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 ThreadDocs : Documentation {
71 let Category = DocCatVariable;
73 The ``__declspec(thread)`` attribute declares a variable with thread local
74 storage. It is available under the ``-fms-extensions`` flag for MSVC
75 compatibility. See the documentation for `__declspec(thread)`_ on MSDN.
77 .. _`__declspec(thread)`: http://msdn.microsoft.com/en-us/library/9w1sdazb.aspx
79 In Clang, ``__declspec(thread)`` is generally equivalent in functionality to the
80 GNU ``__thread`` keyword. The variable must not have a destructor and must have
81 a constant initializer, if any. The attribute only applies to variables
82 declared with static storage duration, such as globals, class static data
83 members, and static locals.
87 def CarriesDependencyDocs : Documentation {
88 let Category = DocCatFunction;
90 The ``carries_dependency`` attribute specifies dependency propagation into and
93 When specified on a function or Objective-C method, the ``carries_dependency``
94 attribute means that the return value carries a dependency out of the function,
95 so that the implementation need not constrain ordering upon return from that
96 function. Implementations of the function and its caller may choose to preserve
97 dependencies instead of emitting memory ordering instructions such as fences.
99 Note, this attribute does not change the meaning of the program, but may result
100 in generation of more efficient code.
104 def C11NoReturnDocs : Documentation {
105 let Category = DocCatFunction;
107 A function declared as ``_Noreturn`` shall not return to its caller. The
108 compiler will generate a diagnostic for a function declared as ``_Noreturn``
109 that appears to be capable of returning to its caller.
113 def CXX11NoReturnDocs : Documentation {
114 let Category = DocCatFunction;
116 A function declared as ``[[noreturn]]`` shall not return to its caller. The
117 compiler will generate a diagnostic for a function declared as ``[[noreturn]]``
118 that appears to be capable of returning to its caller.
122 def AssertCapabilityDocs : Documentation {
123 let Category = DocCatFunction;
124 let Heading = "assert_capability (assert_shared_capability, clang::assert_capability, clang::assert_shared_capability)";
126 Marks a function that dynamically tests whether a capability is held, and halts
127 the program if it is not held.
131 def AcquireCapabilityDocs : Documentation {
132 let Category = DocCatFunction;
133 let Heading = "acquire_capability (acquire_shared_capability, clang::acquire_capability, clang::acquire_shared_capability)";
135 Marks a function as acquiring a capability.
139 def TryAcquireCapabilityDocs : Documentation {
140 let Category = DocCatFunction;
141 let Heading = "try_acquire_capability (try_acquire_shared_capability, clang::try_acquire_capability, clang::try_acquire_shared_capability)";
143 Marks a function that attempts to acquire a capability. This function may fail to
144 actually acquire the capability; they accept a Boolean value determining
145 whether acquiring the capability means success (true), or failing to acquire
146 the capability means success (false).
150 def ReleaseCapabilityDocs : Documentation {
151 let Category = DocCatFunction;
152 let Heading = "release_capability (release_shared_capability, clang::release_capability, clang::release_shared_capability)";
154 Marks a function as releasing a capability.
158 def AssumeAlignedDocs : Documentation {
159 let Category = DocCatFunction;
161 Use ``__attribute__((assume_aligned(<alignment>[,<offset>]))`` on a function
162 declaration to specify that the return value of the function (which must be a
163 pointer type) has the specified offset, in bytes, from an address with the
164 specified alignment. The offset is taken to be zero if omitted.
168 // The returned pointer value has 32-byte alignment.
169 void *a() __attribute__((assume_aligned (32)));
171 // The returned pointer value is 4 bytes greater than an address having
172 // 32-byte alignment.
173 void *b() __attribute__((assume_aligned (32, 4)));
175 Note that this attribute provides information to the compiler regarding a
176 condition that the code already ensures is true. It does not cause the compiler
177 to enforce the provided alignment assumption.
181 def EnableIfDocs : Documentation {
182 let Category = DocCatFunction;
184 .. Note:: Some features of this attribute are experimental. The meaning of
\r
185 multiple enable_if attributes on a single declaration is subject to change in
\r
186 a future version of clang. Also, the ABI is not standardized and the name
\r
187 mangling may change in future versions. To avoid that, use asm labels.
