, then it is inferred to be a delegate if it accesses a variable in an enclosing function, otherwise it is a function pointer.
void test() { int b = 3; auto fp = (uint c) { return c * 2; }; // inferred as function pointer auto dg = (int c) { return 6 + b; }; // inferred as delegate staticassert(!is(typeof(fp) == delegate)); staticassert(is(typeof(dg) == delegate)); }
If a delegate is expected, the literal will be inferred as a delegate even if it accesses no variables from an enclosing function:
void abc(intdelegate(int i)) {} void def(uintfunction(uint s)) {} void test() { int b = 3; abc( (int c) { return 6 + b; } ); // inferred as delegate abc( (int c) { return c * 2; } ); // inferred as delegate def( (uint c) { return c * 2; } ); // inferred as function //def( (uint c) { return c * b; } ); // error! // Because the FunctionLiteral accesses b, its type // is inferred as delegate. But def cannot accept a delegate argument. }
If the type of a function literal can be uniquely determined from its context, parameter type inference is possible.
void foo(intfunction(int) fp); void test() { intfunction(int) fp = (n) { return n * 2; }; // The type of parameter n is inferred as int. foo((n) { return n * 2; }); // The type of parameter n is inferred as int. }
auto fp = (i) { return 1; }; // error, cannot infer type of `i`
Function literals can be aliased. Aliasing a function literal with unspecified parameter types produces a function template with type parameters for each unspecified parameter type of the literal. Type inference for the literal is then done when the template is instantiated.
alias fpt = (i) { return i; }; // ok, infer type of `i` when used //auto fpt(T)(T i) { return i; } // equivalent auto v = fpt(4); // `i` is inferred as int auto d = fpt(10.3); // `i` is inferred as double alias fp = fpt!float; auto f = fp(0); // f is a float
The return type of the FunctionLiteral can be inferred from either the AssignExpression, or any ReturnStatements in the BlockStatement. If there is a different expected type from the context, and the initial inferred return type implicitly converts to the expected type, then the return type is inferred as the expected type.
auto fi = (int i) { return i; }; staticassert(is(typeof(fi(5)) == int)); longfunction(int) fl = (int i) { return i; }; staticassert(is(typeof(fl(5)) == long));
Parameters can be omitted completely for a function literal when there is a BlockStatement function body.
auto f = { writeln("hi"); }; // OK, f has type `void function()` f(); { writeln("hi"); }(); // error () { writeln("hi"); }(); // OK
void loop(int n, voiddelegate() statement) { foreach (_; 0 .. n) { statement(); } } void main() { int n = 0; loop(5, { n += 1; }); assert(n == 5); }
The syntax => AssignExpression is equivalent to { return AssignExpression; }.
void main() { auto i = 3; auto twice = function (int x) => x * 2; assert(twice(i) == 6); auto square = delegate () => i * i; assert(square() == 9); auto n = 5; auto mul_n = (int x) => x * n; assert(mul_n(i) == 15); }
The syntax Identifier => AssignExpression is equivalent to (Identifier) { return AssignExpression; }.
// the following two declarations are equivalent alias fp = i => 1; alias fp = (i) { return 1; };
int motor(alias fp)(int i) { return fp(i) + 1; } int engine() { return motor!(i => i * 2)(6); // returns 13 }
The implicit conversions of built-in scalar types can be explicitly represented by using function call syntax. For example:
auto a = short(1); // implicitly convert an integer literal '1' to short auto b = double(a); // implicitly convert a short variable 'a' to double auto c = byte(128); // error, 128 cannot be represented in a byte
If the argument is omitted, it means default construction of the scalar type:
auto a = ushort(); // same as: ushort.init auto b = wchar(); // same as: wchar.init
The argument may not be given a name:
auto a = short(x: 1); // Error
See also: Usual Arithmetic Conversions.
AssertExpression: assert (AssertArguments)AssertArguments: AssignExpressionAssignExpression,AssignExpression,AssignExpressionAssignExpression,AssignExpression,
The first AssignExpression is evaluated and converted to a boolean value. If the value is not true, an Assert Failure has occurred and the program enters an Invalid State.
int i = fun(); assert(i > 0);
AssertExpression has different semantics if it is in a unittest or in contract.
