Go to the first, previous, next, last section, table of contents.

Extensions to the C Language Family

GNU C provides several language features not found in ANSI standard C. (The `-pedantic' option directs GNU CC to print a warning message if any of these features is used.) To test for the availability of these features in conditional compilation, check for a predefined macro __GNUC__, which is always defined under GNU CC.

These extensions are available in C and Objective C. Most of them are also available in C++. See section Extensions to the C++ Language, for extensions that apply only to C++.

Statements and Declarations in Expressions

A compound statement enclosed in parentheses may appear as an expression in GNU C. This allows you to use loops, switches, and local variables within an expression.

Recall that a compound statement is a sequence of statements surrounded by braces; in this construct, parentheses go around the braces. For example:

({ int y = foo (); int z;
   if (y > 0) z = y;
   else z = - y;
   z; })

is a valid (though slightly more complex than necessary) expression for the absolute value of foo ().

The last thing in the compound statement should be an expression followed by a semicolon; the value of this subexpression serves as the value of the entire construct. (If you use some other kind of statement last within the braces, the construct has type void, and thus effectively no value.)

This feature is especially useful in making macro definitions "safe" (so that they evaluate each operand exactly once). For example, the "maximum" function is commonly defined as a macro in standard C as follows:

#define max(a,b) ((a) > (b) ? (a) : (b))

But this definition computes either a or b twice, with bad results if the operand has side effects. In GNU C, if you know the type of the operands (here let's assume int), you can define the macro safely as follows:

#define maxint(a,b) \
  ({int _a = (a), _b = (b); _a > _b ? _a : _b; })

Embedded statements are not allowed in constant expressions, such as the value of an enumeration constant, the width of a bit field, or the initial value of a static variable.

If you don't know the type of the operand, you can still do this, but you must use typeof (see section Referring to a Type with typeof) or type naming (see section Naming an Expression's Type).

Locally Declared Labels

Each statement expression is a scope in which local labels can be declared. A local label is simply an identifier; you can jump to it with an ordinary goto statement, but only from within the statement expression it belongs to.

A local label declaration looks like this:

__label__ label;

or

__label__ label1, label2, ...;

Local label declarations must come at the beginning of the statement expression, right after the `({', before any ordinary declarations.

The label declaration defines the label name, but does not define the label itself. You must do this in the usual way, with label:, within the statements of the statement expression.

The local label feature is useful because statement expressions are often used in macros. If the macro contains nested loops, a goto can be useful for breaking out of them. However, an ordinary label whose scope is the whole function cannot be used: if the macro can be expanded several times in one function, the label will be multiply defined in that function. A local label avoids this problem. For example:

#define SEARCH(array, target)                     \
({                                               \
  __label__ found;                                \
  typeof (target) _SEARCH_target = (target);      \
  typeof (*(array)) *_SEARCH_array = (array);     \
  int i, j;                                       \
  int value;                                      \
  for (i = 0; i < max; i++)                       \
    for (j = 0; j < max; j++)                     \
      if (_SEARCH_array[i][j] == _SEARCH_target)  \
        { value = i; goto found; }              \
  value = -1;                                     \
 found:                                           \
  value;                                          \
})

Labels as Values

You can get the address of a label defined in the current function (or a containing function) with the unary operator `&&'. The value has type void *. This value is a constant and can be used wherever a constant of that type is valid. For example:

void *ptr;
...
ptr = &&foo;

To use these values, you need to be able to jump to one. This is done with the computed goto statement(1), goto *exp;. For example,

goto *ptr;

Any expression of type void * is allowed.

One way of using these constants is in initializing a static array that will serve as a jump table:

static void *array[] = { &&foo, &&bar, &&hack };

Then you can select a label with indexing, like this:

goto *array[i];

Note that this does not check whether the subscript is in bounds--array indexing in C never does that.

Such an array of label values serves a purpose much like that of the switch statement. The switch statement is cleaner, so use that rather than an array unless the problem does not fit a switch statement very well.

Another use of label values is in an interpreter for threaded code. The labels within the interpreter function can be stored in the threaded code for super-fast dispatching.

You can use this mechanism to jump to code in a different function. If you do that, totally unpredictable things will happen. The best way to avoid this is to store the label address only in automatic variables and never pass it as an argument.

Nested Functions

A nested function is a function defined inside another function. (Nested functions are not supported for GNU C++.) The nested function's name is local to the block where it is defined. For example, here we define a nested function named square, and call it twice:

foo (double a, double b)
{
  double square (double z) { return z * z; }

  return square (a) + square (b);
}

The nested function can access all the variables of the containing function that are visible at the point of its definition. This is called lexical scoping. For example, here we show a nested function which uses an inherited variable named offset:

bar (int *array, int offset, int size)
{
  int access (int *array, int index)
    { return array[index + offset]; }
  int i;
  ...
  for (i = 0; i < size; i++)
    ... access (array, i) ...
}

Nested function definitions are permitted within functions in the places where variable definitions are allowed; that is, in any block, before the first statement in the block.

It is possible to call the nested function from outside the scope of its name by storing its address or passing the address to another function:

hack (int *array, int size)
{
  void store (int index, int value)
    { array[index] = value; }

  intermediate (store, size);
}

Here, the function intermediate receives the address of store as an argument. If intermediate calls store, the arguments given to store are used to store into array. But this technique works only so long as the containing function (hack, in this example) does not exit.

If you try to call the nested function through its address after the containing function has exited, all hell will break loose. If you try to call it after a containing scope level has exited, and if it refers to some of the variables that are no longer in scope, you may be lucky, but it's not wise to take the risk. If, however, the nested function does not refer to anything that has gone out of scope, you should be safe.

GNU CC implements taking the address of a nested function using a technique called trampolines. A paper describing them is available as `http://master.debian.org/~karlheg/Usenix88-lexic.pdf'.

A nested function can jump to a label inherited from a containing function, provided the label was explicitly declared in the containing function (see section Locally Declared Labels). Such a jump returns instantly to the containing function, exiting the nested function which did the goto and any intermediate functions as well. Here is an example:

bar (int *array, int offset, int size)
{
  __label__ failure;
  int access (int *array, int index)
    {
      if (index > size)
        goto failure;
      return array[index + offset];
    }
  int i;
  ...
  for (i = 0; i < size; i++)
    ... access (array, i) ...
  ...
  return 0;

 /* Control comes here from access
    if it detects an error.  */
 failure:
  return -1;
}

A nested function always has internal linkage. Declaring one with extern is erroneous. If you need to declare the nested function before its definition, use auto (which is otherwise meaningless for function declarations).

bar (int *array, int offset, int size)
{
  __label__ failure;
  auto int access (int *, int);
  ...
  int access (int *array, int index)
    {
      if (index > size)
        goto failure;
      return array[index + offset];
    }
  ...
}

Constructing Function Calls

Using the built-in functions described below, you can record the arguments a function received, and call another function with the same arguments, without knowing the number or types of the arguments.

You can also record the return value of that function call, and later return that value, without knowing what data type the function tried to return (as long as your caller expects that data type).

__builtin_apply_args ()
This built-in function returns a pointer of type void * to data describing how to perform a call with the same arguments as were passed to the current function. The function saves the arg pointer register, structure value address, and all registers that might be used to pass arguments to a function into a block of memory allocated on the stack. Then it returns the address of that block.
__builtin_apply (function, arguments, size)
This built-in function invokes function (type void (*)()) with a copy of the parameters described by arguments (type void *) and size (type int). The value of arguments should be the value returned by __builtin_apply_args. The argument size specifies the size of the stack argument data, in bytes. This function returns a pointer of type void * to data describing how to return whatever value was returned by function. The data is saved in a block of memory allocated on the stack. It is not always simple to compute the proper value for size. The value is used by __builtin_apply to compute the amount of data that should be pushed on the stack and copied from the incoming argument area.
__builtin_return (result)
This built-in function returns the value described by result from the containing function. You should specify, for result, a value returned by __builtin_apply.

Naming an Expression's Type

You can give a name to the type of an expression using a typedef declaration with an initializer. Here is how to define name as a type name for the type of exp:

typedef name = exp;

This is useful in conjunction with the statements-within-expressions feature. Here is how the two together can be used to define a safe "maximum" macro that operates on any arithmetic type:

#define max(a,b) \
  ({typedef _ta = (a), _tb = (b);  \
    _ta _a = (a); _tb _b = (b);     \
    _a > _b ? _a : _b; })

The reason for using names that start with underscores for the local variables is to avoid conflicts with variable names that occur within the expressions that are substituted for a and b. Eventually we hope to design a new form of declaration syntax that allows you to declare variables whose scopes start only after their initializers; this will be a more reliable way to prevent such conflicts.

Referring to a Type with typeof

Another way to refer to the type of an expression is with typeof. The syntax of using of this keyword looks like sizeof, but the construct acts semantically like a type name defined with typedef.

There are two ways of writing the argument to typeof: with an expression or with a type. Here is an example with an expression:

typeof (x[0](1))

This assumes that x is an array of functions; the type described is that of the values of the functions.

Here is an example with a typename as the argument:

typeof (int *)

Here the type described is that of pointers to int.