189 The ``enable_if`` attribute can be placed on function declarations to control
190 which overload is selected based on the values of the function's arguments.
191 When combined with the ``overloadable`` attribute, this feature is also
197 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")));
202 isdigit(-10); // results in a compile-time error.
205 The enable_if attribute takes two arguments, the first is an expression written
206 in terms of the function parameters, the second is a string explaining why this
207 overload candidate could not be selected to be displayed in diagnostics. The
208 expression is part of the function signature for the purposes of determining
209 whether it is a redeclaration (following the rules used when determining
210 whether a C++ template specialization is ODR-equivalent), but is not part of
213 The enable_if expression is evaluated as if it were the body of a
214 bool-returning constexpr function declared with the arguments of the function
215 it is being applied to, then called with the parameters at the call site. If the
216 result is false or could not be determined through constant expression
217 evaluation, then this overload will not be chosen and the provided string may
218 be used in a diagnostic if the compile fails as a result.
220 Because the enable_if expression is an unevaluated context, there are no global
221 state changes, nor the ability to pass information from the enable_if
222 expression to the function body. For example, suppose we want calls to
223 strnlen(strbuf, maxlen) to resolve to strnlen_chk(strbuf, maxlen, size of
224 strbuf) only if the size of strbuf can be determined:
228 __attribute__((always_inline))
229 static inline size_t strnlen(const char *s, size_t maxlen)
230 __attribute__((overloadable))
231 __attribute__((enable_if(__builtin_object_size(s, 0) != -1))),
232 "chosen when the buffer size is known but 'maxlen' is not")))
234 return strnlen_chk(s, maxlen, __builtin_object_size(s, 0));
237 Multiple enable_if attributes may be applied to a single declaration. In this
238 case, the enable_if expressions are evaluated from left to right in the
239 following manner. First, the candidates whose enable_if expressions evaluate to
240 false or cannot be evaluated are discarded. If the remaining candidates do not
241 share ODR-equivalent enable_if expressions, the overload resolution is
242 ambiguous. Otherwise, enable_if overload resolution continues with the next
243 enable_if attribute on the candidates that have not been discarded and have
244 remaining enable_if attributes. In this way, we pick the most specific
245 overload out of a number of viable overloads using enable_if.
249 void f() __attribute__((enable_if(true, ""))); // #1
250 void f() __attribute__((enable_if(true, ""))) __attribute__((enable_if(true, ""))); // #2
252 void g(int i, int j) __attribute__((enable_if(i, ""))); // #1
253 void g(int i, int j) __attribute__((enable_if(j, ""))) __attribute__((enable_if(true))); // #2
255 In this example, a call to f() is always resolved to #2, as the first enable_if
256 expression is ODR-equivalent for both declarations, but #1 does not have another
257 enable_if expression to continue evaluating, so the next round of evaluation has
258 only a single candidate. In a call to g(1, 1), the call is ambiguous even though
259 #2 has more enable_if attributes, because the first enable_if expressions are
262 Query for this feature with ``__has_attribute(enable_if)``.
266 def OverloadableDocs : Documentation {
267 let Category = DocCatFunction;
269 Clang provides support for C++ function overloading in C. Function overloading
270 in C is introduced using the ``overloadable`` attribute. For example, one
271 might provide several overloaded versions of a ``tgsin`` function that invokes
272 the appropriate standard function computing the sine of a value with ``float``,
273 ``double``, or ``long double`` precision:
278 float __attribute__((overloadable)) tgsin(float x) { return sinf(x); }
279 double __attribute__((overloadable)) tgsin(double x) { return sin(x); }
280 long double __attribute__((overloadable)) tgsin(long double x) { return sinl(x); }
282 Given these declarations, one can call ``tgsin`` with a ``float`` value to
283 receive a ``float`` result, with a ``double`` to receive a ``double`` result,
284 etc. Function overloading in C follows the rules of C++ function overloading
285 to pick the best overload given the call arguments, with a few C-specific
288 * Conversion from ``float`` or ``double`` to ``long double`` is ranked as a
289 floating-point promotion (per C99) rather than as a floating-point conversion
292 * A conversion from a pointer of type ``T*`` to a pointer of type ``U*`` is
293 considered a pointer conversion (with conversion rank) if ``T`` and ``U`` are
296 * A conversion from type ``T`` to a value of type ``U`` is permitted if ``T``
297 and ``U`` are compatible types. This conversion is given "conversion" rank.