If the first AssignExpression is a reference to a class instance for which a class Invariant exists, the class Invariant must hold.
If the first AssignExpression is a pointer to a struct instance for which a struct Invariant exists, the struct Invariant must hold.
The type of an AssertExpression is void.
auto x = 4; assert(x < 3);When in use, the above will throw an AssertError with a message 4 >= 3.
If the first AssignExpression consists entirely of compile time constants, and evaluates to false, it is a special case - it signifies that subsequent statements are unreachable code. Compile Time Function Execution (CTFE) is not attempted.
This allows the compiler to suppress an error when there is a missing return statement:
int f(int x) { if (x > 0) { return 5 / x; } assert(0); // no need to use a dummy return statement here }
The implementation may handle the case of the first AssignExpression evaluating to false at compile time differently - even when other asserts are ignored, it may still generate a HLT instruction or equivalent.
See also: static assert.
The second AssignExpression, if present, must be implicitly convertible to type const(char)[]. When present, the implementation may evaluate it and print the resulting message upon assert failure:
void main() { assert(0, "an" ~ " error message"); }
When compiled and run, typically it will produce the message:
core.exception.AssertError@test.d(3) an error message
MixinExpression: mixin (ArgumentList)
Each AssignExpression in the ArgumentList is evaluated at compile time, and the result must be representable as a string. The resulting strings are concatenated to form a string. The text contents of the string must be compilable as a valid Expression, and is compiled as such.
int foo(int x) { returnmixin("x +", 1) * 7; // same as ((x + 1) * 7) }
ImportExpression: import (AssignExpression)
The AssignExpression must evaluate at compile time to a constant string. The text contents of the string are interpreted as a file name. The file is read, and the exact contents of the file become a hex string literal.
Implementations may restrict the file name in order to avoid directory traversal security vulnerabilities. A possible restriction might be to disallow any path components in the file name.
Note that by default an import expression will not compile unless one or more paths are passed via the -J switch. This tells the compiler where it should look for the files to import. This is a security feature.
void foo() { // Prints contents of file foo.txt writeln(import("foo.txt")); }
NewExpression: newTypenewType[AssignExpression]newType(NamedArgumentListopt)NewAnonClassExpression
NewExpressions allocate memory on the garbage collected heap by default.
new T constructs an instance of type T and default-initializes it. The result's type is:
int* i = newint; assert(*i == 0); // int.init Object o = new Object; //int[] a = new int[]; // error, need length argument
The Type(NamedArgumentList) form allows passing either a single initializer of the same type, or multiple arguments for more complex types:
int* i = newint(5); assert(*i == 5); Exception e = new Exception("info"); assert(e.msg == "info"); int[] a = newint[](2); assert(a.length == 2); a = newint[2]; // same, see below
The Type[AssignExpression] form allocates a dynamic array with length equal to AssignExpression. It is preferred to use the Type(NamedArgumentList) form when allocating dynamic arrays instead, as it is more general.
The result is a unique expression which can implicitly convert to other qualifiers:
immutable o = new Object;
If a NewExpression is used with a class type as an initializer for a function local variable with scope storage class, then the instance is allocated on the stack.
new can also be used to allocate a nested class.
To allocate multidimensional arrays, the declaration reads in the same order as the prefix array declaration order.
char[][] foo; // dynamic array of strings ... foo = newchar[][30]; // allocate array of 30 strings
The above allocation can also be written as:
foo = newchar[][](30); // allocate array of 30 strings
To allocate the nested arrays, multiple arguments can be used:
int[][][] bar; bar = newint[][][](5, 20, 30); assert(bar.length == 5); assert(bar[0].length == 20); assert(bar[0][0].length == 30);
bar = newint[][][5]; foreach (ref a; bar) { a = newint[][20]; foreach (ref b; a) { b = newint[30]; } }
TypeidExpression: typeid (Type)typeid (Expression)
If Type, returns an instance of class TypeInfo corresponding to Type.