If you are writing a header file that must work when included in ANSI C programs, write __typeof__ instead of typeof. See section Alternate Keywords.

A typeof-construct can be used anywhere a typedef name could be used. For example, you can use it in a declaration, in a cast, or inside of sizeof or typeof.

Generalized Lvalues

Compound expressions, conditional expressions and casts are allowed as lvalues provided their operands are lvalues. This means that you can take their addresses or store values into them.

Standard C++ allows compound expressions and conditional expressions as lvalues, and permits casts to reference type, so use of this extension is deprecated for C++ code.

For example, a compound expression can be assigned, provided the last expression in the sequence is an lvalue. These two expressions are equivalent:

(a, b) += 5
a, (b += 5)

Similarly, the address of the compound expression can be taken. These two expressions are equivalent:

&(a, b)
a, &b

A conditional expression is a valid lvalue if its type is not void and the true and false branches are both valid lvalues. For example, these two expressions are equivalent:

(a ? b : c) = 5
(a ? b = 5 : (c = 5))

A cast is a valid lvalue if its operand is an lvalue. A simple assignment whose left-hand side is a cast works by converting the right-hand side first to the specified type, then to the type of the inner left-hand side expression. After this is stored, the value is converted back to the specified type to become the value of the assignment. Thus, if a has type char *, the following two expressions are equivalent:

(int)a = 5
(int)(a = (char *)(int)5)

An assignment-with-arithmetic operation such as `+=' applied to a cast performs the arithmetic using the type resulting from the cast, and then continues as in the previous case. Therefore, these two expressions are equivalent:

(int)a += 5
(int)(a = (char *)(int) ((int)a + 5))

You cannot take the address of an lvalue cast, because the use of its address would not work out coherently. Suppose that &(int)f were permitted, where f has type float. Then the following statement would try to store an integer bit-pattern where a floating point number belongs:

*&(int)f = 1;

This is quite different from what (int)f = 1 would do--that would convert 1 to floating point and store it. Rather than cause this inconsistency, we think it is better to prohibit use of `&' on a cast.

If you really do want an int * pointer with the address of f, you can simply write (int *)&f.

Conditionals with Omitted Operands

The middle operand in a conditional expression may be omitted. Then if the first operand is nonzero, its value is the value of the conditional expression.

Therefore, the expression

x ? : y

has the value of x if that is nonzero; otherwise, the value of y.

This example is perfectly equivalent to

x ? x : y

In this simple case, the ability to omit the middle operand is not especially useful. When it becomes useful is when the first operand does, or may (if it is a macro argument), contain a side effect. Then repeating the operand in the middle would perform the side effect twice. Omitting the middle operand uses the value already computed without the undesirable effects of recomputing it.

Double-Word Integers

GNU C supports data types for integers that are twice as long as int. Simply write long long int for a signed integer, or unsigned long long int for an unsigned integer. To make an integer constant of type long long int, add the suffix LL to the integer. To make an integer constant of type unsigned long long int, add the suffix ULL to the integer.

You can use these types in arithmetic like any other integer types. Addition, subtraction, and bitwise boolean operations on these types are open-coded on all types of machines. Multiplication is open-coded if the machine supports fullword-to-doubleword a widening multiply instruction. Division and shifts are open-coded only on machines that provide special support. The operations that are not open-coded use special library routines that come with GNU CC.

There may be pitfalls when you use long long types for function arguments, unless you declare function prototypes. If a function expects type int for its argument, and you pass a value of type long long int, confusion will result because the caller and the subroutine will disagree about the number of bytes for the argument. Likewise, if the function expects long long int and you pass int. The best way to avoid such problems is to use prototypes.

Complex Numbers

GNU C supports complex data types. You can declare both complex integer types and complex floating types, using the keyword __complex__.

For example, `__complex__ double x;' declares x as a variable whose real part and imaginary part are both of type double. `__complex__ short int y;' declares y to have real and imaginary parts of type short int; this is not likely to be useful, but it shows that the set of complex types is complete.

To write a constant with a complex data type, use the suffix `i' or `j' (either one; they are equivalent). For example, 2.5fi has type __complex__ float and 3i has type __complex__ int. Such a constant always has a pure imaginary value, but you can form any complex value you like by adding one to a real constant.

To extract the real part of a complex-valued expression exp, write __real__ exp. Likewise, use __imag__ to extract the imaginary part.

The operator `~' performs complex conjugation when used on a value with a complex type.

GNU CC can allocate complex automatic variables in a noncontiguous fashion; it's even possible for the real part to be in a register while the imaginary part is on the stack (or vice-versa). None of the supported debugging info formats has a way to represent noncontiguous allocation like this, so GNU CC describes a noncontiguous complex variable as if it were two separate variables of noncomplex type. If the variable's actual name is foo, the two fictitious variables are named foo$real and foo$imag. You can examine and set these two fictitious variables with your debugger.

A future version of GDB will know how to recognize such pairs and treat them as a single variable with a complex type.

Hex Floats

GNU CC recognizes floating-point numbers written not only in the usual decimal notation, such as 1.55e1, but also numbers such as 0x1.fp3 written in hexadecimal format. In that format the 0x hex introducer and the p or P exponent field are mandatory. The exponent is a decimal number that indicates the power of 2 by which the significand part will be multiplied. Thus 0x1.f is 1 15/16, p3 multiplies it by 8, and the value of 0x1.fp3 is the same as 1.55e1.

Unlike for floating-point numbers in the decimal notation the exponent is always required in the hexadecimal notation. Otherwise the compiler would not be able to resolve the ambiguity of, e.g., 0x1.f. This could mean 1.0f or 1.9375 since f is also the extension for floating-point constants of type float.

Arrays of Length Zero

Zero-length arrays are allowed in GNU C. They are very useful as the last element of a structure which is really a header for a variable-length object:

struct line {
  int length;
  char contents[0];
};

{
  struct line *thisline = (struct line *)
    malloc (sizeof (struct line) + this_length);
  thisline->length = this_length;
}

In standard C, you would have to give contents a length of 1, which means either you waste space or complicate the argument to malloc.

Arrays of Variable Length

Variable-length automatic arrays are allowed in GNU C. These arrays are declared like any other automatic arrays, but with a length that is not a constant expression. The storage is allocated at the point of declaration and deallocated when the brace-level is exited. For example:

FILE *
concat_fopen (char *s1, char *s2, char *mode)
{
  char str[strlen (s1) + strlen (s2) + 1];
  strcpy (str, s1);
  strcat (str, s2);
  return fopen (str, mode);
}

Jumping or breaking out of the scope of the array name deallocates the storage. Jumping into the scope is not allowed; you get an error message for it.

You can use the function alloca to get an effect much like variable-length arrays. The function alloca is available in many other C implementations (but not in all). On the other hand, variable-length arrays are more elegant.

There are other differences between these two methods. Space allocated with alloca exists until the containing function returns. The space for a variable-length array is deallocated as soon as the array name's scope ends. (If you use both variable-length arrays and alloca in the same function, deallocation of a variable-length array will also deallocate anything more recently allocated with alloca.)

You can also use variable-length arrays as arguments to functions:

struct entry
tester (int len, char data[len][len])
{
  ...
}

The length of an array is computed once when the storage is allocated and is remembered for the scope of the array in case you access it with sizeof.

If you want to pass the array first and the length afterward, you can use a forward declaration in the parameter list--another GNU extension.

struct entry
tester (int len; char data[len][len], int len)
{
  ...
}

The `int len' before the semicolon is a parameter forward declaration, and it serves the purpose of making the name len known when the declaration of data is parsed.

You can write any number of such parameter forward declarations in the parameter list. They can be separated by commas or semicolons, but the last one must end with a semicolon, which is followed by the "real" parameter declarations. Each forward declaration must match a "real" declaration in parameter name and data type.

Macros with Variable Numbers of Arguments

In GNU C, a macro can accept a variable number of arguments, much as a function can. The syntax for defining the macro looks much like that used for a function. Here is an example:

#define eprintf(format, args...)  \
 fprintf (stderr, format , ## args)

Here args is a rest argument: it takes in zero or more arguments, as many as the call contains. All of them plus the commas between them form the value of args, which is substituted into the macro body where args is used. Thus, we have this expansion:

eprintf ("%s:%d: ", input_file_name, line_number)
==>
fprintf (stderr, "%s:%d: " , input_file_name, line_number)

Note that the comma after the string constant comes from the definition of eprintf, whereas the last comma comes from the value of args.

The reason for using `##' is to handle the case when args matches no arguments at all. In this case, args has an empty value. In this case, the second comma in the definition becomes an embarrassment: if it got through to the expansion of the macro, we would get something like this:

fprintf (stderr, "success!\n" , )

which is invalid C syntax. `##' gets rid of the comma, so we get the following instead:

fprintf (stderr, "success!\n")

This is a special feature of the GNU C preprocessor: `##' before a rest argument that is empty discards the preceding sequence of non-whitespace characters from the macro definition. (If another macro argument precedes, none of it is discarded.)

It might be better to discard the last preprocessor token instead of the last preceding sequence of non-whitespace characters; in fact, we may someday change this feature to do so. We advise you to write the macro definition so that the preceding sequence of non-whitespace characters is just a single token, so that the meaning will not change if we change the definition of this feature.

