Conditional Compilation

Conditional compilation is the process of selecting which code to compile and which code to not compile.

ConditionalDeclaration:
    Condition DeclarationBlock
    Condition DeclarationBlock else DeclarationBlock
    Condition : DeclDefsopt
    Condition DeclarationBlock else : DeclDefsopt

ConditionalStatement:
    Condition NoScopeNonEmptyStatement
    Condition NoScopeNonEmptyStatement else NoScopeNonEmptyStatement

If the Condition is satisfied, then the following DeclarationBlock or Statement is compiled in. If it is not satisfied, the DeclarationBlock or Statement after the optional else is compiled in.

Any DeclarationBlock or Statement that is not compiled in still must be syntactically correct.

No new scope is introduced, even if the DeclarationBlock or Statement is enclosed by { }.

ConditionalDeclarations and ConditionalStatements can be nested.

The StaticAssert can be used to issue errors at compilation time for branches of the conditional compilation that are errors.

Condition comes in the following forms:

Condition:
    VersionCondition
    DebugCondition
    StaticIfCondition

Version Condition

VersionCondition:
    version ( IntegerLiteral )
    version ( Identifier )
    version ( unittest )
    version ( assert )

Versions enable multiple versions of a module to be implemented with a single source file.

The VersionCondition is satisfied if the IntegerLiteral is greater than or equal to the current version level, or if Identifier matches a version identifier.

The version level and version identifier can be set on the command line by the -version switch or in the module itself with a VersionSpecification, or they can be predefined by the compiler.

Version identifiers are in their own unique name space, they do not conflict with debug identifiers or other symbols in the module. Version identifiers defined in one module have no influence over other imported modules.

int k;
version (Demo) // compile in this code block for the demo version
{
    int i;
    int k;    // error, k already defined

    i = 3;
}
x = i;      // uses the i declared above

version (X86)
{
    ... // implement custom inline assembler version
}
else
{
    ... // use default, but slow, version
}

The version(unittest) is satisfied if and only if the code is compiled with unit tests enabled (the -unittest option on dmd).

Version Specification

VersionSpecification:
    version = Identifier ;
    version = IntegerLiteral ;

The version specification makes it straightforward to group a set of features under one major version, for example:

version (ProfessionalEdition)
{
    version = FeatureA;
    version = FeatureB;
    version = FeatureC;
}
version (HomeEdition)
{
    version = FeatureA;
}
...
version (FeatureB)
{
    ... implement Feature B ...
}

Version identifiers or levels may not be forward referenced:

version (Foo)
{
    int x;
}
version = Foo;  // error, Foo already used

VersionSpecifications may only appear at module scope.

While the debug and version conditions superficially behave the same, they are intended for very different purposes. Debug statements are for adding debug code that is removed for the release version. Version statements are to aid in portability and multiple release versions.

Here's an example of a full version as opposed to a demo version:

class Foo
{
    int a, b;

    version(full)
    {
        int extrafunctionality()
        {
            ...
            return 1;  // extra functionality is supported
        }
    }
    else // demo
    {
        int extrafunctionality()
        {
            return 0;  // extra functionality is not supported
        }
    }
}

Various different version builds can be built with a parameter to version:

version(n) // add in version code if version level is >= n
{
    ... version code ...
}

version(identifier) // add in version code if version
                         // keyword is identifier
{
    ... version code ...
}

These are presumably set by the command line as -version=n and -version=identifier.

Predefined Versions

Several environmental version identifiers and identifier name spaces are predefined for consistent usage. Version identifiers do not conflict with other identifiers in the code, they are in a separate name space. Predefined version identifiers are global, i.e. they apply to all modules being compiled and imported.

The following identifiers are defined, but are deprecated:

Others will be added as they make sense and new implementations appear.

To allow for future growth of the language, the version identifier namespace beginning with "D_" is reserved for identifiers indicating D language specification or new feature conformance. Further, all identifiers derived from the ones listed above by appending any character(s) are reserved. This means that e.g. ARM_foo and Windows_bar are reserved while foo_ARM and bar_Windows are not.

Furthermore, predefined version identifiers from this list cannot be set from the command line or from version statements. (This prevents things like both Windows and linux being simultaneously set.)

Compiler vendor specific versions can be predefined if the trademarked vendor identifier prefixes it, as in:

version(DigitalMars_funky_extension)
{
    ...
}

It is important to use the right version identifier for the right purpose. For example, use the vendor identifier when using a vendor specific feature. Use the operating system identifier when using an operating system specific feature, etc.

Debug Condition

DebugCondition:
    debug
    debug ( IntegerLiteral )
    debug ( Identifier )

Two versions of programs are commonly built, a release build and a debug build. The debug build includes extra error checking code, test harnesses, pretty-printing code, etc. The debug statement conditionally compiles in its statement body. It is D's way of what in C is done with #ifdef DEBUG / #endif pairs.

The debug condition is satisfied when the -debug switch is passed to the compiler or when the debug level is >= 1.

The debug ( IntegerLiteral ) condition is satisfied when the debug level is >= IntegerLiteral.

The debug ( Identifier ) condition is satisfied when the debug identifier matches Identifier.

class Foo
{
    int a, b;
debug:
    int flag;
}

Debug Statement

A ConditionalStatement that has a DebugCondition is called a DebugStatement. DebugStatements have relaxed semantic checks in that pure, @nogc, nothrow and @safe checks are not done. Neither do DebugStatements influence the inference of pure, @nogc, nothrow and @safe attributes.

Debug Specification

DebugSpecification:
    debug = Identifier ;
    debug = IntegerLiteral ;

Debug identifiers and levels are set either by the command line switch -debug or by a DebugSpecification.