If Expression, returns an instance of class TypeInfo corresponding to the type of the Expression. If the type is a class, it returns the TypeInfo of the dynamic type (i.e. the most derived type). The Expression is always executed.
class A { } class B : A { } void main() { import std.stdio; writeln(typeid(int)); // int uint i; writeln(typeid(i++)); // uint writeln(i); // 1 A a = new B(); writeln(typeid(a)); // B writeln(typeid(typeof(a))); // A }
IsExpression: is (Type)is (Type:TypeSpecialization)is (Type==TypeSpecialization)is (Type:TypeSpecialization,TemplateParameterList)is (Type==TypeSpecialization,TemplateParameterList)is (TypeIdentifier)is (TypeIdentifier:TypeSpecialization)is (TypeIdentifier==TypeSpecialization)is (TypeIdentifier:TypeSpecialization,TemplateParameterList)is (TypeIdentifier==TypeSpecialization,TemplateParameterList)TypeSpecialization: TypeTypeCtorstructunionclassinterfaceenum__vectorfunctiondelegatesuperreturn__parametersmodulepackage
An IsExpression is evaluated at compile time and is used to check if an expression is a valid type. In addition, there are forms which can also:
The result of an IsExpression is a boolean which is true if the condition is satisfied and false if not.
Type is the type being tested. It must be syntactically correct, but it need not be semantically correct. If it is not semantically correct, the condition is not satisfied.
TypeSpecialization is the type that Type is being pattern matched against.
IsExpressions may be used in conjunction with typeof to check whether an expression type checks correctly. For example, is(typeof(foo)) will return true if foo has a valid type.
The condition is satisfied if Type is semantically correct. Type must be syntactically correct regardless.
pragma(msg, is(5)); // error pragma(msg, is([][])); // error
int i; staticassert(is(int)); staticassert(is(typeof(i))); // same staticassert(!is(Undefined)); staticassert(!is(typeof(int))); // int is not an expression staticassert(!is(i)); // i is a value alias Func = int(int); // function type staticassert(is(Func)); staticassert(!is(Func[])); // fails as an array of functions is not allowed
The condition is satisfied if Type is semantically correct and it is the same as or can be implicitly converted to TypeSpecialization. TypeSpecialization is only allowed to be a Type.
alias Bar = short; staticassert(is(Bar : int)); // short implicitly converts to int staticassert(!is(Bar : string));
If TypeSpecialization is a type, the condition is satisfied if Type is semantically correct and is the same type as TypeSpecialization.
alias Bar = short; staticassert(is(Bar == short)); staticassert(!is(Bar == int));
If TypeSpecialization is a TypeCtor then the condition is satisfied if Type is of that TypeCtor:
staticassert(is(constint == const)); staticassert(is(constint[] == const)); staticassert(!is(const(int)[] == const)); // head is mutable staticassert(!is(immutableint == const));
If TypeSpecialization is one of structunionclassinterfaceenum__vectorfunctiondelegatemodulepackage then the condition is satisfied if Type is one of those.
staticassert(is(Object == class)); staticassert(is(ModuleInfo == struct)); staticassert(!is(int == class));
The module and package forms are satisfied when Type is a symbol, not a type, unlike the other forms. The isModule and isPackage__traits should be used instead. Package modules are considered to be both packages and modules.
TypeSpecialization can also be one of these keywords:
| keyword | condition |
|---|---|
| super | true if Type is a class or interface |
| return | true if Type is a function, delegate or function pointer |
| __parameters | true if Type is a function, delegate or function pointer |
class C {} staticassert(is(C == super)); void foo(int i); staticassert(!is(foo == return)); staticassert(is(typeof(foo) == return)); staticassert(is(typeof(foo) == __parameters));
See also:Traits.
Identifier is declared to be an alias of the resulting type if the condition is satisfied. The Identifier forms can only be used if the IsExpression appears in a StaticIfCondition or the first argument of a StaticAssert.