Non-Lvalue Arrays May Have Subscripts

Subscripting is allowed on arrays that are not lvalues, even though the unary `&' operator is not. For example, this is valid in GNU C though not valid in other C dialects:

struct foo {int a[4];};

struct foo f();

bar (int index)
{
  return f().a[index];
}

Arithmetic on void- and Function-Pointers

In GNU C, addition and subtraction operations are supported on pointers to void and on pointers to functions. This is done by treating the size of a void or of a function as 1.

A consequence of this is that sizeof is also allowed on void and on function types, and returns 1.

The option `-Wpointer-arith' requests a warning if these extensions are used.

Non-Constant Initializers

As in standard C++, the elements of an aggregate initializer for an automatic variable are not required to be constant expressions in GNU C. Here is an example of an initializer with run-time varying elements:

foo (float f, float g)
{
  float beat_freqs[2] = { f-g, f+g };
  ...
}

Constructor Expressions

GNU C supports constructor expressions. A constructor looks like a cast containing an initializer. Its value is an object of the type specified in the cast, containing the elements specified in the initializer.

Usually, the specified type is a structure. Assume that struct foo and structure are declared as shown:

struct foo {int a; char b[2];} structure;

Here is an example of constructing a struct foo with a constructor:

structure = ((struct foo) {x + y, 'a', 0});

This is equivalent to writing the following:

{
  struct foo temp = {x + y, 'a', 0};
  structure = temp;
}

You can also construct an array. If all the elements of the constructor are (made up of) simple constant expressions, suitable for use in initializers, then the constructor is an lvalue and can be coerced to a pointer to its first element, as shown here:

char **foo = (char *[]) { "x", "y", "z" };

Array constructors whose elements are not simple constants are not very useful, because the constructor is not an lvalue. There are only two valid ways to use it: to subscript it, or initialize an array variable with it. The former is probably slower than a switch statement, while the latter does the same thing an ordinary C initializer would do. Here is an example of subscripting an array constructor:

output = ((int[]) { 2, x, 28 }) [input];

Constructor expressions for scalar types and union types are is also allowed, but then the constructor expression is equivalent to a cast.

Labeled Elements in Initializers

Standard C requires the elements of an initializer to appear in a fixed order, the same as the order of the elements in the array or structure being initialized.

In GNU C you can give the elements in any order, specifying the array indices or structure field names they apply to. This extension is not implemented in GNU C++.

To specify an array index, write `[index]' or `[index] =' before the element value. For example,

int a[6] = { [4] 29, [2] = 15 };

is equivalent to

int a[6] = { 0, 0, 15, 0, 29, 0 };

The index values must be constant expressions, even if the array being initialized is automatic.

To initialize a range of elements to the same value, write `[first ... last] = value'. For example,

int widths[] = { [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 };

Note that the length of the array is the highest value specified plus one.

In a structure initializer, specify the name of a field to initialize with `fieldname:' before the element value. For example, given the following structure,

struct point { int x, y; };

the following initialization

struct point p = { y: yvalue, x: xvalue };

is equivalent to

struct point p = { xvalue, yvalue };

Another syntax which has the same meaning is `.fieldname ='., as shown here:

struct point p = { .y = yvalue, .x = xvalue };

You can also use an element label (with either the colon syntax or the period-equal syntax) when initializing a union, to specify which element of the union should be used. For example,

union foo { int i; double d; };

union foo f = { d: 4 };

will convert 4 to a double to store it in the union using the second element. By contrast, casting 4 to type union foo would store it into the union as the integer i, since it is an integer. (See section Cast to a Union Type.)

You can combine this technique of naming elements with ordinary C initialization of successive elements. Each initializer element that does not have a label applies to the next consecutive element of the array or structure. For example,

int a[6] = { [1] = v1, v2, [4] = v4 };

is equivalent to

int a[6] = { 0, v1, v2, 0, v4, 0 };

Labeling the elements of an array initializer is especially useful when the indices are characters or belong to an enum type. For example:

int whitespace[256]
  = { [' '] = 1, ['\t'] = 1, ['\h'] = 1,
      ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 };

Case Ranges

You can specify a range of consecutive values in a single case label, like this:

case low ... high:

This has the same effect as the proper number of individual case labels, one for each integer value from low to high, inclusive.

This feature is especially useful for ranges of ASCII character codes:

case 'A' ... 'Z':

Be careful: Write spaces around the ..., for otherwise it may be parsed wrong when you use it with integer values. For example, write this:

case 1 ... 5:

rather than this:

case 1...5:

Cast to a Union Type

A cast to union type is similar to other casts, except that the type specified is a union type. You can specify the type either with union tag or with a typedef name. A cast to union is actually a constructor though, not a cast, and hence does not yield an lvalue like normal casts. (See section Constructor Expressions.)

The types that may be cast to the union type are those of the members of the union. Thus, given the following union and variables:

union foo { int i; double d; };
int x;
double y;

both x and y can be cast to type union foo.

Using the cast as the right-hand side of an assignment to a variable of union type is equivalent to storing in a member of the union:

union foo u;
...
u = (union foo) x  ==  u.i = x
u = (union foo) y  ==  u.d = y

You can also use the union cast as a function argument:

void hack (union foo);
...
hack ((union foo) x);

Declaring Attributes of Functions

In GNU C, you declare certain things about functions called in your program which help the compiler optimize function calls and check your code more carefully.

The keyword __attribute__ allows you to specify special attributes when making a declaration. This keyword is followed by an attribute specification inside double parentheses. Nine attributes, noreturn, const, format, no_instrument_function, section, constructor, destructor, unused and weak are currently defined for functions. Other attributes, including section are supported for variables declarations (see section Specifying Attributes of Variables) and for types (see section Specifying Attributes of Types).

You may also specify attributes with `__' preceding and following each keyword. This allows you to use them in header files without being concerned about a possible macro of the same name. For example, you may use __noreturn__ instead of noreturn.

noreturn
A few standard library functions, such as abort and exit, cannot return. GNU CC knows this automatically. Some programs define their own functions that never return. You can declare them noreturn to tell the compiler this fact. For example,
void fatal () __attribute__ ((noreturn));

void
fatal (...)
{
  ... /* Print error message. */ ...
  exit (1);
}
The noreturn keyword tells the compiler to assume that fatal cannot return. It can then optimize without regard to what would happen if fatal ever did return. This makes slightly better code. More importantly, it helps avoid spurious warnings of uninitialized variables. Do not assume that registers saved by the calling function are restored before calling the noreturn function. It does not make sense for a noreturn function to have a return type other than void. The attribute noreturn is not implemented in GNU C versions earlier than 2.5. An alternative way to declare that a function does not return, which works in the current version and in some older versions, is as follows:
typedef void voidfn ();

volatile voidfn fatal;
const
Many functions do not examine any values except their arguments, and have no effects except the return value. Such a function can be subject to common subexpression elimination and loop optimization just as an arithmetic operator would be. These functions should be declared with the attribute const. For example,
int square (int) __attribute__ ((const));
says that the hypothetical function square is safe to call fewer times than the program says. The attribute const is not implemented in GNU C versions earlier than 2.5. An alternative way to declare that a function has no side effects, which works in the current version and in some older versions, is as follows:
typedef int intfn ();

