1 //==--- AttrDocs.td - Attribute documentation ----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===---------------------------------------------------------------------===//
10 def GlobalDocumentation {
12 -------------------------------------------------------------------
13 NOTE: This file is automatically generated by running clang-tblgen
14 -gen-attr-docs. Do not edit this file by hand!!
15 -------------------------------------------------------------------
26 This page lists the attributes currently supported by Clang.
30 def SectionDocs : Documentation {
31 let Category = DocCatVariable;
33 The ``section`` attribute allows you to specify a specific section a
34 global variable or function should be in after translation.
36 let Heading = "section (gnu::section, __declspec(allocate))";
39 def InitSegDocs : Documentation {
40 let Category = DocCatVariable;
42 The attribute applied by ``pragma init_seg()`` controls the section into
43 which global initialization function pointers are emitted. It is only
44 available with ``-fms-extensions``. Typically, this function pointer is
45 emitted into ``.CRT$XCU`` on Windows. The user can change the order of
46 initialization by using a different section name with the same
47 ``.CRT$XC`` prefix and a suffix that sorts lexicographically before or
48 after the standard ``.CRT$XCU`` sections. See the init_seg_
49 documentation on MSDN for more information.
51 .. _init_seg: http://msdn.microsoft.com/en-us/library/7977wcck(v=vs.110).aspx
55 def TLSModelDocs : Documentation {
56 let Category = DocCatVariable;
58 The ``tls_model`` attribute allows you to specify which thread-local storage
59 model to use. It accepts the following strings:
66 TLS models are mutually exclusive.
70 def DLLExportDocs : Documentation {
71 let Category = DocCatVariable;
73 The ``__declspec(dllexport)`` attribute declares a variable, function, or
74 Objective-C interface to be exported from the module. It is available under the
75 ``-fdeclspec`` flag for compatibility with various compilers. The primary use
76 is for COFF object files which explicitly specify what interfaces are available
77 for external use. See the dllexport_ documentation on MSDN for more
80 .. _dllexport: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
84 def DLLImportDocs : Documentation {
85 let Category = DocCatVariable;
87 The ``__declspec(dllimport)`` attribute declares a variable, function, or
88 Objective-C interface to be imported from an external module. It is available
89 under the ``-fdeclspec`` flag for compatibility with various compilers. The
90 primary use is for COFF object files which explicitly specify what interfaces
91 are imported from external modules. See the dllimport_ documentation on MSDN
94 .. _dllimport: https://msdn.microsoft.com/en-us/library/3y1sfaz2.aspx
98 def ThreadDocs : Documentation {
99 let Category = DocCatVariable;
101 The ``__declspec(thread)`` attribute declares a variable with thread local
102 storage. It is available under the ``-fms-extensions`` flag for MSVC
103 compatibility. See the documentation for `__declspec(thread)`_ on MSDN.
105 .. _`__declspec(thread)`: http://msdn.microsoft.com/en-us/library/9w1sdazb.aspx
107 In Clang, ``__declspec(thread)`` is generally equivalent in functionality to the
108 GNU ``__thread`` keyword. The variable must not have a destructor and must have
109 a constant initializer, if any. The attribute only applies to variables
110 declared with static storage duration, such as globals, class static data
111 members, and static locals.
115 def CarriesDependencyDocs : Documentation {
116 let Category = DocCatFunction;
118 The ``carries_dependency`` attribute specifies dependency propagation into and
121 When specified on a function or Objective-C method, the ``carries_dependency``
122 attribute means that the return value carries a dependency out of the function,
123 so that the implementation need not constrain ordering upon return from that
124 function. Implementations of the function and its caller may choose to preserve
125 dependencies instead of emitting memory ordering instructions such as fences.
127 Note, this attribute does not change the meaning of the program, but may result
128 in generation of more efficient code.
132 def C11NoReturnDocs : Documentation {
133 let Category = DocCatFunction;
135 A function declared as ``_Noreturn`` shall not return to its caller. The
136 compiler will generate a diagnostic for a function declared as ``_Noreturn``
137 that appears to be capable of returning to its caller.
141 def CXX11NoReturnDocs : Documentation {
142 let Category = DocCatFunction;
144 A function declared as ``[[noreturn]]`` shall not return to its caller. The
145 compiler will generate a diagnostic for a function declared as ``[[noreturn]]``
146 that appears to be capable of returning to its caller.
150 def AssertCapabilityDocs : Documentation {
151 let Category = DocCatFunction;
152 let Heading = "assert_capability (assert_shared_capability, clang::assert_capability, clang::assert_shared_capability)";
154 Marks a function that dynamically tests whether a capability is held, and halts
155 the program if it is not held.
159 def AcquireCapabilityDocs : Documentation {
160 let Category = DocCatFunction;
161 let Heading = "acquire_capability (acquire_shared_capability, clang::acquire_capability, clang::acquire_shared_capability)";
163 Marks a function as acquiring a capability.
167 def TryAcquireCapabilityDocs : Documentation {
168 let Category = DocCatFunction;
169 let Heading = "try_acquire_capability (try_acquire_shared_capability, clang::try_acquire_capability, clang::try_acquire_shared_capability)";
171 Marks a function that attempts to acquire a capability. This function may fail to
172 actually acquire the capability; they accept a Boolean value determining
173 whether acquiring the capability means success (true), or failing to acquire
174 the capability means success (false).
178 def ReleaseCapabilityDocs : Documentation {
179 let Category = DocCatFunction;
180 let Heading = "release_capability (release_shared_capability, clang::release_capability, clang::release_shared_capability)";
182 Marks a function as releasing a capability.
186 def AssumeAlignedDocs : Documentation {
187 let Category = DocCatFunction;
189 Use ``__attribute__((assume_aligned(<alignment>[,<offset>]))`` on a function
190 declaration to specify that the return value of the function (which must be a
191 pointer type) has the specified offset, in bytes, from an address with the
192 specified alignment. The offset is taken to be zero if omitted.
196 // The returned pointer value has 32-byte alignment.
197 void *a() __attribute__((assume_aligned (32)));
199 // The returned pointer value is 4 bytes greater than an address having
200 // 32-byte alignment.
201 void *b() __attribute__((assume_aligned (32, 4)));
203 Note that this attribute provides information to the compiler regarding a
204 condition that the code already ensures is true. It does not cause the compiler
205 to enforce the provided alignment assumption.
209 def AllocSizeDocs : Documentation {
210 let Category = DocCatFunction;
212 The ``alloc_size`` attribute can be placed on functions that return pointers in
213 order to hint to the compiler how many bytes of memory will be available at the
214 returned poiner. ``alloc_size`` takes one or two arguments.
216 - ``alloc_size(N)`` implies that argument number N equals the number of
217 available bytes at the returned pointer.
218 - ``alloc_size(N, M)`` implies that the product of argument number N and
219 argument number M equals the number of available bytes at the returned
222 Argument numbers are 1-based.
224 An example of how to use ``alloc_size``
228 void *my_malloc(int a) __attribute__((alloc_size(1)));
229 void *my_calloc(int a, int b) __attribute__((alloc_size(1, 2)));
232 void *const p = my_malloc(100);
233 assert(__builtin_object_size(p, 0) == 100);
234 void *const a = my_calloc(20, 5);
235 assert(__builtin_object_size(a, 0) == 100);
238 .. Note:: This attribute works differently in clang than it does in GCC.
239 Specifically, clang will only trace ``const`` pointers (as above); we give up
240 on pointers that are not marked as ``const``. In the vast majority of cases,
241 this is unimportant, because LLVM has support for the ``alloc_size``
242 attribute. However, this may cause mildly unintuitive behavior when used with
243 other attributes, such as ``enable_if``.
247 def EnableIfDocs : Documentation {
248 let Category = DocCatFunction;
250 .. Note:: Some features of this attribute are experimental. The meaning of
251 multiple enable_if attributes on a single declaration is subject to change in
252 a future version of clang. Also, the ABI is not standardized and the name
253 mangling may change in future versions. To avoid that, use asm labels.
255 The ``enable_if`` attribute can be placed on function declarations to control
256 which overload is selected based on the values of the function's arguments.
257 When combined with the ``overloadable`` attribute, this feature is also
263 int isdigit(int c) __attribute__((enable_if(c <= -1 || c > 255, "chosen when 'c' is out of range"))) __attribute__((unavailable("'c' must have the value of an unsigned char or EOF")));
268 isdigit(-10); // results in a compile-time error.
271 The enable_if attribute takes two arguments, the first is an expression written
272 in terms of the function parameters, the second is a string explaining why this
273 overload candidate could not be selected to be displayed in diagnostics. The
274 expression is part of the function signature for the purposes of determining
275 whether it is a redeclaration (following the rules used when determining
276 whether a C++ template specialization is ODR-equivalent), but is not part of
279 The enable_if expression is evaluated as if it were the body of a
280 bool-returning constexpr function declared with the arguments of the function
281 it is being applied to, then called with the parameters at the call site. If the
282 result is false or could not be determined through constant expression
283 evaluation, then this overload will not be chosen and the provided string may
284 be used in a diagnostic if the compile fails as a result.
286 Because the enable_if expression is an unevaluated context, there are no global
287 state changes, nor the ability to pass information from the enable_if
288 expression to the function body. For example, suppose we want calls to
289 strnlen(strbuf, maxlen) to resolve to strnlen_chk(strbuf, maxlen, size of
290 strbuf) only if the size of strbuf can be determined:
294 __attribute__((always_inline))
295 static inline size_t strnlen(const char *s, size_t maxlen)
296 __attribute__((overloadable))
297 __attribute__((enable_if(__builtin_object_size(s, 0) != -1))),
298 "chosen when the buffer size is known but 'maxlen' is not")))
300 return strnlen_chk(s, maxlen, __builtin_object_size(s, 0));
303 Multiple enable_if attributes may be applied to a single declaration. In this
304 case, the enable_if expressions are evaluated from left to right in the
305 following manner. First, the candidates whose enable_if expressions evaluate to
306 false or cannot be evaluated are discarded. If the remaining candidates do not
307 share ODR-equivalent enable_if expressions, the overload resolution is
308 ambiguous. Otherwise, enable_if overload resolution continues with the next
309 enable_if attribute on the candidates that have not been discarded and have
310 remaining enable_if attributes. In this way, we pick the most specific
311 overload out of a number of viable overloads using enable_if.
315 void f() __attribute__((enable_if(true, ""))); // #1
316 void f() __attribute__((enable_if(true, ""))) __attribute__((enable_if(true, ""))); // #2
318 void g(int i, int j) __attribute__((enable_if(i, ""))); // #1
319 void g(int i, int j) __attribute__((enable_if(j, ""))) __attribute__((enable_if(true))); // #2
321 In this example, a call to f() is always resolved to #2, as the first enable_if
322 expression is ODR-equivalent for both declarations, but #1 does not have another
323 enable_if expression to continue evaluating, so the next round of evaluation has
324 only a single candidate. In a call to g(1, 1), the call is ambiguous even though
325 #2 has more enable_if attributes, because the first enable_if expressions are
328 Query for this feature with ``__has_attribute(enable_if)``.
330 Note that functions with one or more ``enable_if`` attributes may not have
331 their address taken, unless all of the conditions specified by said
332 ``enable_if`` are constants that evaluate to ``true``. For example:
336 const int TrueConstant = 1;
337 const int FalseConstant = 0;
338 int f(int a) __attribute__((enable_if(a > 0, "")));
339 int g(int a) __attribute__((enable_if(a == 0 || a != 0, "")));
340 int h(int a) __attribute__((enable_if(1, "")));
341 int i(int a) __attribute__((enable_if(TrueConstant, "")));
342 int j(int a) __attribute__((enable_if(FalseConstant, "")));
346 ptr = &f; // error: 'a > 0' is not always true
347 ptr = &g; // error: 'a == 0 || a != 0' is not a truthy constant
348 ptr = &h; // OK: 1 is a truthy constant
349 ptr = &i; // OK: 'TrueConstant' is a truthy constant
350 ptr = &j; // error: 'FalseConstant' is a constant, but not truthy
353 Because ``enable_if`` evaluation happens during overload resolution,
354 ``enable_if`` may give unintuitive results when used with templates, depending
355 on when overloads are resolved. In the example below, clang will emit a
356 diagnostic about no viable overloads for ``foo`` in ``bar``, but not in ``baz``:
360 double foo(int i) __attribute__((enable_if(i > 0, "")));
361 void *foo(int i) __attribute__((enable_if(i <= 0, "")));
363 auto bar() { return foo(I); }
365 template <typename T>
366 auto baz() { return foo(T::number); }
368 struct WithNumber { constexpr static int number = 1; };
370 bar<sizeof(WithNumber)>();
374 This is because, in ``bar``, ``foo`` is resolved prior to template
375 instantiation, so the value for ``I`` isn't known (thus, both ``enable_if``
376 conditions for ``foo`` fail). However, in ``baz``, ``foo`` is resolved during
377 template instantiation, so the value for ``T::number`` is known.