299 The declaration of ``overloadable`` functions is restricted to function
300 declarations and definitions. Most importantly, if any function with a given
301 name is given the ``overloadable`` attribute, then all function declarations
302 and definitions with that name (and in that scope) must have the
303 ``overloadable`` attribute. This rule even applies to redeclarations of
304 functions whose original declaration had the ``overloadable`` attribute, e.g.,
308 int f(int) __attribute__((overloadable));
309 float f(float); // error: declaration of "f" must have the "overloadable" attribute
311 int g(int) __attribute__((overloadable));
312 int g(int) { } // error: redeclaration of "g" must also have the "overloadable" attribute
314 Functions marked ``overloadable`` must have prototypes. Therefore, the
315 following code is ill-formed:
319 int h() __attribute__((overloadable)); // error: h does not have a prototype
321 However, ``overloadable`` functions are allowed to use a ellipsis even if there
322 are no named parameters (as is permitted in C++). This feature is particularly
323 useful when combined with the ``unavailable`` attribute:
327 void honeypot(...) __attribute__((overloadable, unavailable)); // calling me is an error
329 Functions declared with the ``overloadable`` attribute have their names mangled
330 according to the same rules as C++ function names. For example, the three
331 ``tgsin`` functions in our motivating example get the mangled names
332 ``_Z5tgsinf``, ``_Z5tgsind``, and ``_Z5tgsine``, respectively. There are two
333 caveats to this use of name mangling:
335 * Future versions of Clang may change the name mangling of functions overloaded
336 in C, so you should not depend on an specific mangling. To be completely
337 safe, we strongly urge the use of ``static inline`` with ``overloadable``
340 * The ``overloadable`` attribute has almost no meaning when used in C++,
341 because names will already be mangled and functions are already overloadable.
342 However, when an ``overloadable`` function occurs within an ``extern "C"``
343 linkage specification, it's name *will* be mangled in the same way as it
346 Query for this feature with ``__has_extension(attribute_overloadable)``.
350 def ObjCMethodFamilyDocs : Documentation {
351 let Category = DocCatFunction;
353 Many methods in Objective-C have conventional meanings determined by their
354 selectors. It is sometimes useful to be able to mark a method as having a
355 particular conventional meaning despite not having the right selector, or as
356 not having the conventional meaning that its selector would suggest. For these
357 use cases, we provide an attribute to specifically describe the "method family"
358 that a method belongs to.
360 **Usage**: ``__attribute__((objc_method_family(X)))``, where ``X`` is one of
361 ``none``, ``alloc``, ``copy``, ``init``, ``mutableCopy``, or ``new``. This
362 attribute can only be placed at the end of a method declaration:
366 - (NSString *)initMyStringValue __attribute__((objc_method_family(none)));
368 Users who do not wish to change the conventional meaning of a method, and who
369 merely want to document its non-standard retain and release semantics, should
370 use the retaining behavior attributes (``ns_returns_retained``,
371 ``ns_returns_not_retained``, etc).
373 Query for this feature with ``__has_attribute(objc_method_family)``.
377 def NoDuplicateDocs : Documentation {
378 let Category = DocCatFunction;
380 The ``noduplicate`` attribute can be placed on function declarations to control
381 whether function calls to this function can be duplicated or not as a result of
382 optimizations. This is required for the implementation of functions with
383 certain special requirements, like the OpenCL "barrier" function, that might
384 need to be run concurrently by all the threads that are executing in lockstep
385 on the hardware. For example this attribute applied on the function
386 "nodupfunc" in the code below avoids that:
390 void nodupfunc() __attribute__((noduplicate));
391 // Setting it as a C++11 attribute is also valid
392 // void nodupfunc() [[clang::noduplicate]];
403 gets possibly modified by some optimizations into code similar to this:
415 where the call to "nodupfunc" is duplicated and sunk into the two branches
420 def NoSplitStackDocs : Documentation {
421 let Category = DocCatFunction;
423 The ``no_split_stack`` attribute disables the emission of the split stack
424 preamble for a particular function. It has no effect if ``-fsplit-stack``
429 def ObjCRequiresSuperDocs : Documentation {
430 let Category = DocCatFunction;
432 Some Objective-C classes allow a subclass to override a particular method in a
433 parent class but expect that the overriding method also calls the overridden
434 method in the parent class. For these cases, we provide an attribute to
435 designate that a method requires a "call to ``super``" in the overriding
436 method in the subclass.