extern const intfn square;
This approach does not work in GNU C++ from 2.6.0 on, since the language specifies that the `const' must be attached to the return value. Note that a function that has pointer arguments and examines the data pointed to must not be declared const. Likewise, a function that calls a non-const function usually must not be const. It does not make sense for a const function to return void.
format (archetype, string-index, first-to-check)
The format attribute specifies that a function takes printf, scanf, or strftime style arguments which should be type-checked against a format string. For example, the declaration:
extern int
my_printf (void *my_object, const char *my_format, ...)
      __attribute__ ((format (printf, 2, 3)));
causes the compiler to check the arguments in calls to my_printf for consistency with the printf style format string argument my_format. The parameter archetype determines how the format string is interpreted, and should be either printf, scanf, or strftime. The parameter string-index specifies which argument is the format string argument (starting from 1), while first-to-check is the number of the first argument to check against the format string. For functions where the arguments are not available to be checked (such as vprintf), specify the third parameter as zero. In this case the compiler only checks the format string for consistency. In the example above, the format string (my_format) is the second argument of the function my_print, and the arguments to check start with the third argument, so the correct parameters for the format attribute are 2 and 3. The format attribute allows you to identify your own functions which take format strings as arguments, so that GNU CC can check the calls to these functions for errors. The compiler always checks formats for the ANSI library functions printf, fprintf, sprintf, scanf, fscanf, sscanf, strftime, vprintf, vfprintf and vsprintf whenever such warnings are requested (using `-Wformat'), so there is no need to modify the header file `stdio.h'.
format_arg (string-index)
The format_arg attribute specifies that a function takes printf or scanf style arguments, modifies it (for example, to translate it into another language), and passes it to a printf or scanf style function. For example, the declaration:
extern char *
my_dgettext (char *my_domain, const char *my_format)
      __attribute__ ((format_arg (2)));
causes the compiler to check the arguments in calls to my_dgettext whose result is passed to a printf, scanf, or strftime type function for consistency with the printf style format string argument my_format. The parameter string-index specifies which argument is the format string argument (starting from 1). The format-arg attribute allows you to identify your own functions which modify format strings, so that GNU CC can check the calls to printf, scanf, or strftime function whose operands are a call to one of your own function. The compiler always treats gettext, dgettext, and dcgettext in this manner.
no_instrument_function
If `-finstrument-functions' is given, profiling function calls will be generated at entry and exit of most user-compiled functions. Functions with this attribute will not be so instrumented.
section ("section-name")
Normally, the compiler places the code it generates in the text section. Sometimes, however, you need additional sections, or you need certain particular functions to appear in special sections. The section attribute specifies that a function lives in a particular section. For example, the declaration:
extern void foobar (void) __attribute__ ((section ("bar")));
puts the function foobar in the bar section. Some file formats do not support arbitrary sections so the section attribute is not available on all platforms. If you need to map the entire contents of a module to a particular section, consider using the facilities of the linker instead.
constructor
destructor
The constructor attribute causes the function to be called automatically before execution enters main (). Similarly, the destructor attribute causes the function to be called automatically after main () has completed or exit () has been called. Functions with these attributes are useful for initializing data that will be used implicitly during the execution of the program. These attributes are not currently implemented for Objective C.
unused
This attribute, attached to a function, means that the function is meant to be possibly unused. GNU CC will not produce a warning for this function. GNU C++ does not currently support this attribute as definitions without parameters are valid in C++.
weak
The weak attribute causes the declaration to be emitted as a weak symbol rather than a global. This is primarily useful in defining library functions which can be overridden in user code, though it can also be used with non-function declarations. Weak symbols are supported for ELF targets, and also for a.out targets when using the GNU assembler and linker.
alias ("target")
The alias attribute causes the declaration to be emitted as an alias for another symbol, which must be specified. For instance,
void __f () { /* do something */; }
void f () __attribute__ ((weak, alias ("__f")));
declares `f' to be a weak alias for `__f'. In C++, the mangled name for the target must be used. Not all target machines support this attribute.
no_check_memory_usage
If `-fcheck-memory-usage' is given, calls to support routines will be generated before most memory accesses, to permit support code to record usage and detect uses of uninitialized or unallocated storage. Since the compiler cannot handle them properly, asm statements are not allowed. Declaring a function with this attribute disables the memory checking code for that function, permitting the use of asm statements without requiring separate compilation with different options, and allowing you to write support routines of your own if you wish, without getting infinite recursion if they get compiled with this option.
regparm (number)
On the Intel 386, the regparm attribute causes the compiler to pass up to number integer arguments in registers EAX, EDX, and ECX instead of on the stack. Functions that take a variable number of arguments will continue to be passed all of their arguments on the stack.
stdcall
On the Intel 386, the stdcall attribute causes the compiler to assume that the called function will pop off the stack space used to pass arguments, unless it takes a variable number of arguments. The PowerPC compiler for Windows NT currently ignores the stdcall attribute.
cdecl
On the Intel 386, the cdecl attribute causes the compiler to assume that the calling function will pop off the stack space used to pass arguments. This is useful to override the effects of the `-mrtd' switch. The PowerPC compiler for Windows NT currently ignores the cdecl attribute.
longcall
On the RS/6000 and PowerPC, the longcall attribute causes the compiler to always call the function via a pointer, so that functions which reside further than 64 megabytes (67,108,864 bytes) from the current location can be called.
dllimport
On the PowerPC running Windows NT, the dllimport attribute causes the compiler to call the function via a global pointer to the function pointer that is set up by the Windows NT dll library. The pointer name is formed by combining __imp_ and the function name.
dllexport
On the PowerPC running Windows NT, the dllexport attribute causes the compiler to provide a global pointer to the function pointer, so that it can be called with the dllimport attribute. The pointer name is formed by combining __imp_ and the function name.
exception (except-func [, except-arg])
On the PowerPC running Windows NT, the exception attribute causes the compiler to modify the structured exception table entry it emits for the declared function. The string or identifier except-func is placed in the third entry of the structured exception table. It represents a function, which is called by the exception handling mechanism if an exception occurs. If it was specified, the string or identifier except-arg is placed in the fourth entry of the structured exception table.
function_vector
Use this option on the H8/300 and H8/300H to indicate that the specified function should be called through the function vector. Calling a function through the function vector will reduce code size, however; the function vector has a limited size (maximum 128 entries on the H8/300 and 64 entries on the H8/300H) and shares space with the interrupt vector. You must use GAS and GLD from GNU binutils version 2.7 or later for this option to work correctly.
interrupt_handler
Use this option on the H8/300 and H8/300H to indicate that the specified function is an interrupt handler. The compiler will generate function entry and exit sequences suitable for use in an interrupt handler when this attribute is present.
eightbit_data
Use this option on the H8/300 and H8/300H to indicate that the specified variable should be placed into the eight bit data section. The compiler will generate more efficient code for certain operations on data in the eight bit data area. Note the eight bit data area is limited to 256 bytes of data. You must use GAS and GLD from GNU binutils version 2.7 or later for this option to work correctly.
tiny_data
Use this option on the H8/300H to indicate that the specified variable should be placed into the tiny data section. The compiler will generate more efficient code for loads and stores on data in the tiny data section. Note the tiny data area is limited to slightly under 32kbytes of data.
interrupt
Use this option on the M32R/D to indicate that the specified function is an interrupt handler. The compiler will generate function entry and exit sequences suitable for use in an interrupt handler when this attribute is present.
model (model-name)
Use this attribute on the M32R/D to set the addressability of an object, and the code generated for a function. The identifier model-name is one of small, medium, or large, representing each of the code models. Small model objects live in the lower 16MB of memory (so that their addresses can be loaded with the ld24 instruction), and are callable with the bl instruction. Medium model objects may live anywhere in the 32 bit address space (the compiler will generate seth/add3 instructions to load their addresses), and are callable with the bl instruction. Large model objects may live anywhere in the 32 bit address space (the compiler will generate seth/add3 instructions to load their addresses), and may not be reachable with the bl instruction (the compiler will generate the much slower seth/add3/jl instruction sequence).

You can specify multiple attributes in a declaration by separating them by commas within the double parentheses or by immediately following an attribute declaration with another attribute declaration.

Some people object to the __attribute__ feature, suggesting that ANSI C's #pragma should be used instead. There are two reasons for not doing this.

  1. It is impossible to generate #pragma commands from a macro.
  2. There is no telling what the same #pragma might mean in another compiler.

These two reasons apply to almost any application that might be proposed for #pragma. It is basically a mistake to use #pragma for anything.

Prototypes and Old-Style Function Definitions

GNU C extends ANSI C to allow a function prototype to override a later old-style non-prototype definition. Consider the following example:

/* Use prototypes unless the compiler is old-fashioned.  */
#ifdef __STDC__
#define P(x) x
#else
#define P(x) ()
#endif

/* Prototype function declaration.  */
int isroot P((uid_t));

/* Old-style function definition.  */
int
isroot (x)   /* ??? lossage here ??? */
     uid_t x;
{
  return x == 0;
}

Suppose the type uid_t happens to be short. ANSI C does not allow this example, because subword arguments in old-style non-prototype definitions are promoted. Therefore in this example the function definition's argument is really an int, which does not match the prototype argument type of short.

This restriction of ANSI C makes it hard to write code that is portable to traditional C compilers, because the programmer does not know whether the uid_t type is short, int, or long. Therefore, in cases like these GNU C allows a prototype to override a later old-style definition. More precisely, in GNU C, a function prototype argument type overrides the argument type specified by a later old-style definition if the former type is the same as the latter type before promotion. Thus in GNU C the above example is equivalent to the following:

int isroot (uid_t);

int
isroot (uid_t x)
{
  return x == 0;
}

GNU C++ does not support old-style function definitions, so this extension is irrelevant.

C++ Style Comments

In GNU C, you may use C++ style comments, which start with `//' and continue until the end of the line. Many other C implementations allow such comments, and they are likely to be in a future C standard. However, C++ style comments are not recognized if you specify `-ansi' or `-traditional', since they are incompatible with traditional constructs like dividend//*comment*/divisor.

Dollar Signs in Identifier Names

In GNU C, you may normally use dollar signs in identifier names. This is because many traditional C implementations allow such identifiers. However, dollar signs in identifiers are not supported on a few target machines, typically because the target assembler does not allow them.

The Character ESC in Constants

You can use the sequence `\e' in a string or character constant to stand for the ASCII character ESC.

Inquiring on Alignment of Types or Variables

The keyword __alignof__ allows you to inquire about how an object is aligned, or the minimum alignment usually required by a type. Its syntax is just like sizeof.

For example, if the target machine requires a double value to be aligned on an 8-byte boundary, then __alignof__ (double) is 8. This is true on many RISC machines. On more traditional machine designs, __alignof__ (double) is 4 or even 2.

Some machines never actually require alignment; they allow reference to any data type even at an odd addresses. For these machines, __alignof__ reports the recommended alignment of a type.