Debug specifications only affect the module they appear in, they do not affect any imported modules. Debug identifiers are in their own namespace, independent from version identifiers and other symbols.

It is illegal to forward reference a debug specification:

debug(foo) writeln("Foo");
debug = foo;    // error, foo used before set

DebugSpecifications may only appear at module scope.

Various different debug builds can be built with a parameter to debug:

debug(IntegerLiteral) { } // add in debug code if debug level is >= IntegerLiteral
debug(identifier) { } // add in debug code if debug keyword is identifier

These are presumably set by the command line as -debug=n and -debug=identifier.

Static If Condition

StaticIfCondition:
    static if ( AssignExpression )

AssignExpression is implicitly converted to a boolean type, and is evaluated at compile time. The condition is satisfied if it evaluates to true. It is not satisfied if it evaluates to false.

It is an error if AssignExpression cannot be implicitly converted to a boolean type or if it cannot be evaluated at compile time.

StaticIfConditions can appear in module, class, template, struct, union, or function scope. In function scope, the symbols referred to in the AssignExpression can be any that can normally be referenced by an expression at that point.

const int i = 3;
int j = 4;

static if (i == 3)    // ok, at module scope
    int x;

class C
{
    const int k = 5;

    static if (i == 3) // ok
        int x;
    else
        long x;

    static if (j == 3) // error, j is not a constant
        int y;

    static if (k == 5) // ok, k is in current scope
        int z;
}

template Int(int i)
{
    static if (i == 32)
        alias Int = int;
    else static if (i == 16)
        alias Int = short;
    else
        static assert(0); // not supported
}

Int!(32) a;  // a is an int
Int!(16) b;  // b is a short
Int!(17) c;  // error, static assert trips

A StaticIfCondition differs from an IfStatement in the following ways:

1. It can be used to conditionally compile declarations, not just statements.
2. It does not introduce a new scope even if { } are used for conditionally compiled statements.
3. For unsatisfied conditions, the conditionally compiled code need only be syntactically correct. It does not have to be semantically correct.
4. It must be evaluatable at compile time.

Static Foreach

StaticForeach:
    static AggregateForeach
    static RangeForeach

StaticForeachDeclaration:
    StaticForeach DeclarationBlock
    StaticForeach : DeclDefsopt

StaticForeachStatement:
    StaticForeach NoScopeNonEmptyStatement

The aggregate/range bounds are evaluated at compile time and turned into a sequence of compile-time entities by evaluating corresponding code with a ForeachStatement/ForeachRangeStatement at compile time. The body of the static foreach is then copied a number of times that corresponds to the number of elements of the sequence. Within the i-th copy, the name of the static foreach variable is bound to the i-th entry of the sequence, either as an enum variable declaration (for constants) or an alias declaration (for symbols). (In particular, static foreach variables are never runtime variables.)

static foreach(i; [0, 1, 2, 3])
{
    pragma(msg, i);
}

static foreach supports multiple variables in cases where the corresponding foreach statement supports them. (In this case, static foreach generates a compile-time sequence of tuples, and the tuples are subsequently unpacked during iteration).

static foreach(i, v; ['a', 'b', 'c', 'd'])
{
    static assert(i + 'a' == v);
}

Like bodies of ConditionalDeclarations, a static foreach body does not introduce a new scope. Therefore, it can be used to generate declarations:

import std.range : iota;

static foreach(i; iota(0, 3))
{
    mixin(`enum x`, i, ` = i;`);
}

pragma(msg, x0, " ", x1," ", x2); // 0 1 2

If a new scope is desired for each expansion, use another set of braces:

static foreach(s; ["hi", "hey", "hello"])
{{
    enum len = s.length;    // local to each iteration
    static assert(len <= 5);
}}

static assert(!is(typeof(len)));

break and continue

As static foreach is a code generation construct and not a loop, break and continue cannot be used to change control flow within it. Instead of breaking or continuing a suitable enclosing statement, such an usage yields an error (this is to prevent misunderstandings).

int test(int x)
{
    int r = -1;
    switch(x)
    {
        static foreach(i; 0 .. 100)
        {
            case i:
                r = i;
                break; // error
        }
        default: break;
    }
    return r;
}

static foreach(i; 0 .. 200)
{
    static assert(test(i) == (i < 100 ? i : -1));
}

An explicit break/continue label can be used to avoid this limitation. (Note that static foreach itself cannot be broken nor continued even if it is explicitly labeled.)

int test(int x)
{
    int r = -1;
    Lswitch: switch(x)
    {
        static foreach(i; 0 .. 100)
        {
            case i:
                r = i;
                break Lswitch;
        }
        default: break;
    }
    return r;
}

static foreach(i; 0 .. 200)
{
    static assert(test(i) == (i<100 ? i : -1));
}

Static Assert

StaticAssert:
    static assert ( AssertArguments ) ;

The first AssignExpression is evaluated at compile time, and converted to a boolean value. If the value is true, the static assert is ignored. If the value is false, an error diagnostic is issued and the compile fails.

Unlike AssertExpressions, StaticAsserts are always checked and evaluated by the compiler unless they appear in an unsatisfied conditional.

void foo()
{
    if (0)
    {
        assert(0);  // never trips
        static assert(0); // always trips
    }
    version (BAR)
    {
    }
    else
    {
        static assert(0); // trips when version BAR is not defined
    }
}

StaticAssert is useful tool for drawing attention to conditional configurations not supported in the code.

The optional second AssignExpression can be used to supply additional information, such as a text string, that will be printed out along with the error diagnostic.