381 def DiagnoseIfDocs : Documentation {
382 let Category = DocCatFunction;
384 The ``diagnose_if`` attribute can be placed on function declarations to emit
385 warnings or errors at compile-time if calls to the attributed function meet
386 certain user-defined criteria. For example:
391 __attribute__((diagnose_if(a >= 0, "Redundant abs call", "warning")));
393 __attribute__((diagnose_if(a >= 0, "Redundant abs call", "error")));
395 int val = abs(1); // warning: Redundant abs call
396 int val2 = must_abs(1); // error: Redundant abs call
398 int val4 = must_abs(val); // Because run-time checks are not emitted for
399 // diagnose_if attributes, this executes without
403 ``diagnose_if`` is closely related to ``enable_if``, with a few key differences:
405 * Overload resolution is not aware of ``diagnose_if`` attributes: they're
406 considered only after we select the best candidate from a given candidate set.
407 * Function declarations that differ only in their ``diagnose_if`` attributes are
408 considered to be redeclarations of the same function (not overloads).
409 * If the condition provided to ``diagnose_if`` cannot be evaluated, no
410 diagnostic will be emitted.
412 Otherwise, ``diagnose_if`` is essentially the logical negation of ``enable_if``.
414 As a result of bullet number two, ``diagnose_if`` attributes will stack on the
415 same function. For example:
419 int foo() __attribute__((diagnose_if(1, "diag1", "warning")));
420 int foo() __attribute__((diagnose_if(1, "diag2", "warning")));
422 int bar = foo(); // warning: diag1
424 int (*fooptr)(void) = foo; // warning: diag1
427 constexpr int supportsAPILevel(int N) { return N < 5; }
429 __attribute__((diagnose_if(!supportsAPILevel(10),
430 "Upgrade to API level 10 to use baz", "error")));
432 __attribute__((diagnose_if(!a, "0 is not recommended.", "warning")));
434 int (*bazptr)(int) = baz; // error: Upgrade to API level 10 to use baz
435 int v = baz(0); // error: Upgrade to API level 10 to use baz
437 Query for this feature with ``__has_attribute(diagnose_if)``.
441 def PassObjectSizeDocs : Documentation {
442 let Category = DocCatVariable; // Technically it's a parameter doc, but eh.
444 .. Note:: The mangling of functions with parameters that are annotated with
445 ``pass_object_size`` is subject to change. You can get around this by
446 using ``__asm__("foo")`` to explicitly name your functions, thus preserving
447 your ABI; also, non-overloadable C functions with ``pass_object_size`` are
450 The ``pass_object_size(Type)`` attribute can be placed on function parameters to
451 instruct clang to call ``__builtin_object_size(param, Type)`` at each callsite
452 of said function, and implicitly pass the result of this call in as an invisible
453 argument of type ``size_t`` directly after the parameter annotated with
454 ``pass_object_size``. Clang will also replace any calls to
455 ``__builtin_object_size(param, Type)`` in the function by said implicit
462 int bzero1(char *const p __attribute__((pass_object_size(0))))
463 __attribute__((noinline)) {
465 for (/**/; i < (int)__builtin_object_size(p, 0); ++i) {
473 int n = bzero1(&chars[0]);
474 assert(n == sizeof(chars));
478 If successfully evaluating ``__builtin_object_size(param, Type)`` at the
479 callsite is not possible, then the "failed" value is passed in. So, using the
480 definition of ``bzero1`` from above, the following code would exit cleanly:
484 int main2(int argc, char *argv[]) {
485 int n = bzero1(argv);
490 ``pass_object_size`` plays a part in overload resolution. If two overload
491 candidates are otherwise equally good, then the overload with one or more
492 parameters with ``pass_object_size`` is preferred. This implies that the choice
493 between two identical overloads both with ``pass_object_size`` on one or more
494 parameters will always be ambiguous; for this reason, having two such overloads
495 is illegal. For example:
499 #define PS(N) __attribute__((pass_object_size(N)))
501 void Foo(char *a, char *b); // Overload A
502 // OK -- overload A has no parameters with pass_object_size.
503 void Foo(char *a PS(0), char *b PS(0)); // Overload B
504 // Error -- Same signature (sans pass_object_size) as overload B, and both
505 // overloads have one or more parameters with the pass_object_size attribute.
506 void Foo(void *a PS(0), void *b);
509 void Bar(void *a PS(0)); // Overload C
511 void Bar(char *c PS(1)); // Overload D
514 char known[10], *unknown;
515 Foo(unknown, unknown); // Calls overload B
516 Foo(known, unknown); // Calls overload B
517 Foo(unknown, known); // Calls overload B
518 Foo(known, known); // Calls overload B
520 Bar(known); // Calls overload D
521 Bar(unknown); // Calls overload D
524 Currently, ``pass_object_size`` is a bit restricted in terms of its usage:
526 * Only one use of ``pass_object_size`` is allowed per parameter.
528 * It is an error to take the address of a function with ``pass_object_size`` on
529 any of its parameters. If you wish to do this, you can create an overload
530 without ``pass_object_size`` on any parameters.
532 * It is an error to apply the ``pass_object_size`` attribute to parameters that
533 are not pointers. Additionally, any parameter that ``pass_object_size`` is
534 applied to must be marked ``const`` at its function's definition.
538 def OverloadableDocs : Documentation {
539 let Category = DocCatFunction;
541 Clang provides support for C++ function overloading in C. Function overloading
542 in C is introduced using the ``overloadable`` attribute. For example, one
543 might provide several overloaded versions of a ``tgsin`` function that invokes
544 the appropriate standard function computing the sine of a value with ``float``,
545 ``double``, or ``long double`` precision:
550 float __attribute__((overloadable)) tgsin(float x) { return sinf(x); }
551 double __attribute__((overloadable)) tgsin(double x) { return sin(x); }
552 long double __attribute__((overloadable)) tgsin(long double x) { return sinl(x); }
554 Given these declarations, one can call ``tgsin`` with a ``float`` value to
555 receive a ``float`` result, with a ``double`` to receive a ``double`` result,
556 etc. Function overloading in C follows the rules of C++ function overloading
557 to pick the best overload given the call arguments, with a few C-specific
560 * Conversion from ``float`` or ``double`` to ``long double`` is ranked as a
561 floating-point promotion (per C99) rather than as a floating-point conversion
564 * A conversion from a pointer of type ``T*`` to a pointer of type ``U*`` is
565 considered a pointer conversion (with conversion rank) if ``T`` and ``U`` are
568 * A conversion from type ``T`` to a value of type ``U`` is permitted if ``T``
569 and ``U`` are compatible types. This conversion is given "conversion" rank.
571 * If no viable candidates are otherwise available, we allow a conversion from a
572 pointer of type ``T*`` to a pointer of type ``U*``, where ``T`` and ``U`` are
573 incompatible. This conversion is ranked below all other types of conversions.
574 Please note: ``U`` lacking qualifiers that are present on ``T`` is sufficient
575 for ``T`` and ``U`` to be incompatible.
577 The declaration of ``overloadable`` functions is restricted to function
578 declarations and definitions. Most importantly, if any function with a given
579 name is given the ``overloadable`` attribute, then all function declarations
580 and definitions with that name (and in that scope) must have the
581 ``overloadable`` attribute. This rule even applies to redeclarations of
582 functions whose original declaration had the ``overloadable`` attribute, e.g.,
586 int f(int) __attribute__((overloadable));
587 float f(float); // error: declaration of "f" must have the "overloadable" attribute
589 int g(int) __attribute__((overloadable));
590 int g(int) { } // error: redeclaration of "g" must also have the "overloadable" attribute
592 Functions marked ``overloadable`` must have prototypes. Therefore, the
593 following code is ill-formed:
597 int h() __attribute__((overloadable)); // error: h does not have a prototype
599 However, ``overloadable`` functions are allowed to use a ellipsis even if there
600 are no named parameters (as is permitted in C++). This feature is particularly
601 useful when combined with the ``unavailable`` attribute:
605 void honeypot(...) __attribute__((overloadable, unavailable)); // calling me is an error
607 Functions declared with the ``overloadable`` attribute have their names mangled
608 according to the same rules as C++ function names. For example, the three
609 ``tgsin`` functions in our motivating example get the mangled names
610 ``_Z5tgsinf``, ``_Z5tgsind``, and ``_Z5tgsine``, respectively. There are two
611 caveats to this use of name mangling:
613 * Future versions of Clang may change the name mangling of functions overloaded
614 in C, so you should not depend on an specific mangling. To be completely
615 safe, we strongly urge the use of ``static inline`` with ``overloadable``
618 * The ``overloadable`` attribute has almost no meaning when used in C++,
619 because names will already be mangled and functions are already overloadable.
620 However, when an ``overloadable`` function occurs within an ``extern "C"``
621 linkage specification, it's name *will* be mangled in the same way as it
624 Query for this feature with ``__has_extension(attribute_overloadable)``.
628 def ObjCMethodFamilyDocs : Documentation {
629 let Category = DocCatFunction;
631 Many methods in Objective-C have conventional meanings determined by their
632 selectors. It is sometimes useful to be able to mark a method as having a
633 particular conventional meaning despite not having the right selector, or as
634 not having the conventional meaning that its selector would suggest. For these
635 use cases, we provide an attribute to specifically describe the "method family"
636 that a method belongs to.
638 **Usage**: ``__attribute__((objc_method_family(X)))``, where ``X`` is one of
639 ``none``, ``alloc``, ``copy``, ``init``, ``mutableCopy``, or ``new``. This
640 attribute can only be placed at the end of a method declaration:
644 - (NSString *)initMyStringValue __attribute__((objc_method_family(none)));
646 Users who do not wish to change the conventional meaning of a method, and who
647 merely want to document its non-standard retain and release semantics, should
648 use the retaining behavior attributes (``ns_returns_retained``,
649 ``ns_returns_not_retained``, etc).
651 Query for this feature with ``__has_attribute(objc_method_family)``.
655 def NoDebugDocs : Documentation {
656 let Category = DocCatVariable;
658 The ``nodebug`` attribute allows you to suppress debugging information for a
659 function or method, or for a variable that is not a parameter or a non-static
664 def NoDuplicateDocs : Documentation {
665 let Category = DocCatFunction;
667 The ``noduplicate`` attribute can be placed on function declarations to control
668 whether function calls to this function can be duplicated or not as a result of
669 optimizations. This is required for the implementation of functions with
670 certain special requirements, like the OpenCL "barrier" function, that might
671 need to be run concurrently by all the threads that are executing in lockstep
672 on the hardware. For example this attribute applied on the function
673 "nodupfunc" in the code below avoids that:
677 void nodupfunc() __attribute__((noduplicate));
678 // Setting it as a C++11 attribute is also valid
679 // void nodupfunc() [[clang::noduplicate]];
690 gets possibly modified by some optimizations into code similar to this:
702 where the call to "nodupfunc" is duplicated and sunk into the two branches
707 def ConvergentDocs : Documentation {
708 let Category = DocCatFunction;
710 The ``convergent`` attribute can be placed on a function declaration. It is
711 translated into the LLVM ``convergent`` attribute, which indicates that the call
712 instructions of a function with this attribute cannot be made control-dependent
713 on any additional values.
715 In languages designed for SPMD/SIMT programming model, e.g. OpenCL or CUDA,
716 the call instructions of a function with this attribute must be executed by
717 all work items or threads in a work group or sub group.
719 This attribute is different from ``noduplicate`` because it allows duplicating
720 function calls if it can be proved that the duplicated function calls are
721 not made control-dependent on any additional values, e.g., unrolling a loop
722 executed by all work items.
727 void convfunc(void) __attribute__((convergent));
728 // Setting it as a C++11 attribute is also valid in a C++ program.