438 **Usage**: ``__attribute__((objc_requires_super))``. This attribute can only
439 be placed at the end of a method declaration:
443 - (void)foo __attribute__((objc_requires_super));
445 This attribute can only be applied the method declarations within a class, and
446 not a protocol. Currently this attribute does not enforce any placement of
447 where the call occurs in the overriding method (such as in the case of
448 ``-dealloc`` where the call must appear at the end). It checks only that it
451 Note that on both OS X and iOS that the Foundation framework provides a
452 convenience macro ``NS_REQUIRES_SUPER`` that provides syntactic sugar for this
457 - (void)foo NS_REQUIRES_SUPER;
459 This macro is conditionally defined depending on the compiler's support for
460 this attribute. If the compiler does not support the attribute the macro
463 Operationally, when a method has this annotation the compiler will warn if the
464 implementation of an override in a subclass does not call super. For example:
468 warning: method possibly missing a [super AnnotMeth] call
469 - (void) AnnotMeth{};
474 def ObjCRuntimeNameDocs : Documentation {
475 let Category = DocCatFunction;
477 By default, the Objective-C interface or protocol identifier is used
478 in the metadata name for that object. The `objc_runtime_name`
479 attribute allows annotated interfaces or protocols to use the
480 specified string argument in the object's metadata name instead of the
483 **Usage**: ``__attribute__((objc_runtime_name("MyLocalName")))``. This attribute
484 can only be placed before an @protocol or @interface declaration:
488 __attribute__((objc_runtime_name("MyLocalName")))
495 def ObjCBoxableDocs : Documentation {
496 let Category = DocCatFunction;
498 Structs and unions marked with the ``objc_boxable`` attribute can be used
499 with the Objective-C boxed expression syntax, ``@(...)``.
501 **Usage**: ``__attribute__((objc_boxable))``. This attribute
502 can only be placed on a declaration of a trivially-copyable struct or union:
506 struct __attribute__((objc_boxable)) some_struct {
509 union __attribute__((objc_boxable)) some_union {
513 typedef struct __attribute__((objc_boxable)) _some_struct some_struct;
518 NSValue *boxed = @(ss);
523 def AvailabilityDocs : Documentation {
524 let Category = DocCatFunction;
526 The ``availability`` attribute can be placed on declarations to describe the
527 lifecycle of that declaration relative to operating system versions. Consider
528 the function declaration for a hypothetical function ``f``:
532 void f(void) __attribute__((availability(macosx,introduced=10.4,deprecated=10.6,obsoleted=10.7)));
534 The availability attribute states that ``f`` was introduced in Mac OS X 10.4,
535 deprecated in Mac OS X 10.6, and obsoleted in Mac OS X 10.7. This information
536 is used by Clang to determine when it is safe to use ``f``: for example, if
537 Clang is instructed to compile code for Mac OS X 10.5, a call to ``f()``
538 succeeds. If Clang is instructed to compile code for Mac OS X 10.6, the call
539 succeeds but Clang emits a warning specifying that the function is deprecated.
540 Finally, if Clang is instructed to compile code for Mac OS X 10.7, the call
541 fails because ``f()`` is no longer available.
543 The availability attribute is a comma-separated list starting with the
544 platform name and then including clauses specifying important milestones in the
545 declaration's lifetime (in any order) along with additional information. Those
548 introduced=\ *version*
549 The first version in which this declaration was introduced.
551 deprecated=\ *version*
552 The first version in which this declaration was deprecated, meaning that
553 users should migrate away from this API.
555 obsoleted=\ *version*
556 The first version in which this declaration was obsoleted, meaning that it
557 was removed completely and can no longer be used.