When the operand of __alignof__ is an lvalue rather than a type, the value is the largest alignment that the lvalue is known to have. It may have this alignment as a result of its data type, or because it is part of a structure and inherits alignment from that structure. For example, after this declaration:

struct foo { int x; char y; } foo1;

the value of __alignof__ (foo1.y) is probably 2 or 4, the same as __alignof__ (int), even though the data type of foo1.y does not itself demand any alignment.

A related feature which lets you specify the alignment of an object is __attribute__ ((aligned (alignment))); see the following section.

Specifying Attributes of Variables

The keyword __attribute__ allows you to specify special attributes of variables or structure fields. This keyword is followed by an attribute specification inside double parentheses. Eight attributes are currently defined for variables: aligned, mode, nocommon, packed, section, transparent_union, unused, and weak. Other attributes are available for functions (see section Declaring Attributes of Functions) and for types (see section Specifying Attributes of Types).

You may also specify attributes with `__' preceding and following each keyword. This allows you to use them in header files without being concerned about a possible macro of the same name. For example, you may use __aligned__ instead of aligned.

aligned (alignment)
This attribute specifies a minimum alignment for the variable or structure field, measured in bytes. For example, the declaration:
int x __attribute__ ((aligned (16))) = 0;
causes the compiler to allocate the global variable x on a 16-byte boundary. On a 68040, this could be used in conjunction with an asm expression to access the move16 instruction which requires 16-byte aligned operands. You can also specify the alignment of structure fields. For example, to create a double-word aligned int pair, you could write:
struct foo { int x[2] __attribute__ ((aligned (8))); };
This is an alternative to creating a union with a double member that forces the union to be double-word aligned. It is not possible to specify the alignment of functions; the alignment of functions is determined by the machine's requirements and cannot be changed. You cannot specify alignment for a typedef name because such a name is just an alias, not a distinct type. As in the preceding examples, you can explicitly specify the alignment (in bytes) that you wish the compiler to use for a given variable or structure field. Alternatively, you can leave out the alignment factor and just ask the compiler to align a variable or field to the maximum useful alignment for the target machine you are compiling for. For example, you could write:
short array[3] __attribute__ ((aligned));
Whenever you leave out the alignment factor in an aligned attribute specification, the compiler automatically sets the alignment for the declared variable or field to the largest alignment which is ever used for any data type on the target machine you are compiling for. Doing this can often make copy operations more efficient, because the compiler can use whatever instructions copy the biggest chunks of memory when performing copies to or from the variables or fields that you have aligned this way. The aligned attribute can only increase the alignment; but you can decrease it by specifying packed as well. See below. Note that the effectiveness of aligned attributes may be limited by inherent limitations in your linker. On many systems, the linker is only able to arrange for variables to be aligned up to a certain maximum alignment. (For some linkers, the maximum supported alignment may be very very small.) If your linker is only able to align variables up to a maximum of 8 byte alignment, then specifying aligned(16) in an __attribute__ will still only provide you with 8 byte alignment. See your linker documentation for further information.
mode (mode)
This attribute specifies the data type for the declaration--whichever type corresponds to the mode mode. This in effect lets you request an integer or floating point type according to its width. You may also specify a mode of `byte' or `__byte__' to indicate the mode corresponding to a one-byte integer, `word' or `__word__' for the mode of a one-word integer, and `pointer' or `__pointer__' for the mode used to represent pointers.
nocommon
This attribute specifies requests GNU CC not to place a variable "common" but instead to allocate space for it directly. If you specify the `-fno-common' flag, GNU CC will do this for all variables. Specifying the nocommon attribute for a variable provides an initialization of zeros. A variable may only be initialized in one source file.
packed
The packed attribute specifies that a variable or structure field should have the smallest possible alignment--one byte for a variable, and one bit for a field, unless you specify a larger value with the aligned attribute. Here is a structure in which the field x is packed, so that it immediately follows a:
struct foo
{
  char a;
  int x[2] __attribute__ ((packed));
};
section ("section-name")
Normally, the compiler places the objects it generates in sections like data and bss. Sometimes, however, you need additional sections, or you need certain particular variables to appear in special sections, for example to map to special hardware. The section attribute specifies that a variable (or function) lives in a particular section. For example, this small program uses several specific section names:
struct duart a __attribute__ ((section ("DUART_A"))) = { 0 };
struct duart b __attribute__ ((section ("DUART_B"))) = { 0 };
char stack[10000] __attribute__ ((section ("STACK"))) = { 0 };
int init_data __attribute__ ((section ("INITDATA"))) = 0;

main()
{
  /* Initialize stack pointer */
  init_sp (stack + sizeof (stack));

  /* Initialize initialized data */
  memcpy (&init_data, &data, &edata - &data);

  /* Turn on the serial ports */
  init_duart (&a);
  init_duart (&b);
}
Use the section attribute with an initialized definition of a global variable, as shown in the example. GNU CC issues a warning and otherwise ignores the section attribute in uninitialized variable declarations. You may only use the section attribute with a fully initialized global definition because of the way linkers work. The linker requires each object be defined once, with the exception that uninitialized variables tentatively go in the common (or bss) section and can be multiply "defined". You can force a variable to be initialized with the `-fno-common' flag or the nocommon attribute. Some file formats do not support arbitrary sections so the section attribute is not available on all platforms. If you need to map the entire contents of a module to a particular section, consider using the facilities of the linker instead.
transparent_union
This attribute, attached to a function parameter which is a union, means that the corresponding argument may have the type of any union member, but the argument is passed as if its type were that of the first union member. For more details see See section Specifying Attributes of Types. You can also use this attribute on a typedef for a union data type; then it applies to all function parameters with that type.
unused
This attribute, attached to a variable, means that the variable is meant to be possibly unused. GNU CC will not produce a warning for this variable.
weak
The weak attribute is described in See section Declaring Attributes of Functions.
model (model-name)
Use this attribute on the M32R/D to set the addressability of an object. The identifier model-name is one of small, medium, or large, representing each of the code models. Small model objects live in the lower 16MB of memory (so that their addresses can be loaded with the ld24 instruction). Medium and large model objects may live anywhere in the 32 bit address space (the compiler will generate seth/add3 instructions to load their addresses).

To specify multiple attributes, separate them by commas within the double parentheses: for example, `__attribute__ ((aligned (16), packed))'.

Specifying Attributes of Types

The keyword __attribute__ allows you to specify special attributes of struct and union types when you define such types. This keyword is followed by an attribute specification inside double parentheses. Three attributes are currently defined for types: aligned, packed, and transparent_union. Other attributes are defined for functions (see section Declaring Attributes of Functions) and for variables (see section Specifying Attributes of Variables).

You may also specify any one of these attributes with `__' preceding and following its keyword. This allows you to use these attributes in header files without being concerned about a possible macro of the same name. For example, you may use __aligned__ instead of aligned.

You may specify the aligned and transparent_union attributes either in a typedef declaration or just past the closing curly brace of a complete enum, struct or union type definition and the packed attribute only past the closing brace of a definition.

You may also specify attributes between the enum, struct or union tag and the name of the type rather than after the closing brace.