729 // void convfunc(void) [[clang::convergent]];
734 def NoSplitStackDocs : Documentation {
735 let Category = DocCatFunction;
737 The ``no_split_stack`` attribute disables the emission of the split stack
738 preamble for a particular function. It has no effect if ``-fsplit-stack``
743 def ObjCRequiresSuperDocs : Documentation {
744 let Category = DocCatFunction;
746 Some Objective-C classes allow a subclass to override a particular method in a
747 parent class but expect that the overriding method also calls the overridden
748 method in the parent class. For these cases, we provide an attribute to
749 designate that a method requires a "call to ``super``" in the overriding
750 method in the subclass.
752 **Usage**: ``__attribute__((objc_requires_super))``. This attribute can only
753 be placed at the end of a method declaration:
757 - (void)foo __attribute__((objc_requires_super));
759 This attribute can only be applied the method declarations within a class, and
760 not a protocol. Currently this attribute does not enforce any placement of
761 where the call occurs in the overriding method (such as in the case of
762 ``-dealloc`` where the call must appear at the end). It checks only that it
765 Note that on both OS X and iOS that the Foundation framework provides a
766 convenience macro ``NS_REQUIRES_SUPER`` that provides syntactic sugar for this
771 - (void)foo NS_REQUIRES_SUPER;
773 This macro is conditionally defined depending on the compiler's support for
774 this attribute. If the compiler does not support the attribute the macro
777 Operationally, when a method has this annotation the compiler will warn if the
778 implementation of an override in a subclass does not call super. For example:
782 warning: method possibly missing a [super AnnotMeth] call
783 - (void) AnnotMeth{};
788 def ObjCRuntimeNameDocs : Documentation {
789 let Category = DocCatFunction;
791 By default, the Objective-C interface or protocol identifier is used
792 in the metadata name for that object. The `objc_runtime_name`
793 attribute allows annotated interfaces or protocols to use the
794 specified string argument in the object's metadata name instead of the
797 **Usage**: ``__attribute__((objc_runtime_name("MyLocalName")))``. This attribute
798 can only be placed before an @protocol or @interface declaration:
802 __attribute__((objc_runtime_name("MyLocalName")))
809 def ObjCRuntimeVisibleDocs : Documentation {
810 let Category = DocCatFunction;
812 This attribute specifies that the Objective-C class to which it applies is visible to the Objective-C runtime but not to the linker. Classes annotated with this attribute cannot be subclassed and cannot have categories defined for them.
816 def ObjCBoxableDocs : Documentation {
817 let Category = DocCatFunction;
819 Structs and unions marked with the ``objc_boxable`` attribute can be used
820 with the Objective-C boxed expression syntax, ``@(...)``.
822 **Usage**: ``__attribute__((objc_boxable))``. This attribute
823 can only be placed on a declaration of a trivially-copyable struct or union:
827 struct __attribute__((objc_boxable)) some_struct {
830 union __attribute__((objc_boxable)) some_union {
834 typedef struct __attribute__((objc_boxable)) _some_struct some_struct;
839 NSValue *boxed = @(ss);
844 def AvailabilityDocs : Documentation {
845 let Category = DocCatFunction;
847 The ``availability`` attribute can be placed on declarations to describe the
848 lifecycle of that declaration relative to operating system versions. Consider
849 the function declaration for a hypothetical function ``f``:
853 void f(void) __attribute__((availability(macos,introduced=10.4,deprecated=10.6,obsoleted=10.7)));
855 The availability attribute states that ``f`` was introduced in Mac OS X 10.4,
856 deprecated in Mac OS X 10.6, and obsoleted in Mac OS X 10.7. This information
857 is used by Clang to determine when it is safe to use ``f``: for example, if
858 Clang is instructed to compile code for Mac OS X 10.5, a call to ``f()``
859 succeeds. If Clang is instructed to compile code for Mac OS X 10.6, the call
860 succeeds but Clang emits a warning specifying that the function is deprecated.
861 Finally, if Clang is instructed to compile code for Mac OS X 10.7, the call
862 fails because ``f()`` is no longer available.
864 The availability attribute is a comma-separated list starting with the
865 platform name and then including clauses specifying important milestones in the
866 declaration's lifetime (in any order) along with additional information. Those
869 introduced=\ *version*
870 The first version in which this declaration was introduced.
872 deprecated=\ *version*
873 The first version in which this declaration was deprecated, meaning that
874 users should migrate away from this API.
876 obsoleted=\ *version*
877 The first version in which this declaration was obsoleted, meaning that it
878 was removed completely and can no longer be used.
881 This declaration is never available on this platform.
883 message=\ *string-literal*
884 Additional message text that Clang will provide when emitting a warning or
885 error about use of a deprecated or obsoleted declaration. Useful to direct
886 users to replacement APIs.
888 replacement=\ *string-literal*
889 Additional message text that Clang will use to provide Fix-It when emitting
890 a warning about use of a deprecated declaration. The Fix-It will replace
891 the deprecated declaration with the new declaration specified.
893 Multiple availability attributes can be placed on a declaration, which may
894 correspond to different platforms. Only the availability attribute with the
895 platform corresponding to the target platform will be used; any others will be
896 ignored. If no availability attribute specifies availability for the current
897 target platform, the availability attributes are ignored. Supported platforms
901 Apple's iOS operating system. The minimum deployment target is specified by
902 the ``-mios-version-min=*version*`` or ``-miphoneos-version-min=*version*``
903 command-line arguments.
906 Apple's Mac OS X operating system. The minimum deployment target is
907 specified by the ``-mmacosx-version-min=*version*`` command-line argument.
908 ``macosx`` is supported for backward-compatibility reasons, but it is
912 Apple's tvOS operating system. The minimum deployment target is specified by
913 the ``-mtvos-version-min=*version*`` command-line argument.
916 Apple's watchOS operating system. The minimum deployment target is specified by
917 the ``-mwatchos-version-min=*version*`` command-line argument.
919 A declaration can typically be used even when deploying back to a platform
920 version prior to when the declaration was introduced. When this happens, the
921 declaration is `weakly linked
922 <https://developer.apple.com/library/mac/#documentation/MacOSX/Conceptual/BPFrameworks/Concepts/WeakLinking.html>`_,
923 as if the ``weak_import`` attribute were added to the declaration. A
924 weakly-linked declaration may or may not be present a run-time, and a program
925 can determine whether the declaration is present by checking whether the
926 address of that declaration is non-NULL.
928 The flag ``strict`` disallows using API when deploying back to a
929 platform version prior to when the declaration was introduced. An
930 attempt to use such API before its introduction causes a hard error.
931 Weakly-linking is almost always a better API choice, since it allows
932 users to query availability at runtime.
934 If there are multiple declarations of the same entity, the availability
935 attributes must either match on a per-platform basis or later
936 declarations must not have availability attributes for that
937 platform. For example:
941 void g(void) __attribute__((availability(macos,introduced=10.4)));
942 void g(void) __attribute__((availability(macos,introduced=10.4))); // okay, matches
943 void g(void) __attribute__((availability(ios,introduced=4.0))); // okay, adds a new platform
944 void g(void); // okay, inherits both macos and ios availability from above.
945 void g(void) __attribute__((availability(macos,introduced=10.5))); // error: mismatch
947 When one method overrides another, the overriding method can be more widely available than the overridden method, e.g.,:
952 - (id)method __attribute__((availability(macos,introduced=10.4)));
953 - (id)method2 __attribute__((availability(macos,introduced=10.4)));
957 - (id)method __attribute__((availability(macos,introduced=10.3))); // okay: method moved into base class later
958 - (id)method __attribute__((availability(macos,introduced=10.5))); // error: this method was available via the base class in 10.4
964 def RequireConstantInitDocs : Documentation {
965 let Category = DocCatVariable;
967 This attribute specifies that the variable to which it is attached is intended
968 to have a `constant initializer <http://en.cppreference.com/w/cpp/language/constant_initialization>`_
969 according to the rules of [basic.start.static]. The variable is required to
970 have static or thread storage duration. If the initialization of the variable
971 is not a constant initializer an error will be produced. This attribute may
974 Note that in C++03 strict constant expression checking is not done. Instead
975 the attribute reports if Clang can emit the variable as a constant, even if it's
976 not technically a 'constant initializer'. This behavior is non-portable.
978 Static storage duration variables with constant initializers avoid hard-to-find
979 bugs caused by the indeterminate order of dynamic initialization. They can also
980 be safely used during dynamic initialization across translation units.
982 This attribute acts as a compile time assertion that the requirements
983 for constant initialization have been met. Since these requirements change
984 between dialects and have subtle pitfalls it's important to fail fast instead
985 of silently falling back on dynamic initialization.
990 #define SAFE_STATIC [[clang::require_constant_initialization]]
995 SAFE_STATIC T x = {42}; // Initialization OK. Doesn't check destructor.
996 SAFE_STATIC T y = 42; // error: variable does not have a constant initializer
997 // copy initialization is not a constant expression on a non-literal type.
1001 def WarnMaybeUnusedDocs : Documentation {
1002 let Category = DocCatVariable;
1003 let Heading = "maybe_unused, unused, gnu::unused";
1005 When passing the ``-Wunused`` flag to Clang, entities that are unused by the
1006 program may be diagnosed. The ``[[maybe_unused]]`` (or
1007 ``__attribute__((unused))``) attribute can be used to silence such diagnostics
1008 when the entity cannot be removed. For instance, a local variable may exist
1009 solely for use in an ``assert()`` statement, which makes the local variable
1010 unused when ``NDEBUG`` is defined.
1012 The attribute may be applied to the declaration of a class, a typedef, a
1013 variable, a function or method, a function parameter, an enumeration, an
1014 enumerator, a non-static data member, or a label.
1019 [[maybe_unused]] void f([[maybe_unused]] bool thing1,
1020 [[maybe_unused]] bool thing2) {
1021 [[maybe_unused]] bool b = thing1 && thing2;
1027 def WarnUnusedResultsDocs : Documentation {
1028 let Category = DocCatFunction;
1029 let Heading = "nodiscard, warn_unused_result, clang::warn_unused_result, gnu::warn_unused_result";
1031 Clang supports the ability to diagnose when the results of a function call
1032 expression are discarded under suspicious circumstances. A diagnostic is
1033 generated when a function or its return type is marked with ``[[nodiscard]]``
1034 (or ``__attribute__((warn_unused_result))``) and the function call appears as a
1035 potentially-evaluated discarded-value expression that is not explicitly cast to
1039 struct [[nodiscard]] error_info { /*...*/ };
1040 error_info enable_missile_safety_mode();
1042 void launch_missiles();
1043 void test_missiles() {
1044 enable_missile_safety_mode(); // diagnoses
1048 void f() { foo(); } // Does not diagnose, error_info is a reference.
1052 def FallthroughDocs : Documentation {
1053 let Category = DocCatStmt;
1054 let Heading = "fallthrough, clang::fallthrough";
1056 The ``fallthrough`` (or ``clang::fallthrough``) attribute is used
1057 to annotate intentional fall-through
1058 between switch labels. It can only be applied to a null statement placed at a
1059 point of execution between any statement and the next switch label. It is
1060 common to mark these places with a specific comment, but this attribute is
1061 meant to replace comments with a more strict annotation, which can be checked
1062 by the compiler. This attribute doesn't change semantics of the code and can
1063 be used wherever an intended fall-through occurs. It is designed to mimic
1064 control-flow statements like ``break;``, so it can be placed in most places
1065 where ``break;`` can, but only if there are no statements on the execution path
1066 between it and the next switch label.
1068 By default, Clang does not warn on unannotated fallthrough from one ``switch``
1069 case to another. Diagnostics on fallthrough without a corresponding annotation
1070 can be enabled with the ``-Wimplicit-fallthrough`` argument.
1076 // compile with -Wimplicit-fallthrough
1079 case 33: // no warning: no statements between case labels
1081 case 44: // warning: unannotated fall-through
1083 [[clang::fallthrough]];
1084 case 55: // no warning
1091 [[clang::fallthrough]];
1093 case 66: // no warning
1095 [[clang::fallthrough]]; // warning: fallthrough annotation does not
1096 // directly precede case label
1098 case 77: // warning: unannotated fall-through
1104 def ARMInterruptDocs : Documentation {
1105 let Category = DocCatFunction;
1107 Clang supports the GNU style ``__attribute__((interrupt("TYPE")))`` attribute on
1108 ARM targets. This attribute may be attached to a function definition and
1109 instructs the backend to generate appropriate function entry/exit code so that
1110 it can be used directly as an interrupt service routine.
1112 The parameter passed to the interrupt attribute is optional, but if
1113 provided it must be a string literal with one of the following values: "IRQ",
1114 "FIQ", "SWI", "ABORT", "UNDEF".