560 This declaration is never available on this platform.
562 message=\ *string-literal*
563 Additional message text that Clang will provide when emitting a warning or
564 error about use of a deprecated or obsoleted declaration. Useful to direct
565 users to replacement APIs.
567 Multiple availability attributes can be placed on a declaration, which may
568 correspond to different platforms. Only the availability attribute with the
569 platform corresponding to the target platform will be used; any others will be
570 ignored. If no availability attribute specifies availability for the current
571 target platform, the availability attributes are ignored. Supported platforms
575 Apple's iOS operating system. The minimum deployment target is specified by
576 the ``-mios-version-min=*version*`` or ``-miphoneos-version-min=*version*``
577 command-line arguments.
580 Apple's Mac OS X operating system. The minimum deployment target is
581 specified by the ``-mmacosx-version-min=*version*`` command-line argument.
583 A declaration can be used even when deploying back to a platform version prior
584 to when the declaration was introduced. When this happens, the declaration is
586 <https://developer.apple.com/library/mac/#documentation/MacOSX/Conceptual/BPFrameworks/Concepts/WeakLinking.html>`_,
587 as if the ``weak_import`` attribute were added to the declaration. A
588 weakly-linked declaration may or may not be present a run-time, and a program
589 can determine whether the declaration is present by checking whether the
590 address of that declaration is non-NULL.
592 If there are multiple declarations of the same entity, the availability
593 attributes must either match on a per-platform basis or later
594 declarations must not have availability attributes for that
595 platform. For example:
599 void g(void) __attribute__((availability(macosx,introduced=10.4)));
600 void g(void) __attribute__((availability(macosx,introduced=10.4))); // okay, matches
601 void g(void) __attribute__((availability(ios,introduced=4.0))); // okay, adds a new platform
602 void g(void); // okay, inherits both macosx and ios availability from above.
603 void g(void) __attribute__((availability(macosx,introduced=10.5))); // error: mismatch
605 When one method overrides another, the overriding method can be more widely available than the overridden method, e.g.,:
610 - (id)method __attribute__((availability(macosx,introduced=10.4)));
611 - (id)method2 __attribute__((availability(macosx,introduced=10.4)));
615 - (id)method __attribute__((availability(macosx,introduced=10.3))); // okay: method moved into base class later
616 - (id)method __attribute__((availability(macosx,introduced=10.5))); // error: this method was available via the base class in 10.4
621 def FallthroughDocs : Documentation {
622 let Category = DocCatStmt;
624 The ``clang::fallthrough`` attribute is used along with the
625 ``-Wimplicit-fallthrough`` argument to annotate intentional fall-through
626 between switch labels. It can only be applied to a null statement placed at a
627 point of execution between any statement and the next switch label. It is
628 common to mark these places with a specific comment, but this attribute is
629 meant to replace comments with a more strict annotation, which can be checked
630 by the compiler. This attribute doesn't change semantics of the code and can
631 be used wherever an intended fall-through occurs. It is designed to mimic
632 control-flow statements like ``break;``, so it can be placed in most places
633 where ``break;`` can, but only if there are no statements on the execution path
634 between it and the next switch label.
640 // compile with -Wimplicit-fallthrough
643 case 33: // no warning: no statements between case labels
645 case 44: // warning: unannotated fall-through
647 [[clang::fallthrough]];
648 case 55: // no warning
655 [[clang::fallthrough]];
657 case 66: // no warning
659 [[clang::fallthrough]]; // warning: fallthrough annotation does not
660 // directly precede case label
662 case 77: // warning: unannotated fall-through
668 def ARMInterruptDocs : Documentation {
669 let Category = DocCatFunction;
671 Clang supports the GNU style ``__attribute__((interrupt("TYPE")))`` attribute on
672 ARM targets. This attribute may be attached to a function definition and
673 instructs the backend to generate appropriate function entry/exit code so that
674 it can be used directly as an interrupt service routine.
676 The parameter passed to the interrupt attribute is optional, but if
677 provided it must be a string literal with one of the following values: "IRQ",
678 "FIQ", "SWI", "ABORT", "UNDEF".