The condition is satisfied if Type is semantically correct. If so, Identifier is declared to be an alias of Type.
struct S { int i, j; } staticassert(is(typeof(S.i) T) && T.sizeof == 4);
alias Bar = short; void foo() { staticif (is(Bar T)) alias S = T; elsealias S = long; pragma(msg, S); // short // if T was defined, it remains in scope if (is(T)) pragma(msg, T); // short //if (is(Bar U)) {} // error, cannot declare U here }
If TypeSpecialization is a type, the condition is satisfied if Type is semantically correct and it is the same as or can be implicitly converted to TypeSpecialization. Identifier is declared to be an alias of the TypeSpecialization.
alias Bar = int; staticif (is(Bar T : int)) alias S = T; elsealias S = long; staticassert(is(S == int));
If TypeSpecialization is a type pattern involving Identifier, type deduction of Identifier is attempted based on either Type or a type that it implicitly converts to. The condition is only satisfied if the type pattern is matched.
struct S { long* i; alias i this; // S converts to long* } staticif (is(S U : U*)) // S is matched against the pattern U* { U u; } staticassert(is(U == long));
The way the type of Identifier is determined is analogous to the way template parameter types are determined by TemplateTypeParameterSpecialization.
If TypeSpecialization is a type, the condition is satisfied if Type is semantically correct and is the same type as TypeSpecialization. Identifier is declared to be an alias of the TypeSpecialization.
const x = 5; staticif (is(typeof(x) T == constint)) // satisfied, T is now defined alias S = T; staticassert(is(T)); // T is in scope pragma(msg, T); // const int
If TypeSpecialization is a type pattern involving Identifier, type deduction of Identifier is attempted based on Type. The condition is only satisfied if the type pattern is matched.
alias Foo = long*; staticif (is(Foo U == U*)) // Foo is matched against the pattern U* { U u; } staticassert(is(U == long));
If TypeSpecialization is a valid keyword for the is(Type == Keyword) form, the condition is satisfied in the same manner. Identifier is set as follows:
| keyword | alias type for Identifier |
|---|---|
| struct | Type |
| union | Type |
| class | Type |
| interface | Type |
| super | TypeSeq of base classes and interfaces |
| enum | the base type of the enum |
| __vector | the static array type of the vector |
| function | TypeSeq of the function parameter types. For C- and D-style variadic functions, only the non-variadic parameters are included. For typesafe variadic functions, the ... is ignored. |
| delegate | the function type of the delegate |
| return | the return type of the function, delegate, or function pointer |
| __parameters | the parameter sequence of a function, delegate, or function pointer. This includes the parameter types, names, and default values. |
| const | Type |
| immutable | Type |
| inout | Type |
| shared | Type |
| module | the module |
| package | the package |
enum E : byte { Emember } staticif (is(E V == enum)) // satisfied, E is an enum V v; // v is declared to be a byte staticassert(is(V == byte));
is (Type:TypeSpecialization,TemplateParameterList)is (Type==TypeSpecialization,TemplateParameterList)is (TypeIdentifier:TypeSpecialization,TemplateParameterList)is (TypeIdentifier==TypeSpecialization,TemplateParameterList)
More complex types can be pattern matched. The TemplateParameterList declares symbols based on the parts of the pattern that are matched, analogously to the way implied template parameters are matched.
Example: Matching a Template Instantiation
struct Tuple(T...) { // ... } alias Tup2 = Tuple!(int, string); staticif (is(Tup2 : Template!Args, alias Template, Args...)) { staticassert(__traits(isSame, Template, Tuple)); staticassert(is(Template!(int, string) == Tup2)); // same struct } staticassert(is(Args[0] == int)); staticassert(is(Args[1] == string));
Type cannot be matched when TypeSpecialization is an alias template instance:
struct S(T) {} alias A(T) = S!T; staticassert(is(A!int : S!T, T)); //static assert(!is(A!int : A!T, T));
Example: Matching an Associative Array
alias AA = long[string]; staticif (is(AA T : T[U], U : string)) // T[U] is the pattern { pragma(msg, T); // long pragma(msg, U); // string } // no match, B is not an int staticassert(!is(AA A : A[B], B : int));
Example: Matching a Static Array
staticif (is(int[10] W : W[len], int len)) // W[len] is the pattern { staticassert(len == 10); } staticassert(is(W == int)); // no match, len should be 10 staticassert(!is(int[10] X : X[len], int len : 5));
RvalueExpression: __rvalue (AssignExpression)
The RvalueExpression causes the embedded AssignExpression to be treated as an rvalue whether it is an rvalue or an lvalue.