1116 The semantics are as follows:
1118 - If the function is AAPCS, Clang instructs the backend to realign the stack to
1119 8 bytes on entry. This is a general requirement of the AAPCS at public
1120 interfaces, but may not hold when an exception is taken. Doing this allows
1121 other AAPCS functions to be called.
1122 - If the CPU is M-class this is all that needs to be done since the architecture
1123 itself is designed in such a way that functions obeying the normal AAPCS ABI
1124 constraints are valid exception handlers.
1125 - If the CPU is not M-class, the prologue and epilogue are modified to save all
1126 non-banked registers that are used, so that upon return the user-mode state
1127 will not be corrupted. Note that to avoid unnecessary overhead, only
1128 general-purpose (integer) registers are saved in this way. If VFP operations
1129 are needed, that state must be saved manually.
1131 Specifically, interrupt kinds other than "FIQ" will save all core registers
1132 except "lr" and "sp". "FIQ" interrupts will save r0-r7.
1133 - If the CPU is not M-class, the return instruction is changed to one of the
1134 canonical sequences permitted by the architecture for exception return. Where
1135 possible the function itself will make the necessary "lr" adjustments so that
1136 the "preferred return address" is selected.
1138 Unfortunately the compiler is unable to make this guarantee for an "UNDEF"
1139 handler, where the offset from "lr" to the preferred return address depends on
1140 the execution state of the code which generated the exception. In this case
1141 a sequence equivalent to "movs pc, lr" will be used.
1145 def MipsInterruptDocs : Documentation {
1146 let Category = DocCatFunction;
1148 Clang supports the GNU style ``__attribute__((interrupt("ARGUMENT")))`` attribute on
1149 MIPS targets. This attribute may be attached to a function definition and instructs
1150 the backend to generate appropriate function entry/exit code so that it can be used
1151 directly as an interrupt service routine.
1153 By default, the compiler will produce a function prologue and epilogue suitable for
1154 an interrupt service routine that handles an External Interrupt Controller (eic)
1155 generated interrupt. This behaviour can be explicitly requested with the "eic"
1158 Otherwise, for use with vectored interrupt mode, the argument passed should be
1159 of the form "vector=LEVEL" where LEVEL is one of the following values:
1160 "sw0", "sw1", "hw0", "hw1", "hw2", "hw3", "hw4", "hw5". The compiler will
1161 then set the interrupt mask to the corresponding level which will mask all
1162 interrupts up to and including the argument.
1164 The semantics are as follows:
1166 - The prologue is modified so that the Exception Program Counter (EPC) and
1167 Status coprocessor registers are saved to the stack. The interrupt mask is
1168 set so that the function can only be interrupted by a higher priority
1169 interrupt. The epilogue will restore the previous values of EPC and Status.
1171 - The prologue and epilogue are modified to save and restore all non-kernel
1172 registers as necessary.
1174 - The FPU is disabled in the prologue, as the floating pointer registers are not
1175 spilled to the stack.
1177 - The function return sequence is changed to use an exception return instruction.
1179 - The parameter sets the interrupt mask for the function corresponding to the
1180 interrupt level specified. If no mask is specified the interrupt mask
1185 def TargetDocs : Documentation {
1186 let Category = DocCatFunction;
1188 Clang supports the GNU style ``__attribute__((target("OPTIONS")))`` attribute.
1189 This attribute may be attached to a function definition and instructs
1190 the backend to use different code generation options than were passed on the
1193 The current set of options correspond to the existing "subtarget features" for
1194 the target with or without a "-mno-" in front corresponding to the absence
1195 of the feature, as well as ``arch="CPU"`` which will change the default "CPU"
1198 Example "subtarget features" from the x86 backend include: "mmx", "sse", "sse4.2",
1199 "avx", "xop" and largely correspond to the machine specific options handled by
1204 def DocCatAMDGPUAttributes : DocumentationCategory<"AMD GPU Attributes">;
1206 def AMDGPUFlatWorkGroupSizeDocs : Documentation {
1207 let Category = DocCatAMDGPUAttributes;
1209 The flat work-group size is the number of work-items in the work-group size
1210 specified when the kernel is dispatched. It is the product of the sizes of the
1211 x, y, and z dimension of the work-group.
1214 ``__attribute__((amdgpu_flat_work_group_size(<min>, <max>)))`` attribute for the
1215 AMDGPU target. This attribute may be attached to a kernel function definition
1216 and is an optimization hint.
1218 ``<min>`` parameter specifies the minimum flat work-group size, and ``<max>``
1219 parameter specifies the maximum flat work-group size (must be greater than
1220 ``<min>``) to which all dispatches of the kernel will conform. Passing ``0, 0``
1221 as ``<min>, <max>`` implies the default behavior (``128, 256``).
1223 If specified, the AMDGPU target backend might be able to produce better machine
1224 code for barriers and perform scratch promotion by estimating available group
1227 An error will be given if:
1228 - Specified values violate subtarget specifications;
1229 - Specified values are not compatible with values provided through other
1234 def AMDGPUWavesPerEUDocs : Documentation {
1235 let Category = DocCatAMDGPUAttributes;
1237 A compute unit (CU) is responsible for executing the wavefronts of a work-group.
1238 It is composed of one or more execution units (EU), which are responsible for
1239 executing the wavefronts. An EU can have enough resources to maintain the state
1240 of more than one executing wavefront. This allows an EU to hide latency by
1241 switching between wavefronts in a similar way to symmetric multithreading on a
1242 CPU. In order to allow the state for multiple wavefronts to fit on an EU, the
1243 resources used by a single wavefront have to be limited. For example, the number
1244 of SGPRs and VGPRs. Limiting such resources can allow greater latency hiding,
1245 but can result in having to spill some register state to memory.
1247 Clang supports the ``__attribute__((amdgpu_waves_per_eu(<min>[, <max>])))``
1248 attribute for the AMDGPU target. This attribute may be attached to a kernel
1249 function definition and is an optimization hint.
1251 ``<min>`` parameter specifies the requested minimum number of waves per EU, and
1252 *optional* ``<max>`` parameter specifies the requested maximum number of waves
1253 per EU (must be greater than ``<min>`` if specified). If ``<max>`` is omitted,
1254 then there is no restriction on the maximum number of waves per EU other than
1255 the one dictated by the hardware for which the kernel is compiled. Passing
1256 ``0, 0`` as ``<min>, <max>`` implies the default behavior (no limits).
1258 If specified, this attribute allows an advanced developer to tune the number of
1259 wavefronts that are capable of fitting within the resources of an EU. The AMDGPU
1260 target backend can use this information to limit resources, such as number of
1261 SGPRs, number of VGPRs, size of available group and private memory segments, in
1262 such a way that guarantees that at least ``<min>`` wavefronts and at most
1263 ``<max>`` wavefronts are able to fit within the resources of an EU. Requesting
1264 more wavefronts can hide memory latency but limits available registers which
1265 can result in spilling. Requesting fewer wavefronts can help reduce cache
1266 thrashing, but can reduce memory latency hiding.
1268 This attribute controls the machine code generated by the AMDGPU target backend
1269 to ensure it is capable of meeting the requested values. However, when the
1270 kernel is executed, there may be other reasons that prevent meeting the request,
1271 for example, there may be wavefronts from other kernels executing on the EU.
1273 An error will be given if:
1274 - Specified values violate subtarget specifications;
1275 - Specified values are not compatible with values provided through other
1277 - The AMDGPU target backend is unable to create machine code that can meet the
1282 def AMDGPUNumSGPRNumVGPRDocs : Documentation {
1283 let Category = DocCatAMDGPUAttributes;
1285 Clang supports the ``__attribute__((amdgpu_num_sgpr(<num_sgpr>)))`` and
1286 ``__attribute__((amdgpu_num_vgpr(<num_vgpr>)))`` attributes for the AMDGPU
1287 target. These attributes may be attached to a kernel function definition and are
1288 an optimization hint.
1290 If these attributes are specified, then the AMDGPU target backend will attempt
1291 to limit the number of SGPRs and/or VGPRs used to the specified value(s). The
1292 number of used SGPRs and/or VGPRs may further be rounded up to satisfy the
1293 allocation requirements or constraints of the subtarget. Passing ``0`` as
1294 ``num_sgpr`` and/or ``num_vgpr`` implies the default behavior (no limits).
1296 These attributes can be used to test the AMDGPU target backend. It is
1297 recommended that the ``amdgpu_waves_per_eu`` attribute be used to control
1298 resources such as SGPRs and VGPRs since it is aware of the limits for different
1301 An error will be given if:
1302 - Specified values violate subtarget specifications;
1303 - Specified values are not compatible with values provided through other
1305 - The AMDGPU target backend is unable to create machine code that can meet the
1310 def DocCatCallingConvs : DocumentationCategory<"Calling Conventions"> {
1312 Clang supports several different calling conventions, depending on the target
1313 platform and architecture. The calling convention used for a function determines
1314 how parameters are passed, how results are returned to the caller, and other
1315 low-level details of calling a function.
1319 def PcsDocs : Documentation {
1320 let Category = DocCatCallingConvs;
1322 On ARM targets, this attribute can be used to select calling conventions
1323 similar to ``stdcall`` on x86. Valid parameter values are "aapcs" and
1328 def RegparmDocs : Documentation {
1329 let Category = DocCatCallingConvs;
1331 On 32-bit x86 targets, the regparm attribute causes the compiler to pass
1332 the first three integer parameters in EAX, EDX, and ECX instead of on the
1333 stack. This attribute has no effect on variadic functions, and all parameters
1334 are passed via the stack as normal.
1338 def SysVABIDocs : Documentation {
1339 let Category = DocCatCallingConvs;
1341 On Windows x86_64 targets, this attribute changes the calling convention of a
1342 function to match the default convention used on Sys V targets such as Linux,
1343 Mac, and BSD. This attribute has no effect on other targets.
1347 def MSABIDocs : Documentation {
1348 let Category = DocCatCallingConvs;
1350 On non-Windows x86_64 targets, this attribute changes the calling convention of
1351 a function to match the default convention used on Windows x86_64. This
1352 attribute has no effect on Windows targets or non-x86_64 targets.
1356 def StdCallDocs : Documentation {
1357 let Category = DocCatCallingConvs;
1359 On 32-bit x86 targets, this attribute changes the calling convention of a
1360 function to clear parameters off of the stack on return. This convention does
1361 not support variadic calls or unprototyped functions in C, and has no effect on
1362 x86_64 targets. This calling convention is used widely by the Windows API and
1363 COM applications. See the documentation for `__stdcall`_ on MSDN.
1365 .. _`__stdcall`: http://msdn.microsoft.com/en-us/library/zxk0tw93.aspx
1369 def FastCallDocs : Documentation {
1370 let Category = DocCatCallingConvs;
1372 On 32-bit x86 targets, this attribute changes the calling convention of a
1373 function to use ECX and EDX as register parameters and clear parameters off of
1374 the stack on return. This convention does not support variadic calls or
1375 unprototyped functions in C, and has no effect on x86_64 targets. This calling
1376 convention is supported primarily for compatibility with existing code. Users
1377 seeking register parameters should use the ``regparm`` attribute, which does
1378 not require callee-cleanup. See the documentation for `__fastcall`_ on MSDN.
1380 .. _`__fastcall`: http://msdn.microsoft.com/en-us/library/6xa169sk.aspx
1384 def RegCallDocs : Documentation {
1385 let Category = DocCatCallingConvs;
1387 On x86 targets, this attribute changes the calling convention to
1388 `__regcall`_ convention. This convention aims to pass as many arguments
1389 as possible in registers. It also tries to utilize registers for the
1390 return value whenever it is possible.
1392 .. _`__regcall`: https://software.intel.com/en-us/node/693069
1396 def ThisCallDocs : Documentation {
1397 let Category = DocCatCallingConvs;
1399 On 32-bit x86 targets, this attribute changes the calling convention of a
1400 function to use ECX for the first parameter (typically the implicit ``this``
1401 parameter of C++ methods) and clear parameters off of the stack on return. This
1402 convention does not support variadic calls or unprototyped functions in C, and
1403 has no effect on x86_64 targets. See the documentation for `__thiscall`_ on
1406 .. _`__thiscall`: http://msdn.microsoft.com/en-us/library/ek8tkfbw.aspx
1410 def VectorCallDocs : Documentation {
1411 let Category = DocCatCallingConvs;
1413 On 32-bit x86 *and* x86_64 targets, this attribute changes the calling
1414 convention of a function to pass vector parameters in SSE registers.