aligned (alignment)
This attribute specifies a minimum alignment (in bytes) for variables of the specified type. For example, the declarations:
struct S { short f[3]; } __attribute__ ((aligned (8)));
typedef int more_aligned_int __attribute__ ((aligned (8)));
force the compiler to insure (as far as it can) that each variable whose type is struct S or more_aligned_int will be allocated and aligned at least on a 8-byte boundary. On a Sparc, having all variables of type struct S aligned to 8-byte boundaries allows the compiler to use the ldd and std (doubleword load and store) instructions when copying one variable of type struct S to another, thus improving run-time efficiency. Note that the alignment of any given struct or union type is required by the ANSI C standard to be at least a perfect multiple of the lowest common multiple of the alignments of all of the members of the struct or union in question. This means that you can effectively adjust the alignment of a struct or union type by attaching an aligned attribute to any one of the members of such a type, but the notation illustrated in the example above is a more obvious, intuitive, and readable way to request the compiler to adjust the alignment of an entire struct or union type. As in the preceding example, you can explicitly specify the alignment (in bytes) that you wish the compiler to use for a given struct or union type. Alternatively, you can leave out the alignment factor and just ask the compiler to align a type to the maximum useful alignment for the target machine you are compiling for. For example, you could write:
struct S { short f[3]; } __attribute__ ((aligned));
Whenever you leave out the alignment factor in an aligned attribute specification, the compiler automatically sets the alignment for the type to the largest alignment which is ever used for any data type on the target machine you are compiling for. Doing this can often make copy operations more efficient, because the compiler can use whatever instructions copy the biggest chunks of memory when performing copies to or from the variables which have types that you have aligned this way. In the example above, if the size of each short is 2 bytes, then the size of the entire struct S type is 6 bytes. The smallest power of two which is greater than or equal to that is 8, so the compiler sets the alignment for the entire struct S type to 8 bytes. Note that although you can ask the compiler to select a time-efficient alignment for a given type and then declare only individual stand-alone objects of that type, the compiler's ability to select a time-efficient alignment is primarily useful only when you plan to create arrays of variables having the relevant (efficiently aligned) type. If you declare or use arrays of variables of an efficiently-aligned type, then it is likely that your program will also be doing pointer arithmetic (or subscripting, which amounts to the same thing) on pointers to the relevant type, and the code that the compiler generates for these pointer arithmetic operations will often be more efficient for efficiently-aligned types than for other types. The aligned attribute can only increase the alignment; but you can decrease it by specifying packed as well. See below. Note that the effectiveness of aligned attributes may be limited by inherent limitations in your linker. On many systems, the linker is only able to arrange for variables to be aligned up to a certain maximum alignment. (For some linkers, the maximum supported alignment may be very very small.) If your linker is only able to align variables up to a maximum of 8 byte alignment, then specifying aligned(16) in an __attribute__ will still only provide you with 8 byte alignment. See your linker documentation for further information.
packed
This attribute, attached to an enum, struct, or union type definition, specified that the minimum required memory be used to represent the type. Specifying this attribute for struct and union types is equivalent to specifying the packed attribute on each of the structure or union members. Specifying the `-fshort-enums' flag on the line is equivalent to specifying the packed attribute on all enum definitions. You may only specify this attribute after a closing curly brace on an enum definition, not in a typedef declaration, unless that declaration also contains the definition of the enum.
transparent_union
This attribute, attached to a union type definition, indicates that any function parameter having that union type causes calls to that function to be treated in a special way. First, the argument corresponding to a transparent union type can be of any type in the union; no cast is required. Also, if the union contains a pointer type, the corresponding argument can be a null pointer constant or a void pointer expression; and if the union contains a void pointer type, the corresponding argument can be any pointer expression. If the union member type is a pointer, qualifiers like const on the referenced type must be respected, just as with normal pointer conversions. Second, the argument is passed to the function using the calling conventions of first member of the transparent union, not the calling conventions of the union itself. All members of the union must have the same machine representation; this is necessary for this argument passing to work properly. Transparent unions are designed for library functions that have multiple interfaces for compatibility reasons. For example, suppose the wait function must accept either a value of type int * to comply with Posix, or a value of type union wait * to comply with the 4.1BSD interface. If wait's parameter were void *, wait would accept both kinds of arguments, but it would also accept any other pointer type and this would make argument type checking less useful. Instead, <sys/wait.h> might define the interface as follows:
typedef union
  {
    int *__ip;
    union wait *__up;
  } wait_status_ptr_t __attribute__ ((__transparent_union__));

pid_t wait (wait_status_ptr_t);
This interface allows either int * or union wait * arguments to be passed, using the int * calling convention. The program can call wait with arguments of either type:
int w1 () { int w; return wait (&w); }
int w2 () { union wait w; return wait (&w); }
With this interface, wait's implementation might look like this:
pid_t wait (wait_status_ptr_t p)
{
  return waitpid (-1, p.__ip, 0);
}
unused
When attached to a type (including a union or a struct), this attribute means that variables of that type are meant to appear possibly unused. GNU CC will not produce a warning for any variables of that type, even if the variable appears to do nothing. This is often the case with lock or thread classes, which are usually defined and then not referenced, but contain constructors and destructors that have nontrivial bookkeeping functions.

To specify multiple attributes, separate them by commas within the double parentheses: for example, `__attribute__ ((aligned (16), packed))'.

An Inline Function is As Fast As a Macro

By declaring a function inline, you can direct GNU CC to integrate that function's code into the code for its callers. This makes execution faster by eliminating the function-call overhead; in addition, if any of the actual argument values are constant, their known values may permit simplifications at compile time so that not all of the inline function's code needs to be included. The effect on code size is less predictable; object code may be larger or smaller with function inlining, depending on the particular case. Inlining of functions is an optimization and it really "works" only in optimizing compilation. If you don't use `-O', no function is really inline.

To declare a function inline, use the inline keyword in its declaration, like this:

inline int
inc (int *a)
{
  (*a)++;
}

(If you are writing a header file to be included in ANSI C programs, write __inline__ instead of inline. See section Alternate Keywords.) You can also make all "simple enough" functions inline with the option `-finline-functions'.

Note that certain usages in a function definition can make it unsuitable for inline substitution. Among these usages are: use of varargs, use of alloca, use of variable sized data types (see section Arrays of Variable Length), use of computed goto (see section Labels as Values), use of nonlocal goto, and nested functions (see section Nested Functions). Using `-Winline' will warn when a function marked inline could not be substituted, and will give the reason for the failure.

Note that in C and Objective C, unlike C++, the inline keyword does not affect the linkage of the function.

GNU CC automatically inlines member functions defined within the class body of C++ programs even if they are not explicitly declared inline. (You can override this with `-fno-default-inline'; see section Options Controlling C++ Dialect.)

When a function is both inline and static, if all calls to the function are integrated into the caller, and the function's address is never used, then the function's own assembler code is never referenced. In this case, GNU CC does not actually output assembler code for the function, unless you specify the option `-fkeep-inline-functions'. Some calls cannot be integrated for various reasons (in particular, calls that precede the function's definition cannot be integrated, and neither can recursive calls within the definition). If there is a nonintegrated call, then the function is compiled to assembler code as usual. The function must also be compiled as usual if the program refers to its address, because that can't be inlined.

When an inline function is not static, then the compiler must assume that there may be calls from other source files; since a global symbol can be defined only once in any program, the function must not be defined in the other source files, so the calls therein cannot be integrated. Therefore, a non-static inline function is always compiled on its own in the usual fashion.

If you specify both inline and extern in the function definition, then the definition is used only for inlining. In no case is the function compiled on its own, not even if you refer to its address explicitly. Such an address becomes an external reference, as if you had only declared the function, and had not defined it.

This combination of inline and extern has almost the effect of a macro. The way to use it is to put a function definition in a header file with these keywords, and put another copy of the definition (lacking inline and extern) in a library file. The definition in the header file will cause most calls to the function to be inlined. If any uses of the function remain, they will refer to the single copy in the library.

GNU C does not inline any functions when not optimizing. It is not clear whether it is better to inline or not, in this case, but we found that a correct implementation when not optimizing was difficult. So we did the easy thing, and turned it off.

Assembler Instructions with C Expression Operands

In an assembler instruction using asm, you can specify the operands of the instruction using C expressions. This means you need not guess which registers or memory locations will contain the data you want to use.

You must specify an assembler instruction template much like what appears in a machine description, plus an operand constraint string for each operand.

For example, here is how to use the 68881's fsinx instruction:

asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));

Here angle is the C expression for the input operand while result is that of the output operand. Each has `"f"' as its operand constraint, saying that a floating point register is required. The `=' in `=f' indicates that the operand is an output; all output operands' constraints must use `='. The constraints use the same language used in the machine description (see section Operand Constraints).

Each operand is described by an operand-constraint string followed by the C expression in parentheses. A colon separates the assembler template from the first output operand and another separates the last output operand from the first input, if any. Commas separate the operands within each group. The total number of operands is limited to ten or to the maximum number of operands in any instruction pattern in the machine description, whichever is greater.

If there are no output operands but there are input operands, you must place two consecutive colons surrounding the place where the output operands would go.

Output operand expressions must be lvalues; the compiler can check this. The input operands need not be lvalues. The compiler cannot check whether the operands have data types that are reasonable for the instruction being executed. It does not parse the assembler instruction template and does not know what it means or even whether it is valid assembler input. The extended asm feature is most often used for machine instructions the compiler itself does not know exist. If the output expression cannot be directly addressed (for example, it is a bit field), your constraint must allow a register. In that case, GNU CC will use the register as the output of the asm, and then store that register into the output.

The ordinary output operands must be write-only; GNU CC will assume that the values in these operands before the instruction are dead and need not be generated. Extended asm supports input-output or read-write operands. Use the constraint character `+' to indicate such an operand and list it with the output operands.

When the constraints for the read-write operand (or the operand in which only some of the bits are to be changed) allows a register, you may, as an alternative, logically split its function into two separate operands, one input operand and one write-only output operand. The connection between them is expressed by constraints which say they need to be in the same location when the instruction executes. You can use the same C expression for both operands, or different expressions. For example, here we write the (fictitious) `combine' instruction with bar as its read-only source operand and foo as its read-write destination:

asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));

The constraint `"0"' for operand 1 says that it must occupy the same location as operand 0. A digit in constraint is allowed only in an input operand and it must refer to an output operand.

Only a digit in the constraint can guarantee that one operand will be in the same place as another. The mere fact that foo is the value of both operands is not enough to guarantee that they will be in the same place in the generated assembler code. The following would not work reliably:

asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));

Various optimizations or reloading could cause operands 0 and 1 to be in different registers; GNU CC knows no reason not to do so. For example, the compiler might find a copy of the value of foo in one register and use it for operand 1, but generate the output operand 0 in a different register (copying it afterward to foo's own address). Of course, since the register for operand 1 is not even mentioned in the assembler code, the result will not work, but GNU CC can't tell that.