If both ref and non-ref parameter overloads are present, an rvalue is preferably matched to the non-ref parameters, and an lvalue is preferably matched to the ref parameter. An RvalueExpression will preferably match with the non-ref parameter.
An rvalue argument is owned by the function called. Hence, if an lvalue is matched to the rvalue argument, a copy is made of the lvalue to be passed to the function. The function will then call the destructor (if any) on the parameter at the conclusion of the function. An rvalue argument is not copied, as it is assumed to already be unique, and is also destroyed at the conclusion of the function.
The called function's semantics are the same whether a parameter originated as an rvalue or is a copy of an lvalue. This means that an RvalueExpression argument destroys the expression upon function return. Attempts to continue to use the lvalue expression are invalid. The compiler won't always be able to detect a use after being passed to the function, which means that the destructor for the object must reset the object's contents to its initial value, or at least a benign value that can be destructed more than once.
struct S { ubyte* p; ~this() { free(p); // add `p = null;` here to prevent double free } } void aggh(S s) { } void oops() { S s; s.p = cast(ubyte*)malloc(10); aggh(__rvalue(s)); // destructor of s called at end of scope, double-freeing s.p }
RvalueExpressions enable the use of move constructors and move assignments.
It is also allowed to use the __rvalue keyword as a function attribute. The presence of this attribute will infer the RvalueExpression upon the function call expression at the call-site, as if applied to the function's return value. This is only accepted on functions that return by reference.
ref S fun(returnref S s) __rvalue { return s; } S s; S t = fun(s); // call inferred as: __rvalue(fun(s))
SpecialKeyword: __FILE____FILE_FULL_PATH____MODULE____LINE____FUNCTION____PRETTY_FUNCTION__
__FILE__ and __LINE__ expand to the source file name and line number at the point of instantiation. The path of the source file is left up to the compiler.
__FILE_FULL_PATH__ expands to the absolute source file name at the point of instantiation.
__MODULE__ expands to the module name at the point of instantiation.
__FUNCTION__ expands to the fully qualified name of the function at the point of instantiation.
__PRETTY_FUNCTION__ is similar to __FUNCTION__, but also expands the function return type, its parameter types, and its attributes.
Example:
module test; import std.stdio; void test(string file = __FILE__, size_t line = __LINE__, string mod = __MODULE__, string func = __FUNCTION__, string pretty = __PRETTY_FUNCTION__, string fileFullPath = __FILE_FULL_PATH__) { writefln("file: '%s', line: '%s', module: '%s',\nfunction: '%s', " ~ "pretty function: '%s',\nfile full path: '%s'", file, line, mod, func, pretty, fileFullPath); } int main(string[] args) { test(); return 0; }
Assuming the file was at /example/test.d, this will output:
file: 'test.d', line: '13', module: 'test', function: 'test.main', pretty function: 'int test.main(string[] args)', file full path: '/example/test.d'
Warning: Do not use mixin(__FUNCTION__) to get the symbol for the current function. This seems to be a common thing for programmers to attempt when trying to get the symbol for the current function in order to do introspection on it, since D does not currently have a direct way to get that symbol. However, using mixin(__FUNCTION__) means that the symbol for the function will be looked up by name, which means that it's subject to the various rules that go with symbol lookup, which can cause various problems. One such problem would that if a function is overloaded, the result will be the first overload whether that's the current function or not.
Given that D doesn't currently have a way to directly get the symbol for the current function, the best way to do it is to get the parent symbol of a symbol within the function, since that avoids any issues surrounding symbol lookup rules. An example of that which doesn't rely on any other symbols is __traits(parent {}). It declares an anonymous, nested function, whose parent is then the current function. So, getting its parent gets the symbol for the current function.
An implementation may rearrange the evaluation of expressions according to arithmetic associativity and commutativity rules as long as, within that thread of execution, no observable difference is possible.
This rule precludes any associative or commutative reordering of floating point expressions.