1416 On 32-bit x86 targets, this calling convention is similar to ``__fastcall``.
1417 The first two integer parameters are passed in ECX and EDX. Subsequent integer
1418 parameters are passed in memory, and callee clears the stack. On x86_64
1419 targets, the callee does *not* clear the stack, and integer parameters are
1420 passed in RCX, RDX, R8, and R9 as is done for the default Windows x64 calling
1423 On both 32-bit x86 and x86_64 targets, vector and floating point arguments are
1424 passed in XMM0-XMM5. Homogeneous vector aggregates of up to four elements are
1425 passed in sequential SSE registers if enough are available. If AVX is enabled,
1426 256 bit vectors are passed in YMM0-YMM5. Any vector or aggregate type that
1427 cannot be passed in registers for any reason is passed by reference, which
1428 allows the caller to align the parameter memory.
1430 See the documentation for `__vectorcall`_ on MSDN for more details.
1432 .. _`__vectorcall`: http://msdn.microsoft.com/en-us/library/dn375768.aspx
1436 def DocCatConsumed : DocumentationCategory<"Consumed Annotation Checking"> {
1438 Clang supports additional attributes for checking basic resource management
1439 properties, specifically for unique objects that have a single owning reference.
1440 The following attributes are currently supported, although **the implementation
1441 for these annotations is currently in development and are subject to change.**
1445 def SetTypestateDocs : Documentation {
1446 let Category = DocCatConsumed;
1448 Annotate methods that transition an object into a new state with
1449 ``__attribute__((set_typestate(new_state)))``. The new state must be
1450 unconsumed, consumed, or unknown.
1454 def CallableWhenDocs : Documentation {
1455 let Category = DocCatConsumed;
1457 Use ``__attribute__((callable_when(...)))`` to indicate what states a method
1458 may be called in. Valid states are unconsumed, consumed, or unknown. Each
1459 argument to this attribute must be a quoted string. E.g.:
1461 ``__attribute__((callable_when("unconsumed", "unknown")))``
1465 def TestTypestateDocs : Documentation {
1466 let Category = DocCatConsumed;
1468 Use ``__attribute__((test_typestate(tested_state)))`` to indicate that a method
1469 returns true if the object is in the specified state..
1473 def ParamTypestateDocs : Documentation {
1474 let Category = DocCatConsumed;
1476 This attribute specifies expectations about function parameters. Calls to an
1477 function with annotated parameters will issue a warning if the corresponding
1478 argument isn't in the expected state. The attribute is also used to set the
1479 initial state of the parameter when analyzing the function's body.
1483 def ReturnTypestateDocs : Documentation {
1484 let Category = DocCatConsumed;
1486 The ``return_typestate`` attribute can be applied to functions or parameters.
1487 When applied to a function the attribute specifies the state of the returned
1488 value. The function's body is checked to ensure that it always returns a value
1489 in the specified state. On the caller side, values returned by the annotated
1490 function are initialized to the given state.
1492 When applied to a function parameter it modifies the state of an argument after
1493 a call to the function returns. The function's body is checked to ensure that
1494 the parameter is in the expected state before returning.
1498 def ConsumableDocs : Documentation {
1499 let Category = DocCatConsumed;
1501 Each ``class`` that uses any of the typestate annotations must first be marked
1502 using the ``consumable`` attribute. Failure to do so will result in a warning.
1504 This attribute accepts a single parameter that must be one of the following:
1505 ``unknown``, ``consumed``, or ``unconsumed``.
1509 def NoSanitizeDocs : Documentation {
1510 let Category = DocCatFunction;
1512 Use the ``no_sanitize`` attribute on a function declaration to specify
1513 that a particular instrumentation or set of instrumentations should not be
1514 applied to that function. The attribute takes a list of string literals,
1515 which have the same meaning as values accepted by the ``-fno-sanitize=``
1516 flag. For example, ``__attribute__((no_sanitize("address", "thread")))``
1517 specifies that AddressSanitizer and ThreadSanitizer should not be applied
1520 See :ref:`Controlling Code Generation <controlling-code-generation>` for a
1521 full list of supported sanitizer flags.
1525 def NoSanitizeAddressDocs : Documentation {
1526 let Category = DocCatFunction;
1527 // This function has multiple distinct spellings, and so it requires a custom
1528 // heading to be specified. The most common spelling is sufficient.
1529 let Heading = "no_sanitize_address (no_address_safety_analysis, gnu::no_address_safety_analysis, gnu::no_sanitize_address)";
1531 .. _langext-address_sanitizer:
1533 Use ``__attribute__((no_sanitize_address))`` on a function declaration to
1534 specify that address safety instrumentation (e.g. AddressSanitizer) should
1535 not be applied to that function.
1539 def NoSanitizeThreadDocs : Documentation {
1540 let Category = DocCatFunction;
1541 let Heading = "no_sanitize_thread";
1543 .. _langext-thread_sanitizer:
1545 Use ``__attribute__((no_sanitize_thread))`` on a function declaration to
1546 specify that checks for data races on plain (non-atomic) memory accesses should
1547 not be inserted by ThreadSanitizer. The function is still instrumented by the
1548 tool to avoid false positives and provide meaningful stack traces.
1552 def NoSanitizeMemoryDocs : Documentation {
1553 let Category = DocCatFunction;
1554 let Heading = "no_sanitize_memory";
1556 .. _langext-memory_sanitizer:
1558 Use ``__attribute__((no_sanitize_memory))`` on a function declaration to
1559 specify that checks for uninitialized memory should not be inserted
1560 (e.g. by MemorySanitizer). The function may still be instrumented by the tool
1561 to avoid false positives in other places.
1565 def DocCatTypeSafety : DocumentationCategory<"Type Safety Checking"> {
1567 Clang supports additional attributes to enable checking type safety properties
1568 that can't be enforced by the C type system. To see warnings produced by these
1569 checks, ensure that -Wtype-safety is enabled. Use cases include:
1571 * MPI library implementations, where these attributes enable checking that
1572 the buffer type matches the passed ``MPI_Datatype``;
1573 * for HDF5 library there is a similar use case to MPI;
1574 * checking types of variadic functions' arguments for functions like
1575 ``fcntl()`` and ``ioctl()``.
1577 You can detect support for these attributes with ``__has_attribute()``. For
1582 #if defined(__has_attribute)
1583 # if __has_attribute(argument_with_type_tag) && \
1584 __has_attribute(pointer_with_type_tag) && \
1585 __has_attribute(type_tag_for_datatype)
1586 # define ATTR_MPI_PWT(buffer_idx, type_idx) __attribute__((pointer_with_type_tag(mpi,buffer_idx,type_idx)))
1587 /* ... other macros ... */
1591 #if !defined(ATTR_MPI_PWT)
1592 # define ATTR_MPI_PWT(buffer_idx, type_idx)
1595 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1600 def ArgumentWithTypeTagDocs : Documentation {
1601 let Category = DocCatTypeSafety;
1602 let Heading = "argument_with_type_tag";
1604 Use ``__attribute__((argument_with_type_tag(arg_kind, arg_idx,
1605 type_tag_idx)))`` on a function declaration to specify that the function
1606 accepts a type tag that determines the type of some other argument.
1608 This attribute is primarily useful for checking arguments of variadic functions
1609 (``pointer_with_type_tag`` can be used in most non-variadic cases).
1611 In the attribute prototype above:
1612 * ``arg_kind`` is an identifier that should be used when annotating all
1613 applicable type tags.
1614 * ``arg_idx`` provides the position of a function argument. The expected type of
1615 this function argument will be determined by the function argument specified
1616 by ``type_tag_idx``. In the code example below, "3" means that the type of the
1617 function's third argument will be determined by ``type_tag_idx``.
1618 * ``type_tag_idx`` provides the position of a function argument. This function
1619 argument will be a type tag. The type tag will determine the expected type of
1620 the argument specified by ``arg_idx``. In the code example below, "2" means
1621 that the type tag associated with the function's second argument should agree
1622 with the type of the argument specified by ``arg_idx``.
1628 int fcntl(int fd, int cmd, ...)
1629 __attribute__(( argument_with_type_tag(fcntl,3,2) ));
1630 // The function's second argument will be a type tag; this type tag will
1631 // determine the expected type of the function's third argument.
1635 def PointerWithTypeTagDocs : Documentation {
1636 let Category = DocCatTypeSafety;
1637 let Heading = "pointer_with_type_tag";
1639 Use ``__attribute__((pointer_with_type_tag(ptr_kind, ptr_idx, type_tag_idx)))``
1640 on a function declaration to specify that the function accepts a type tag that
1641 determines the pointee type of some other pointer argument.
1643 In the attribute prototype above:
1644 * ``ptr_kind`` is an identifier that should be used when annotating all
1645 applicable type tags.
1646 * ``ptr_idx`` provides the position of a function argument; this function
1647 argument will have a pointer type. The expected pointee type of this pointer
1648 type will be determined by the function argument specified by
1649 ``type_tag_idx``. In the code example below, "1" means that the pointee type
1650 of the function's first argument will be determined by ``type_tag_idx``.
1651 * ``type_tag_idx`` provides the position of a function argument; this function
1652 argument will be a type tag. The type tag will determine the expected pointee
1653 type of the pointer argument specified by ``ptr_idx``. In the code example
1654 below, "3" means that the type tag associated with the function's third
1655 argument should agree with the pointee type of the pointer argument specified
1662 typedef int MPI_Datatype;
1663 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1664 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1665 // The function's 3rd argument will be a type tag; this type tag will
1666 // determine the expected pointee type of the function's 1st argument.
1670 def TypeTagForDatatypeDocs : Documentation {
1671 let Category = DocCatTypeSafety;
1673 When declaring a variable, use
1674 ``__attribute__((type_tag_for_datatype(kind, type)))`` to create a type tag that
1675 is tied to the ``type`` argument given to the attribute.
1677 In the attribute prototype above:
1678 * ``kind`` is an identifier that should be used when annotating all applicable
1680 * ``type`` indicates the name of the type.
1682 Clang supports annotating type tags of two forms.
1684 * **Type tag that is a reference to a declared identifier.**
1685 Use ``__attribute__((type_tag_for_datatype(kind, type)))`` when declaring that
1690 typedef int MPI_Datatype;
1691 extern struct mpi_datatype mpi_datatype_int
1692 __attribute__(( type_tag_for_datatype(mpi,int) ));
1693 #define MPI_INT ((MPI_Datatype) &mpi_datatype_int)
1694 // &mpi_datatype_int is a type tag. It is tied to type "int".
1696 * **Type tag that is an integral literal.**
1697 Declare a ``static const`` variable with an initializer value and attach
1698 ``__attribute__((type_tag_for_datatype(kind, type)))`` on that declaration:
1702 typedef int MPI_Datatype;
1703 static const MPI_Datatype mpi_datatype_int
1704 __attribute__(( type_tag_for_datatype(mpi,int) )) = 42;
1705 #define MPI_INT ((MPI_Datatype) 42)
1706 // The number 42 is a type tag. It is tied to type "int".
1709 The ``type_tag_for_datatype`` attribute also accepts an optional third argument
1710 that determines how the type of the function argument specified by either
1711 ``arg_idx`` or ``ptr_idx`` is compared against the type associated with the type
1712 tag. (Recall that for the ``argument_with_type_tag`` attribute, the type of the
1713 function argument specified by ``arg_idx`` is compared against the type
1714 associated with the type tag. Also recall that for the ``pointer_with_type_tag``
1715 attribute, the pointee type of the function argument specified by ``ptr_idx`` is
1716 compared against the type associated with the type tag.) There are two supported
1717 values for this optional third argument:
1719 * ``layout_compatible`` will cause types to be compared according to
1720 layout-compatibility rules (In C++11 [class.mem] p 17, 18, see the
1721 layout-compatibility rules for two standard-layout struct types and for two
1722 standard-layout union types). This is useful when creating a type tag
1723 associated with a struct or union type. For example:
1728 typedef int MPI_Datatype;
1729 struct internal_mpi_double_int { double d; int i; };
1730 extern struct mpi_datatype mpi_datatype_double_int
1731 __attribute__(( type_tag_for_datatype(mpi,
1732 struct internal_mpi_double_int, layout_compatible) ));
1734 #define MPI_DOUBLE_INT ((MPI_Datatype) &mpi_datatype_double_int)
1736 int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
1737 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1740 struct my_pair { double a; int b; };
1741 struct my_pair *buffer;
1742 MPI_Send(buffer, 1, MPI_DOUBLE_INT /*, ... */); // no warning because the
1743 // layout of my_pair is
1744 // compatible with that of
1745 // internal_mpi_double_int
1747 struct my_int_pair { int a; int b; }
1748 struct my_int_pair *buffer2;
1749 MPI_Send(buffer2, 1, MPI_DOUBLE_INT /*, ... */); // warning because the
1750 // layout of my_int_pair
1751 // does not match that of
1752 // internal_mpi_double_int
1754 * ``must_be_null`` specifies that the function argument specified by either
1755 ``arg_idx`` (for the ``argument_with_type_tag`` attribute) or ``ptr_idx`` (for
1756 the ``pointer_with_type_tag`` attribute) should be a null pointer constant.