Some instructions clobber specific hard registers. To describe this, write a third colon after the input operands, followed by the names of the clobbered hard registers (given as strings). Here is a realistic example for the VAX:

asm volatile ("movc3 %0,%1,%2"
              : /* no outputs */
              : "g" (from), "g" (to), "g" (count)
              : "r0", "r1", "r2", "r3", "r4", "r5");

It is an error for a clobber description to overlap an input or output operand (for example, an operand describing a register class with one member, mentioned in the clobber list). Most notably, it is invalid to describe that an input operand is modified, but unused as output. It has to be specified as an input and output operand anyway. Note that if there are only unused output operands, you will then also need to specify volatile for the asm construct, as described below.

If you refer to a particular hardware register from the assembler code, you will probably have to list the register after the third colon to tell the compiler the register's value is modified. In some assemblers, the register names begin with `%'; to produce one `%' in the assembler code, you must write `%%' in the input.

If your assembler instruction can alter the condition code register, add `cc' to the list of clobbered registers. GNU CC on some machines represents the condition codes as a specific hardware register; `cc' serves to name this register. On other machines, the condition code is handled differently, and specifying `cc' has no effect. But it is valid no matter what the machine.

If your assembler instruction modifies memory in an unpredictable fashion, add `memory' to the list of clobbered registers. This will cause GNU CC to not keep memory values cached in registers across the assembler instruction.

You can put multiple assembler instructions together in a single asm template, separated either with newlines (written as `\n') or with semicolons if the assembler allows such semicolons. The GNU assembler allows semicolons and most Unix assemblers seem to do so. The input operands are guaranteed not to use any of the clobbered registers, and neither will the output operands' addresses, so you can read and write the clobbered registers as many times as you like. Here is an example of multiple instructions in a template; it assumes the subroutine _foo accepts arguments in registers 9 and 10:

asm ("movl %0,r9;movl %1,r10;call _foo"
     : /* no outputs */
     : "g" (from), "g" (to)
     : "r9", "r10");

Unless an output operand has the `&' constraint modifier, GNU CC may allocate it in the same register as an unrelated input operand, on the assumption the inputs are consumed before the outputs are produced. This assumption may be false if the assembler code actually consists of more than one instruction. In such a case, use `&' for each output operand that may not overlap an input. See section Constraint Modifier Characters.

If you want to test the condition code produced by an assembler instruction, you must include a branch and a label in the asm construct, as follows:

asm ("clr %0;frob %1;beq 0f;mov #1,%0;0:"
     : "g" (result)
     : "g" (input));

This assumes your assembler supports local labels, as the GNU assembler and most Unix assemblers do.

Speaking of labels, jumps from one asm to another are not supported. The compiler's optimizers do not know about these jumps, and therefore they cannot take account of them when deciding how to optimize.

Usually the most convenient way to use these asm instructions is to encapsulate them in macros that look like functions. For example,

#define sin(x)       \
({ double __value, __arg = (x);   \
   asm ("fsinx %1,%0": "=f" (__value): "f" (__arg));  \
   __value; })

Here the variable __arg is used to make sure that the instruction operates on a proper double value, and to accept only those arguments x which can convert automatically to a double.

Another way to make sure the instruction operates on the correct data type is to use a cast in the asm. This is different from using a variable __arg in that it converts more different types. For example, if the desired type were int, casting the argument to int would accept a pointer with no complaint, while assigning the argument to an int variable named __arg would warn about using a pointer unless the caller explicitly casts it.

If an asm has output operands, GNU CC assumes for optimization purposes the instruction has no side effects except to change the output operands. This does not mean instructions with a side effect cannot be used, but you must be careful, because the compiler may eliminate them if the output operands aren't used, or move them out of loops, or replace two with one if they constitute a common subexpression. Also, if your instruction does have a side effect on a variable that otherwise appears not to change, the old value of the variable may be reused later if it happens to be found in a register.

You can prevent an asm instruction from being deleted, moved significantly, or combined, by writing the keyword volatile after the asm. For example:

#define get_and_set_priority(new)  \
({ int __old; \
   asm volatile ("get_and_set_priority %0, %1": "=g" (__old) : "g" (new)); \
   __old; })

If you write an asm instruction with no outputs, GNU CC will know the instruction has side-effects and will not delete the instruction or move it outside of loops. If the side-effects of your instruction are not purely external, but will affect variables in your program in ways other than reading the inputs and clobbering the specified registers or memory, you should write the volatile keyword to prevent future versions of GNU CC from moving the instruction around within a core region.

An asm instruction without any operands or clobbers (and "old style" asm) will not be deleted or moved significantly, regardless, unless it is unreachable, the same wasy as if you had written a volatile keyword.

Note that even a volatile asm instruction can be moved in ways that appear insignificant to the compiler, such as across jump instructions. You can't expect a sequence of volatile asm instructions to remain perfectly consecutive. If you want consecutive output, use a single asm.

It is a natural idea to look for a way to give access to the condition code left by the assembler instruction. However, when we attempted to implement this, we found no way to make it work reliably. The problem is that output operands might need reloading, which would result in additional following "store" instructions. On most machines, these instructions would alter the condition code before there was time to test it. This problem doesn't arise for ordinary "test" and "compare" instructions because they don't have any output operands.

If you are writing a header file that should be includable in ANSI C programs, write __asm__ instead of asm. See section Alternate Keywords.

i386 floating point asm operands

There are several rules on the usage of stack-like regs in asm_operands insns. These rules apply only to the operands that are stack-like regs:

  1. Given a set of input regs that die in an asm_operands, it is necessary to know which are implicitly popped by the asm, and which must be explicitly popped by gcc. An input reg that is implicitly popped by the asm must be explicitly clobbered, unless it is constrained to match an output operand.
  2. For any input reg that is implicitly popped by an asm, it is necessary to know how to adjust the stack to compensate for the pop. If any non-popped input is closer to the top of the reg-stack than the implicitly popped reg, it would not be possible to know what the stack looked like -- it's not clear how the rest of the stack "slides up". All implicitly popped input regs must be closer to the top of the reg-stack than any input that is not implicitly popped. It is possible that if an input dies in an insn, reload might use the input reg for an output reload. Consider this example:
    asm ("foo" : "=t" (a) : "f" (b));
    
    This asm says that input B is not popped by the asm, and that the asm pushes a result onto the reg-stack, ie, the stack is one deeper after the asm than it was before. But, it is possible that reload will think that it can use the same reg for both the input and the output, if input B dies in this insn. If any input operand uses the f constraint, all output reg constraints must use the & earlyclobber. The asm above would be written as
    asm ("foo" : "=&t" (a) : "f" (b));
    
  3. Some operands need to be in particular places on the stack. All output operands fall in this category -- there is no other way to know which regs the outputs appear in unless the user indicates this in the constraints. Output operands must specifically indicate which reg an output appears in after an asm. =f is not allowed: the operand constraints must select a class with a single reg.
  4. Output operands may not be "inserted" between existing stack regs. Since no 387 opcode uses a read/write operand, all output operands are dead before the asm_operands, and are pushed by the asm_operands. It makes no sense to push anywhere but the top of the reg-stack. Output operands must start at the top of the reg-stack: output operands may not "skip" a reg.
  5. Some asm statements may need extra stack space for internal calculations. This can be guaranteed by clobbering stack registers unrelated to the inputs and outputs.

Here are a couple of reasonable asms to want to write. This asm takes one input, which is internally popped, and produces two outputs.

asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));

This asm takes two inputs, which are popped by the fyl2xp1 opcode, and replaces them with one output. The user must code the st(1) clobber for reg-stack.c to know that fyl2xp1 pops both inputs.

asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");

Controlling Names Used in Assembler Code

You can specify the name to be used in the assembler code for a C function or variable by writing the asm (or __asm__) keyword after the declarator as follows:

int foo asm ("myfoo") = 2;

This specifies that the name to be used for the variable foo in the assembler code should be `myfoo' rather than the usual `_foo'.

On systems where an underscore is normally prepended to the name of a C function or variable, this feature allows you to define names for the linker that do not start with an underscore.

You cannot use asm in this way in a function definition; but you can get the same effect by writing a declaration for the function before its definition and putting asm there, like this:

extern func () asm ("FUNC");

func (x, y)
     int x, y;
...

It is up to you to make sure that the assembler names you choose do not conflict with any other assembler symbols. Also, you must not use a register name; that would produce completely invalid assembler code. GNU CC does not as yet have the ability to store static variables in registers. Perhaps that will be added.

Variables in Specified Registers

GNU C allows you to put a few global variables into specified hardware registers. You can also specify the register in which an ordinary register variable should be allocated.

Defining Global Register Variables

You can define a global register variable in GNU C like this:

register int *foo asm ("a5");

Here a5 is the name of the register which should be used. Choose a register which is normally saved and restored by function calls on your machine, so that library routines will not clobber it.

Naturally the register name is cpu-dependent, so you would need to conditionalize your program according to cpu type. The register a5 would be a good choice on a 68000 for a variable of pointer type. On machines with register windows, be sure to choose a "global" register that is not affected magically by the function call mechanism.

In addition, operating systems on one type of cpu may differ in how they name the registers; then you would need additional conditionals. For example, some 68000 operating systems call this register %a5.

Eventually there may be a way of asking the compiler to choose a register automatically, but first we need to figure out how it should choose and how to enable you to guide the choice. No solution is evident.