1757 The second argument to the ``type_tag_for_datatype`` attribute is ignored. For
1763 typedef int MPI_Datatype;
1764 extern struct mpi_datatype mpi_datatype_null
1765 __attribute__(( type_tag_for_datatype(mpi, void, must_be_null) ));
1767 #define MPI_DATATYPE_NULL ((MPI_Datatype) &mpi_datatype_null)
1768 int MPI_Send(void *buf, int count, MPI_Datatype datatype, ...)
1769 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1772 struct my_pair { double a; int b; };
1773 struct my_pair *buffer;
1774 MPI_Send(buffer, 1, MPI_DATATYPE_NULL /*, ... */); // warning: MPI_DATATYPE_NULL
1775 // was specified but buffer
1776 // is not a null pointer
1780 def FlattenDocs : Documentation {
1781 let Category = DocCatFunction;
1783 The ``flatten`` attribute causes calls within the attributed function to
1784 be inlined unless it is impossible to do so, for example if the body of the
1785 callee is unavailable or if the callee has the ``noinline`` attribute.
1789 def FormatDocs : Documentation {
1790 let Category = DocCatFunction;
1793 Clang supports the ``format`` attribute, which indicates that the function
1794 accepts a ``printf`` or ``scanf``-like format string and corresponding
1795 arguments or a ``va_list`` that contains these arguments.
1797 Please see `GCC documentation about format attribute
1798 <http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html>`_ to find details
1799 about attribute syntax.
1801 Clang implements two kinds of checks with this attribute.
1803 #. Clang checks that the function with the ``format`` attribute is called with
1804 a format string that uses format specifiers that are allowed, and that
1805 arguments match the format string. This is the ``-Wformat`` warning, it is
1808 #. Clang checks that the format string argument is a literal string. This is
1809 the ``-Wformat-nonliteral`` warning, it is off by default.
1811 Clang implements this mostly the same way as GCC, but there is a difference
1812 for functions that accept a ``va_list`` argument (for example, ``vprintf``).
1813 GCC does not emit ``-Wformat-nonliteral`` warning for calls to such
1814 functions. Clang does not warn if the format string comes from a function
1815 parameter, where the function is annotated with a compatible attribute,
1816 otherwise it warns. For example:
1820 __attribute__((__format__ (__scanf__, 1, 3)))
1821 void foo(const char* s, char *buf, ...) {
1825 vprintf(s, ap); // warning: format string is not a string literal
1828 In this case we warn because ``s`` contains a format string for a
1829 ``scanf``-like function, but it is passed to a ``printf``-like function.
1831 If the attribute is removed, clang still warns, because the format string is
1832 not a string literal.
1838 __attribute__((__format__ (__printf__, 1, 3)))
1839 void foo(const char* s, char *buf, ...) {
1843 vprintf(s, ap); // warning
1846 In this case Clang does not warn because the format string ``s`` and
1847 the corresponding arguments are annotated. If the arguments are
1848 incorrect, the caller of ``foo`` will receive a warning.
1852 def AlignValueDocs : Documentation {
1853 let Category = DocCatType;
1855 The align_value attribute can be added to the typedef of a pointer type or the
1856 declaration of a variable of pointer or reference type. It specifies that the
1857 pointer will point to, or the reference will bind to, only objects with at
1858 least the provided alignment. This alignment value must be some positive power
1863 typedef double * aligned_double_ptr __attribute__((align_value(64)));
1864 void foo(double & x __attribute__((align_value(128)),
1865 aligned_double_ptr y) { ... }
1867 If the pointer value does not have the specified alignment at runtime, the
1868 behavior of the program is undefined.
1872 def FlagEnumDocs : Documentation {
1873 let Category = DocCatType;
1875 This attribute can be added to an enumerator to signal to the compiler that it
1876 is intended to be used as a flag type. This will cause the compiler to assume
1877 that the range of the type includes all of the values that you can get by
1878 manipulating bits of the enumerator when issuing warnings.
1882 def EmptyBasesDocs : Documentation {
1883 let Category = DocCatType;
1885 The empty_bases attribute permits the compiler to utilize the
1886 empty-base-optimization more frequently.
1887 This attribute only applies to struct, class, and union types.
1888 It is only supported when using the Microsoft C++ ABI.
1892 def LayoutVersionDocs : Documentation {
1893 let Category = DocCatType;
1895 The layout_version attribute requests that the compiler utilize the class
1896 layout rules of a particular compiler version.
1897 This attribute only applies to struct, class, and union types.
1898 It is only supported when using the Microsoft C++ ABI.
1902 def MSInheritanceDocs : Documentation {
1903 let Category = DocCatType;
1904 let Heading = "__single_inhertiance, __multiple_inheritance, __virtual_inheritance";
1906 This collection of keywords is enabled under ``-fms-extensions`` and controls
1907 the pointer-to-member representation used on ``*-*-win32`` targets.
1909 The ``*-*-win32`` targets utilize a pointer-to-member representation which
1910 varies in size and alignment depending on the definition of the underlying
1913 However, this is problematic when a forward declaration is only available and
1914 no definition has been made yet. In such cases, Clang is forced to utilize the
1915 most general representation that is available to it.
1917 These keywords make it possible to use a pointer-to-member representation other
1918 than the most general one regardless of whether or not the definition will ever
1919 be present in the current translation unit.
1921 This family of keywords belong between the ``class-key`` and ``class-name``:
1925 struct __single_inheritance S;
1929 This keyword can be applied to class templates but only has an effect when used
1930 on full specializations:
1934 template <typename T, typename U> struct __single_inheritance A; // warning: inheritance model ignored on primary template
1935 template <typename T> struct __multiple_inheritance A<T, T>; // warning: inheritance model ignored on partial specialization
1936 template <> struct __single_inheritance A<int, float>;
1938 Note that choosing an inheritance model less general than strictly necessary is
1943 struct __multiple_inheritance S; // error: inheritance model does not match definition
1949 def MSNoVTableDocs : Documentation {
1950 let Category = DocCatType;
1952 This attribute can be added to a class declaration or definition to signal to
1953 the compiler that constructors and destructors will not reference the virtual
1954 function table. It is only supported when using the Microsoft C++ ABI.
1958 def OptnoneDocs : Documentation {
1959 let Category = DocCatFunction;
1961 The ``optnone`` attribute suppresses essentially all optimizations
1962 on a function or method, regardless of the optimization level applied to
1963 the compilation unit as a whole. This is particularly useful when you
1964 need to debug a particular function, but it is infeasible to build the
1965 entire application without optimization. Avoiding optimization on the
1966 specified function can improve the quality of the debugging information
1969 This attribute is incompatible with the ``always_inline`` and ``minsize``
1974 def LoopHintDocs : Documentation {
1975 let Category = DocCatStmt;
1976 let Heading = "#pragma clang loop";
1978 The ``#pragma clang loop`` directive allows loop optimization hints to be
1979 specified for the subsequent loop. The directive allows vectorization,
1980 interleaving, and unrolling to be enabled or disabled. Vector width as well
1981 as interleave and unrolling count can be manually specified. See
1982 `language extensions
1983 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
1988 def UnrollHintDocs : Documentation {
1989 let Category = DocCatStmt;
1990 let Heading = "#pragma unroll, #pragma nounroll";
1992 Loop unrolling optimization hints can be specified with ``#pragma unroll`` and
1993 ``#pragma nounroll``. The pragma is placed immediately before a for, while,
1994 do-while, or c++11 range-based for loop.
1996 Specifying ``#pragma unroll`` without a parameter directs the loop unroller to
1997 attempt to fully unroll the loop if the trip count is known at compile time and
1998 attempt to partially unroll the loop if the trip count is not known at compile
2008 Specifying the optional parameter, ``#pragma unroll _value_``, directs the
2009 unroller to unroll the loop ``_value_`` times. The parameter may optionally be
2010 enclosed in parentheses:
2024 Specifying ``#pragma nounroll`` indicates that the loop should not be unrolled:
2033 ``#pragma unroll`` and ``#pragma unroll _value_`` have identical semantics to
2034 ``#pragma clang loop unroll(full)`` and
2035 ``#pragma clang loop unroll_count(_value_)`` respectively. ``#pragma nounroll``
2036 is equivalent to ``#pragma clang loop unroll(disable)``. See
2037 `language extensions
2038 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
2039 for further details including limitations of the unroll hints.
2043 def OpenCLUnrollHintDocs : Documentation {
2044 let Category = DocCatStmt;
2045 let Heading = "__attribute__((opencl_unroll_hint))";
2047 The opencl_unroll_hint attribute qualifier can be used to specify that a loop
2048 (for, while and do loops) can be unrolled. This attribute qualifier can be
2049 used to specify full unrolling or partial unrolling by a specified amount.
2050 This is a compiler hint and the compiler may ignore this directive. See
2051 `OpenCL v2.0 <https://www.khronos.org/registry/cl/specs/opencl-2.0.pdf>`_
2052 s6.11.5 for details.
2056 def OpenCLAccessDocs : Documentation {
2057 let Category = DocCatStmt;
2058 let Heading = "__read_only, __write_only, __read_write (read_only, write_only, read_write)";
2060 The access qualifiers must be used with image object arguments or pipe arguments
2061 to declare if they are being read or written by a kernel or function.
2063 The read_only/__read_only, write_only/__write_only and read_write/__read_write
2064 names are reserved for use as access qualifiers and shall not be used otherwise.
2069 foo (read_only image2d_t imageA,
2070 write_only image2d_t imageB) {
2074 In the above example imageA is a read-only 2D image object, and imageB is a
2075 write-only 2D image object.
2077 The read_write (or __read_write) qualifier can not be used with pipe.
2079 More details can be found in the OpenCL C language Spec v2.0, Section 6.6.
2083 def DocOpenCLAddressSpaces : DocumentationCategory<"OpenCL Address Spaces"> {
2085 The address space qualifier may be used to specify the region of memory that is
2086 used to allocate the object. OpenCL supports the following address spaces:
2087 __generic(generic), __global(global), __local(local), __private(private),
2088 __constant(constant).
2092 __constant int c = ...;
2094 __generic int* foo(global int* g) {
2101 More details can be found in the OpenCL C language Spec v2.0, Section 6.5.
2105 def OpenCLAddressSpaceGenericDocs : Documentation {
2106 let Category = DocOpenCLAddressSpaces;
2108 The generic address space attribute is only available with OpenCL v2.0 and later.
2109 It can be used with pointer types. Variables in global and local scope and
2110 function parameters in non-kernel functions can have the generic address space
2111 type attribute. It is intended to be a placeholder for any other address space
2112 except for '__constant' in OpenCL code which can be used with multiple address
2117 def OpenCLAddressSpaceConstantDocs : Documentation {
2118 let Category = DocOpenCLAddressSpaces;
2120 The constant address space attribute signals that an object is located in
2121 a constant (non-modifiable) memory region. It is available to all work items.
2122 Any type can be annotated with the constant address space attribute. Objects
2123 with the constant address space qualifier can be declared in any scope and must
2124 have an initializer.
2128 def OpenCLAddressSpaceGlobalDocs : Documentation {
2129 let Category = DocOpenCLAddressSpaces;
2131 The global address space attribute specifies that an object is allocated in
2132 global memory, which is accessible by all work items. The content stored in this
2133 memory area persists between kernel executions. Pointer types to the global
2134 address space are allowed as function parameters or local variables. Starting
2135 with OpenCL v2.0, the global address space can be used with global (program
2136 scope) variables and static local variable as well.
2140 def OpenCLAddressSpaceLocalDocs : Documentation {
2141 let Category = DocOpenCLAddressSpaces;
2143 The local address space specifies that an object is allocated in the local (work
2144 group) memory area, which is accessible to all work items in the same work
2145 group. The content stored in this memory region is not accessible after
2146 the kernel execution ends. In a kernel function scope, any variable can be in
2147 the local address space. In other scopes, only pointer types to the local address
2148 space are allowed. Local address space variables cannot have an initializer.
2152 def OpenCLAddressSpacePrivateDocs : Documentation {
2153 let Category = DocOpenCLAddressSpaces;
2155 The private address space specifies that an object is allocated in the private
2156 (work item) memory. Other work items cannot access the same memory area and its
2157 content is destroyed after work item execution ends. Local variables can be
2158 declared in the private address space. Function arguments are always in the
2159 private address space. Kernel function arguments of a pointer or an array type
2160 cannot point to the private address space.
2164 def OpenCLNoSVMDocs : Documentation {
2165 let Category = DocCatVariable;
2167 OpenCL 2.0 supports the optional ``__attribute__((nosvm))`` qualifier for
2168 pointer variable. It informs the compiler that the pointer does not refer
2169 to a shared virtual memory region. See OpenCL v2.0 s6.7.2 for details.
2171 Since it is not widely used and has been removed from OpenCL 2.1, it is ignored
2175 def NullabilityDocs : DocumentationCategory<"Nullability Attributes"> {
2177 Whether a particular pointer may be "null" is an important concern when working with pointers in the C family of languages. The various nullability attributes indicate whether a particular pointer can be null or not, which makes APIs more expressive and can help static analysis tools identify bugs involving null pointers. Clang supports several kinds of nullability attributes: the ``nonnull`` and ``returns_nonnull`` attributes indicate which function or method parameters and result types can never be null, while nullability type qualifiers indicate which pointer types can be null (``_Nullable``) or cannot be null (``_Nonnull``).
2179 The nullability (type) qualifiers express whether a value of a given pointer type can be null (the ``_Nullable`` qualifier), doesn't have a defined meaning for null (the ``_Nonnull`` qualifier), or for which the purpose of null is unclear (the ``_Null_unspecified`` qualifier). Because nullability qualifiers are expressed within the type system, they are more general than the ``nonnull`` and ``returns_nonnull`` attributes, allowing one to express (for example) a nullable pointer to an array of nonnull pointers. Nullability qualifiers are written to the right of the pointer to which they apply. For example:
2183 // No meaningful result when 'ptr' is null (here, it happens to be undefined behavior).
2184 int fetch(int * _Nonnull ptr) { return *ptr; }
2186 // 'ptr' may be null.
2187 int fetch_or_zero(int * _Nullable ptr) {
2188 return ptr ? *ptr : 0;
2191 // A nullable pointer to non-null pointers to const characters.
2192 const char *join_strings(const char * _Nonnull * _Nullable strings, unsigned n);
2194 In Objective-C, there is an alternate spelling for the nullability qualifiers that can be used in Objective-C methods and properties using context-sensitive, non-underscored keywords. For example:
2196 .. code-block:: objective-c
2198 @interface NSView : NSResponder
2199 - (nullable NSView *)ancestorSharedWithView:(nonnull NSView *)aView;
2200 @property (assign, nullable) NSView *superview;
2201 @property (readonly, nonnull) NSArray *subviews;
2206 def TypeNonNullDocs : Documentation {
2207 let Category = NullabilityDocs;
2209 The ``_Nonnull`` nullability qualifier indicates that null is not a meaningful value for a value of the ``_Nonnull`` pointer type. For example, given a declaration such as:
2213 int fetch(int * _Nonnull ptr);
2215 a caller of ``fetch`` should not provide a null value, and the compiler will produce a warning if it sees a literal null value passed to ``fetch``. Note that, unlike the declaration attribute ``nonnull``, the presence of ``_Nonnull`` does not imply that passing null is undefined behavior: ``fetch`` is free to consider null undefined behavior or (perhaps for backward-compatibility reasons) defensively handle null.
2219 def TypeNullableDocs : Documentation {
2220 let Category = NullabilityDocs;
2222 The ``_Nullable`` nullability qualifier indicates that a value of the ``_Nullable`` pointer type can be null. For example, given:
2226 int fetch_or_zero(int * _Nullable ptr);
2228 a caller of ``fetch_or_zero`` can provide null.
2232 def TypeNullUnspecifiedDocs : Documentation {
2233 let Category = NullabilityDocs;
2235 The ``_Null_unspecified`` nullability qualifier indicates that neither the ``_Nonnull`` nor ``_Nullable`` qualifiers make sense for a particular pointer type. It is used primarily to indicate that the role of null with specific pointers in a nullability-annotated header is unclear, e.g., due to overly-complex implementations or historical factors with a long-lived API.
2239 def NonNullDocs : Documentation {
2240 let Category = NullabilityDocs;
2242 The ``nonnull`` attribute indicates that some function parameters must not be null, and can be used in several different ways. It's original usage (`from GCC <https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#Common-Function-Attributes>`_) is as a function (or Objective-C method) attribute that specifies which parameters of the function are nonnull in a comma-separated list. For example:
2246 extern void * my_memcpy (void *dest, const void *src, size_t len)
2247 __attribute__((nonnull (1, 2)));
2249 Here, the ``nonnull`` attribute indicates that parameters 1 and 2
2250 cannot have a null value. Omitting the parenthesized list of parameter indices means that all parameters of pointer type cannot be null:
2254 extern void * my_memcpy (void *dest, const void *src, size_t len)
2255 __attribute__((nonnull));
2257 Clang also allows the ``nonnull`` attribute to be placed directly on a function (or Objective-C method) parameter, eliminating the need to specify the parameter index ahead of type. For example:
2261 extern void * my_memcpy (void *dest __attribute__((nonnull)),
2262 const void *src __attribute__((nonnull)), size_t len);
2264 Note that the ``nonnull`` attribute indicates that passing null to a non-null parameter is undefined behavior, which the optimizer may take advantage of to, e.g., remove null checks. The ``_Nonnull`` type qualifier indicates that a pointer cannot be null in a more general manner (because it is part of the type system) and does not imply undefined behavior, making it more widely applicable.
2268 def ReturnsNonNullDocs : Documentation {
2269 let Category = NullabilityDocs;
2271 The ``returns_nonnull`` attribute indicates that a particular function (or Objective-C method) always returns a non-null pointer. For example, a particular system ``malloc`` might be defined to terminate a process when memory is not available rather than returning a null pointer:
2275 extern void * malloc (size_t size) __attribute__((returns_nonnull));
2277 The ``returns_nonnull`` attribute implies that returning a null pointer is undefined behavior, which the optimizer may take advantage of. The ``_Nonnull`` type qualifier indicates that a pointer cannot be null in a more general manner (because it is part of the type system) and does not imply undefined behavior, making it more widely applicable
2281 def NoAliasDocs : Documentation {
2282 let Category = DocCatFunction;
2284 The ``noalias`` attribute indicates that the only memory accesses inside
2285 function are loads and stores from objects pointed to by its pointer-typed
2286 arguments, with arbitrary offsets.
2290 def OMPDeclareSimdDocs : Documentation {
2291 let Category = DocCatFunction;
2292 let Heading = "#pragma omp declare simd";
2294 The `declare simd` construct can be applied to a function to enable the creation
2295 of one or more versions that can process multiple arguments using SIMD
2296 instructions from a single invocation in a SIMD loop. The `declare simd`
2297 directive is a declarative directive. There may be multiple `declare simd`
2298 directives for a function. The use of a `declare simd` construct on a function
2299 enables the creation of SIMD versions of the associated function that can be
2300 used to process multiple arguments from a single invocation from a SIMD loop
2302 The syntax of the `declare simd` construct is as follows:
2306 #pragma omp declare simd [clause[[,] clause] ...] new-line
2307 [#pragma omp declare simd [clause[[,] clause] ...] new-line]
2309 function definition or declaration
2311 where clause is one of the following:
2316 linear(argument-list[:constant-linear-step])
2317 aligned(argument-list[:alignment])
2318 uniform(argument-list)
2325 def OMPDeclareTargetDocs : Documentation {
2326 let Category = DocCatFunction;
2327 let Heading = "#pragma omp declare target";
2329 The `declare target` directive specifies that variables and functions are mapped
2330 to a device for OpenMP offload mechanism.
2332 The syntax of the declare target directive is as follows:
2336 #pragma omp declare target new-line
2337 declarations-definition-seq
2338 #pragma omp end declare target new-line
2342 def NotTailCalledDocs : Documentation {
2343 let Category = DocCatFunction;
2345 The ``not_tail_called`` attribute prevents tail-call optimization on statically bound calls. It has no effect on indirect calls. Virtual functions, objective-c methods, and functions marked as ``always_inline`` cannot be marked as ``not_tail_called``.
2347 For example, it prevents tail-call optimization in the following case:
2351 int __attribute__((not_tail_called)) foo1(int);
2354 return foo1(a); // No tail-call optimization on direct calls.
2357 However, it doesn't prevent tail-call optimization in this case:
2361 int __attribute__((not_tail_called)) foo1(int);
2364 int (*fn)(int) = &foo1;
2366 // not_tail_called has no effect on an indirect call even if the call can be
2367 // resolved at compile time.
2371 Marking virtual functions as ``not_tail_called`` is an error:
2377 // not_tail_called on a virtual function is an error.
2378 [[clang::not_tail_called]] virtual int foo1();
2382 // Non-virtual functions can be marked ``not_tail_called``.
2383 [[clang::not_tail_called]] int foo3();
2386 class Derived1 : public Base {
2388 int foo1() override;
2390 // not_tail_called on a virtual function is an error.
2391 [[clang::not_tail_called]] int foo2() override;
2396 def InternalLinkageDocs : Documentation {
2397 let Category = DocCatFunction;
2399 The ``internal_linkage`` attribute changes the linkage type of the declaration to internal.
2400 This is similar to C-style ``static``, but can be used on classes and class methods. When applied to a class definition,
2401 this attribute affects all methods and static data members of that class.
2402 This can be used to contain the ABI of a C++ library by excluding unwanted class methods from the export tables.
2406 def DisableTailCallsDocs : Documentation {
2407 let Category = DocCatFunction;
2409 The ``disable_tail_calls`` attribute instructs the backend to not perform tail call optimization inside the marked function.
2417 int foo(int a) __attribute__((disable_tail_calls)) {
2418 return callee(a); // This call is not tail-call optimized.
2421 Marking virtual functions as ``disable_tail_calls`` is legal.
2429 [[clang::disable_tail_calls]] virtual int foo1() {
2430 return callee(); // This call is not tail-call optimized.
2434 class Derived1 : public Base {
2436 int foo1() override {
2437 return callee(); // This call is tail-call optimized.
2444 def AnyX86InterruptDocs : Documentation {
2445 let Category = DocCatFunction;
2447 Clang supports the GNU style ``__attribute__((interrupt))`` attribute on
2448 x86/x86-64 targets.The compiler generates function entry and exit sequences
2449 suitable for use in an interrupt handler when this attribute is present.
2450 The 'IRET' instruction, instead of the 'RET' instruction, is used to return
2451 from interrupt or exception handlers. All registers, except for the EFLAGS
2452 register which is restored by the 'IRET' instruction, are preserved by the
2455 Any interruptible-without-stack-switch code must be compiled with
2456 -mno-red-zone since interrupt handlers can and will, because of the
2457 hardware design, touch the red zone.
2459 1. interrupt handler must be declared with a mandatory pointer argument:
2463 struct interrupt_frame
2472 __attribute__ ((interrupt))
2473 void f (struct interrupt_frame *frame) {
2477 2. exception handler:
2479 The exception handler is very similar to the interrupt handler with
2480 a different mandatory function signature:
2484 __attribute__ ((interrupt))
2485 void f (struct interrupt_frame *frame, uword_t error_code) {
2489 and compiler pops 'ERROR_CODE' off stack before the 'IRET' instruction.
2491 The exception handler should only be used for exceptions which push an
2492 error code and all other exceptions must use the interrupt handler.
2493 The system will crash if the wrong handler is used.
2497 def SwiftCallDocs : Documentation {
2498 let Category = DocCatVariable;
2500 The ``swiftcall`` attribute indicates that a function should be called
2501 using the Swift calling convention for a function or function pointer.