Defining a global register variable in a certain register reserves that register entirely for this use, at least within the current compilation. The register will not be allocated for any other purpose in the functions in the current compilation. The register will not be saved and restored by these functions. Stores into this register are never deleted even if they would appear to be dead, but references may be deleted or moved or simplified.

It is not safe to access the global register variables from signal handlers, or from more than one thread of control, because the system library routines may temporarily use the register for other things (unless you recompile them specially for the task at hand).

It is not safe for one function that uses a global register variable to call another such function foo by way of a third function lose that was compiled without knowledge of this variable (i.e. in a different source file in which the variable wasn't declared). This is because lose might save the register and put some other value there. For example, you can't expect a global register variable to be available in the comparison-function that you pass to qsort, since qsort might have put something else in that register. (If you are prepared to recompile qsort with the same global register variable, you can solve this problem.)

If you want to recompile qsort or other source files which do not actually use your global register variable, so that they will not use that register for any other purpose, then it suffices to specify the compiler option `-ffixed-reg'. You need not actually add a global register declaration to their source code.

A function which can alter the value of a global register variable cannot safely be called from a function compiled without this variable, because it could clobber the value the caller expects to find there on return. Therefore, the function which is the entry point into the part of the program that uses the global register variable must explicitly save and restore the value which belongs to its caller.

On most machines, longjmp will restore to each global register variable the value it had at the time of the setjmp. On some machines, however, longjmp will not change the value of global register variables. To be portable, the function that called setjmp should make other arrangements to save the values of the global register variables, and to restore them in a longjmp. This way, the same thing will happen regardless of what longjmp does.

All global register variable declarations must precede all function definitions. If such a declaration could appear after function definitions, the declaration would be too late to prevent the register from being used for other purposes in the preceding functions.

Global register variables may not have initial values, because an executable file has no means to supply initial contents for a register.

On the Sparc, there are reports that g3 ... g7 are suitable registers, but certain library functions, such as getwd, as well as the subroutines for division and remainder, modify g3 and g4. g1 and g2 are local temporaries.

On the 68000, a2 ... a5 should be suitable, as should d2 ... d7. Of course, it will not do to use more than a few of those.

Specifying Registers for Local Variables

You can define a local register variable with a specified register like this:

register int *foo asm ("a5");

Here a5 is the name of the register which should be used. Note that this is the same syntax used for defining global register variables, but for a local variable it would appear within a function.

Naturally the register name is cpu-dependent, but this is not a problem, since specific registers are most often useful with explicit assembler instructions (see section Assembler Instructions with C Expression Operands). Both of these things generally require that you conditionalize your program according to cpu type.

In addition, operating systems on one type of cpu may differ in how they name the registers; then you would need additional conditionals. For example, some 68000 operating systems call this register %a5.

Defining such a register variable does not reserve the register; it remains available for other uses in places where flow control determines the variable's value is not live. However, these registers are made unavailable for use in the reload pass; excessive use of this feature leaves the compiler too few available registers to compile certain functions.

This option does not guarantee that GNU CC will generate code that has this variable in the register you specify at all times. You may not code an explicit reference to this register in an asm statement and assume it will always refer to this variable.

Stores into local register variables may be deleted when they appear to be dead according to dataflow analysis. References to local register variables may be deleted or moved or simplified.

Alternate Keywords

The option `-traditional' disables certain keywords; `-ansi' disables certain others. This causes trouble when you want to use GNU C extensions, or ANSI C features, in a general-purpose header file that should be usable by all programs, including ANSI C programs and traditional ones. The keywords asm, typeof and inline cannot be used since they won't work in a program compiled with `-ansi', while the keywords const, volatile, signed, typeof and inline won't work in a program compiled with `-traditional'.

The way to solve these problems is to put `__' at the beginning and end of each problematical keyword. For example, use __asm__ instead of asm, __const__ instead of const, and __inline__ instead of inline.

Other C compilers won't accept these alternative keywords; if you want to compile with another compiler, you can define the alternate keywords as macros to replace them with the customary keywords. It looks like this:

#ifndef __GNUC__
#define __asm__ asm
#endif

`-pedantic' causes warnings for many GNU C extensions. You can prevent such warnings within one expression by writing __extension__ before the expression. __extension__ has no effect aside from this.

Incomplete enum Types

You can define an enum tag without specifying its possible values. This results in an incomplete type, much like what you get if you write struct foo without describing the elements. A later declaration which does specify the possible values completes the type.

You can't allocate variables or storage using the type while it is incomplete. However, you can work with pointers to that type.

This extension may not be very useful, but it makes the handling of enum more consistent with the way struct and union are handled.

This extension is not supported by GNU C++.

Function Names as Strings

GNU CC predefines two string variables to be the name of the current function. The variable __FUNCTION__ is the name of the function as it appears in the source. The variable __PRETTY_FUNCTION__ is the name of the function pretty printed in a language specific fashion.

These names are always the same in a C function, but in a C++ function they may be different. For example, this program:

extern "C" {
extern int printf (char *, ...);
}

class a {
 public:
  sub (int i)
    {
      printf ("__FUNCTION__ = %s\n", __FUNCTION__);
      printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
    }
};

int
main (void)
{
  a ax;
  ax.sub (0);
  return 0;
}

gives this output:

__FUNCTION__ = sub
__PRETTY_FUNCTION__ = int  a::sub (int)

These names are not macros: they are predefined string variables. For example, `#ifdef __FUNCTION__' does not have any special meaning inside a function, since the preprocessor does not do anything special with the identifier __FUNCTION__.

Getting the Return or Frame Address of a Function

These functions may be used to get information about the callers of a function.

__builtin_return_address (level)
This function returns the return address of the current function, or of one of its callers. The level argument is number of frames to scan up the call stack. A value of 0 yields the return address of the current function, a value of 1 yields the return address of the caller of the current function, and so forth. The level argument must be a constant integer. On some machines it may be impossible to determine the return address of any function other than the current one; in such cases, or when the top of the stack has been reached, this function will return 0. This function should only be used with a non-zero argument for debugging purposes.
__builtin_frame_address (level)
This function is similar to __builtin_return_address, but it returns the address of the function frame rather than the return address of the function. Calling __builtin_frame_address with a value of 0 yields the frame address of the current function, a value of 1 yields the frame address of the caller of the current function, and so forth. The frame is the area on the stack which holds local variables and saved registers. The frame address is normally the address of the first word pushed on to the stack by the function. However, the exact definition depends upon the processor and the calling convention. If the processor has a dedicated frame pointer register, and the function has a frame, then __builtin_frame_address will return the value of the frame pointer register. The caveats that apply to __builtin_return_address apply to this function as well.

Other built-in functions provided by GNU CC

GNU CC provides a large number of built-in functions other than the ones mentioned above. Some of these are for internal use in the processing of exceptions or variable-length argument lists and will not be documented here because they may change from time to time; we do not recommend general use of these functions.

The remaining functions are provided for optimization purposes.

GNU CC includes builtin versions of many of the functions in the standard C library. These will always be treated as having the same meaning as the C library function even if you specify the `-fno-builtin' (see section Options Controlling C Dialect) option. These functions correspond to the C library functions alloca, ffs, abs, fabsf, fabs, fabsl, labs, memcpy, memcmp, strcmp, strcpy, strlen, sqrtf, sqrt, sqrtl, sinf, sin, sinl, cosf, cos, and cosl.

You can use the builtin function __builtin_constant_p to determine if a value is known to be constant at compile-time and hence that GNU CC can perform constant-folding on expressions involving that value. The argument of the function is the value to test. The function returns the integer 1 if the argument is known to be a compile-time constant and 0 if it is not known to be a compile-time constant. A return of 0 does not indicate that the value is not a constant, but merely that GNU CC cannot prove it is a constant with the specified value of the `-O' option.

You would typically use this function in an embedded application where memory was a critical resource. If you have some complex calculation, you may want it to be folded if it involves constants, but need to call a function if it does not. For example:

#define Scale_Value(X)  \
  (__builtin_constant_p (X) ? ((X) * SCALE + OFFSET) : Scale (X))

You may use this builtin function in either a macro or an inline function. However, if you use it in an inlined function and pass an argument of the function as the argument to the builtin, GNU CC will never return 1 when you call the inline function with a string constant or constructor expression (see section Constructor Expressions) and will not return 1 when you pass a constant numeric value to the inline function unless you specify the `-O' option.

Deprecated Features

In the past, the GNU C++ compiler was extended to experiment with new features, at a time when the C++ language was still evolving. Now that the C++ standard is complete, some of those features are superceded by superior alternatives. Using the old features might cause a warning in some cases that the feature will be dropped in the future. In other cases, the feature might be gone already.

While the list below is not exhaustive, it documents some of the options that are now deprecated:

-fthis-is-variable
In early versions of C++, assignment to this could be used to implement application-defined memory allocation. Now, allocation functions (`operator new') are the standard-conforming way to achieve the same effect.
-fexternal-templates
-falt-external-templates
These are two of the many ways for g++ to implement template instantiation. See section Where's the Template?. The C++ standard clearly defines how template definitions have to be organized across implementation units. g++ has an implicit instantiation mechanism that should work just fine for standard-conforming code.

Go to the first, previous, next, last section, table of contents.