2503 The lowering for the Swift calling convention, as described by the Swift
2504 ABI documentation, occurs in multiple phases. The first, "high-level"
2505 phase breaks down the formal parameters and results into innately direct
2506 and indirect components, adds implicit paraameters for the generic
2507 signature, and assigns the context and error ABI treatments to parameters
2508 where applicable. The second phase breaks down the direct parameters
2509 and results from the first phase and assigns them to registers or the
2510 stack. The ``swiftcall`` convention only handles this second phase of
2511 lowering; the C function type must accurately reflect the results
2512 of the first phase, as follows:
2514 - Results classified as indirect by high-level lowering should be
2515 represented as parameters with the ``swift_indirect_result`` attribute.
2517 - Results classified as direct by high-level lowering should be represented
2520 - First, remove any empty direct results.
2522 - If there are no direct results, the C result type should be ``void``.
2524 - If there is one direct result, the C result type should be a type with
2525 the exact layout of that result type.
2527 - If there are a multiple direct results, the C result type should be
2528 a struct type with the exact layout of a tuple of those results.
2530 - Parameters classified as indirect by high-level lowering should be
2531 represented as parameters of pointer type.
2533 - Parameters classified as direct by high-level lowering should be
2534 omitted if they are empty types; otherwise, they should be represented
2535 as a parameter type with a layout exactly matching the layout of the
2536 Swift parameter type.
2538 - The context parameter, if present, should be represented as a trailing
2539 parameter with the ``swift_context`` attribute.
2541 - The error result parameter, if present, should be represented as a
2542 trailing parameter (always following a context parameter) with the
2543 ``swift_error_result`` attribute.
2545 ``swiftcall`` does not support variadic arguments or unprototyped functions.
2547 The parameter ABI treatment attributes are aspects of the function type.
2548 A function type which which applies an ABI treatment attribute to a
2549 parameter is a different type from an otherwise-identical function type
2550 that does not. A single parameter may not have multiple ABI treatment
2553 Support for this feature is target-dependent, although it should be
2554 supported on every target that Swift supports. Query for this support
2555 with ``__has_attribute(swiftcall)``. This implies support for the
2556 ``swift_context``, ``swift_error_result``, and ``swift_indirect_result``
2561 def SwiftContextDocs : Documentation {
2562 let Category = DocCatVariable;
2564 The ``swift_context`` attribute marks a parameter of a ``swiftcall``
2565 function as having the special context-parameter ABI treatment.
2567 This treatment generally passes the context value in a special register
2568 which is normally callee-preserved.
2570 A ``swift_context`` parameter must either be the last parameter or must be
2571 followed by a ``swift_error_result`` parameter (which itself must always be
2572 the last parameter).
2574 A context parameter must have pointer or reference type.
2578 def SwiftErrorResultDocs : Documentation {
2579 let Category = DocCatVariable;
2581 The ``swift_error_result`` attribute marks a parameter of a ``swiftcall``
2582 function as having the special error-result ABI treatment.
2584 This treatment generally passes the underlying error value in and out of
2585 the function through a special register which is normally callee-preserved.
2586 This is modeled in C by pretending that the register is addressable memory:
2588 - The caller appears to pass the address of a variable of pointer type.
2589 The current value of this variable is copied into the register before
2590 the call; if the call returns normally, the value is copied back into the
2593 - The callee appears to receive the address of a variable. This address
2594 is actually a hidden location in its own stack, initialized with the
2595 value of the register upon entry. When the function returns normally,
2596 the value in that hidden location is written back to the register.
2598 A ``swift_error_result`` parameter must be the last parameter, and it must be
2599 preceded by a ``swift_context`` parameter.
2601 A ``swift_error_result`` parameter must have type ``T**`` or ``T*&`` for some
2602 type T. Note that no qualifiers are permitted on the intermediate level.
2604 It is undefined behavior if the caller does not pass a pointer or
2605 reference to a valid object.
2607 The standard convention is that the error value itself (that is, the
2608 value stored in the apparent argument) will be null upon function entry,
2609 but this is not enforced by the ABI.
2613 def SwiftIndirectResultDocs : Documentation {
2614 let Category = DocCatVariable;
2616 The ``swift_indirect_result`` attribute marks a parameter of a ``swiftcall``
2617 function as having the special indirect-result ABI treatment.
2619 This treatment gives the parameter the target's normal indirect-result
2620 ABI treatment, which may involve passing it differently from an ordinary
2621 parameter. However, only the first indirect result will receive this
2622 treatment. Furthermore, low-level lowering may decide that a direct result
2623 must be returned indirectly; if so, this will take priority over the
2624 ``swift_indirect_result`` parameters.
2626 A ``swift_indirect_result`` parameter must either be the first parameter or
2627 follow another ``swift_indirect_result`` parameter.
2629 A ``swift_indirect_result`` parameter must have type ``T*`` or ``T&`` for
2630 some object type ``T``. If ``T`` is a complete type at the point of
2631 definition of a function, it is undefined behavior if the argument
2632 value does not point to storage of adequate size and alignment for a
2633 value of type ``T``.
2635 Making indirect results explicit in the signature allows C functions to
2636 directly construct objects into them without relying on language
2637 optimizations like C++'s named return value optimization (NRVO).
2641 def AbiTagsDocs : Documentation {
2642 let Category = DocCatFunction;
2644 The ``abi_tag`` attribute can be applied to a function, variable, class or
2645 inline namespace declaration to modify the mangled name of the entity. It gives
2646 the ability to distinguish between different versions of the same entity but
2647 with different ABI versions supported. For example, a newer version of a class
2648 could have a different set of data members and thus have a different size. Using
2649 the ``abi_tag`` attribute, it is possible to have different mangled names for
2650 a global variable of the class type. Therefor, the old code could keep using
2651 the old manged name and the new code will use the new mangled name with tags.
2655 def PreserveMostDocs : Documentation {
2656 let Category = DocCatCallingConvs;
2658 On X86-64 and AArch64 targets, this attribute changes the calling convention of
2659 a function. The ``preserve_most`` calling convention attempts to make the code
2660 in the caller as unintrusive as possible. This convention behaves identically
2661 to the ``C`` calling convention on how arguments and return values are passed,
2662 but it uses a different set of caller/callee-saved registers. This alleviates
2663 the burden of saving and recovering a large register set before and after the
2664 call in the caller. If the arguments are passed in callee-saved registers,
2665 then they will be preserved by the callee across the call. This doesn't
2666 apply for values returned in callee-saved registers.
2668 - On X86-64 the callee preserves all general purpose registers, except for
2669 R11. R11 can be used as a scratch register. Floating-point registers
2670 (XMMs/YMMs) are not preserved and need to be saved by the caller.
2672 The idea behind this convention is to support calls to runtime functions
2673 that have a hot path and a cold path. The hot path is usually a small piece
2674 of code that doesn't use many registers. The cold path might need to call out to
2675 another function and therefore only needs to preserve the caller-saved
2676 registers, which haven't already been saved by the caller. The
2677 `preserve_most` calling convention is very similar to the ``cold`` calling
2678 convention in terms of caller/callee-saved registers, but they are used for
2679 different types of function calls. ``coldcc`` is for function calls that are
2680 rarely executed, whereas `preserve_most` function calls are intended to be
2681 on the hot path and definitely executed a lot. Furthermore ``preserve_most``
2682 doesn't prevent the inliner from inlining the function call.
2684 This calling convention will be used by a future version of the Objective-C
2685 runtime and should therefore still be considered experimental at this time.
2686 Although this convention was created to optimize certain runtime calls to
2687 the Objective-C runtime, it is not limited to this runtime and might be used
2688 by other runtimes in the future too. The current implementation only
2689 supports X86-64 and AArch64, but the intention is to support more architectures
2694 def PreserveAllDocs : Documentation {
2695 let Category = DocCatCallingConvs;
2697 On X86-64 and AArch64 targets, this attribute changes the calling convention of
2698 a function. The ``preserve_all`` calling convention attempts to make the code
2699 in the caller even less intrusive than the ``preserve_most`` calling convention.
2700 This calling convention also behaves identical to the ``C`` calling convention
2701 on how arguments and return values are passed, but it uses a different set of
2702 caller/callee-saved registers. This removes the burden of saving and
2703 recovering a large register set before and after the call in the caller. If
2704 the arguments are passed in callee-saved registers, then they will be
2705 preserved by the callee across the call. This doesn't apply for values
2706 returned in callee-saved registers.
2708 - On X86-64 the callee preserves all general purpose registers, except for
2709 R11. R11 can be used as a scratch register. Furthermore it also preserves
2710 all floating-point registers (XMMs/YMMs).
2712 The idea behind this convention is to support calls to runtime functions
2713 that don't need to call out to any other functions.
2715 This calling convention, like the ``preserve_most`` calling convention, will be
2716 used by a future version of the Objective-C runtime and should be considered
2717 experimental at this time.
2721 def DeprecatedDocs : Documentation {
2722 let Category = DocCatFunction;
2724 The ``deprecated`` attribute can be applied to a function, a variable, or a
2725 type. This is useful when identifying functions, variables, or types that are
2726 expected to be removed in a future version of a program.
2728 Consider the function declaration for a hypothetical function ``f``:
2732 void f(void) __attribute__((deprecated("message", "replacement")));
2734 When spelled as `__attribute__((deprecated))`, the deprecated attribute can have
2735 two optional string arguments. The first one is the message to display when
2736 emitting the warning; the second one enables the compiler to provide a Fix-It
2737 to replace the deprecated name with a new name. Otherwise, when spelled as
2738 `[[gnu::deprecated]] or [[deprecated]]`, the attribute can have one optional
2739 string argument which is the message to display when emitting the warning.
2743 def IFuncDocs : Documentation {
2744 let Category = DocCatFunction;
2746 ``__attribute__((ifunc("resolver")))`` is used to mark that the address of a declaration should be resolved at runtime by calling a resolver function.
2748 The symbol name of the resolver function is given in quotes. A function with this name (after mangling) must be defined in the current translation unit; it may be ``static``. The resolver function should take no arguments and return a pointer.
2750 The ``ifunc`` attribute may only be used on a function declaration. A function declaration with an ``ifunc`` attribute is considered to be a definition of the declared entity. The entity must not have weak linkage; for example, in C++, it cannot be applied to a declaration if a definition at that location would be considered inline.
2752 Not all targets support this attribute. ELF targets support this attribute when using binutils v2.20.1 or higher and glibc v2.11.1 or higher. Non-ELF targets currently do not support this attribute.
2756 def LTOVisibilityDocs : Documentation {
2757 let Category = DocCatType;
2759 See :doc:`LTOVisibility`.
2763 def RenderScriptKernelAttributeDocs : Documentation {
2764 let Category = DocCatFunction;
2766 ``__attribute__((kernel))`` is used to mark a ``kernel`` function in
2769 In RenderScript, ``kernel`` functions are used to express data-parallel
2770 computations. The RenderScript runtime efficiently parallelizes ``kernel``
2771 functions to run on computational resources such as multi-core CPUs and GPUs.
2772 See the RenderScript_ documentation for more information.
2774 .. _RenderScript: https://developer.android.com/guide/topics/renderscript/compute.html
2778 def XRayDocs : Documentation {
2779 let Category = DocCatFunction;
2780 let Heading = "xray_always_instrument (clang::xray_always_instrument), xray_never_instrument (clang::xray_never_instrument)";
2782 ``__attribute__((xray_always_instrument))`` or ``[[clang::xray_always_instrument]]`` is used to mark member functions (in C++), methods (in Objective C), and free functions (in C, C++, and Objective C) to be instrumented with XRay. This will cause the function to always have space at the beginning and exit points to allow for runtime patching.
2784 Conversely, ``__attribute__((xray_never_instrument))`` or ``[[clang::xray_never_instrument]]`` will inhibit the insertion of these instrumentation points.
2786 If a function has neither of these attributes, they become subject to the XRay heuristics used to determine whether a function should be instrumented or otherwise.
2790 def TransparentUnionDocs : Documentation {
2791 let Category = DocCatType;
2793 This attribute can be applied to a union to change the behaviour of calls to
2794 functions that have an argument with a transparent union type. The compiler
2795 behaviour is changed in the following manner:
2797 - A value whose type is any member of the transparent union can be passed as an
2798 argument without the need to cast that value.
2800 - The argument is passed to the function using the calling convention of the
2801 first member of the transparent union. Consequently, all the members of the
2802 transparent union should have the same calling convention as its first member.
2804 Transparent unions are not supported in C++.
2808 def ObjCSubclassingRestrictedDocs : Documentation {
2809 let Category = DocCatType;
2811 This attribute can be added to an Objective-C ``@interface`` declaration to
2812 ensure that this class cannot be subclassed.