680 The semantics are as follows:
682 - If the function is AAPCS, Clang instructs the backend to realign the stack to
683 8 bytes on entry. This is a general requirement of the AAPCS at public
684 interfaces, but may not hold when an exception is taken. Doing this allows
685 other AAPCS functions to be called.
686 - If the CPU is M-class this is all that needs to be done since the architecture
687 itself is designed in such a way that functions obeying the normal AAPCS ABI
688 constraints are valid exception handlers.
689 - If the CPU is not M-class, the prologue and epilogue are modified to save all
690 non-banked registers that are used, so that upon return the user-mode state
691 will not be corrupted. Note that to avoid unnecessary overhead, only
692 general-purpose (integer) registers are saved in this way. If VFP operations
693 are needed, that state must be saved manually.
695 Specifically, interrupt kinds other than "FIQ" will save all core registers
696 except "lr" and "sp". "FIQ" interrupts will save r0-r7.
697 - If the CPU is not M-class, the return instruction is changed to one of the
698 canonical sequences permitted by the architecture for exception return. Where
699 possible the function itself will make the necessary "lr" adjustments so that
700 the "preferred return address" is selected.
702 Unfortunately the compiler is unable to make this guarantee for an "UNDEF"
703 handler, where the offset from "lr" to the preferred return address depends on
704 the execution state of the code which generated the exception. In this case
705 a sequence equivalent to "movs pc, lr" will be used.
709 def TargetDocs : Documentation {
710 let Category = DocCatFunction;
712 Clang supports the GNU style ``__attribute__((target("OPTIONS")))`` attribute.
713 This attribute may be attached to a function definition and instructs
714 the backend to use different code generation options than were passed on the
717 The current set of options correspond to the existing "subtarget features" for
718 the target with or without a "-mno-" in front corresponding to the absence
719 of the feature, as well as ``arch="CPU"`` which will change the default "CPU"
722 Example "subtarget features" from the x86 backend include: "mmx", "sse", "sse4.2",
723 "avx", "xop" and largely correspond to the machine specific options handled by
728 def DocCatAMDGPURegisterAttributes :
729 DocumentationCategory<"AMD GPU Register Attributes"> {
731 Clang supports attributes for controlling register usage on AMD GPU
732 targets. These attributes may be attached to a kernel function
733 definition and is an optimization hint to the backend for the maximum
734 number of registers to use. This is useful in cases where register
735 limited occupancy is known to be an important factor for the
736 performance for the kernel.
738 The semantics are as follows:
740 - The backend will attempt to limit the number of used registers to
741 the specified value, but the exact number used is not
742 guaranteed. The number used may be rounded up to satisfy the
743 allocation requirements or ABI constraints of the subtarget. For
744 example, on Southern Islands VGPRs may only be allocated in
745 increments of 4, so requesting a limit of 39 VGPRs will really
746 attempt to use up to 40. Requesting more registers than the
747 subtarget supports will truncate to the maximum allowed. The backend
748 may also use fewer registers than requested whenever possible.
750 - 0 implies the default no limit on register usage.
752 - Ignored on older VLIW subtargets which did not have separate scalar
753 and vector registers, R600 through Northern Islands.
759 def AMDGPUNumVGPRDocs : Documentation {
760 let Category = DocCatAMDGPURegisterAttributes;
763 ``__attribute__((amdgpu_num_vgpr(<num_registers>)))`` attribute on AMD
764 Southern Islands GPUs and later for controlling the number of vector
765 registers. A typical value would be between 4 and 256 in increments
770 def AMDGPUNumSGPRDocs : Documentation {
771 let Category = DocCatAMDGPURegisterAttributes;
775 ``__attribute__((amdgpu_num_sgpr(<num_registers>)))`` attribute on AMD
776 Southern Islands GPUs and later for controlling the number of scalar
777 registers. A typical value would be between 8 and 104 in increments of
780 Due to common instruction constraints, an additional 2-4 SGPRs are
781 typically required for internal use depending on features used. This
782 value is a hint for the total number of SGPRs to use, and not the
783 number of user SGPRs, so no special consideration needs to be given
788 def DocCatCallingConvs : DocumentationCategory<"Calling Conventions"> {
790 Clang supports several different calling conventions, depending on the target
791 platform and architecture. The calling convention used for a function determines
792 how parameters are passed, how results are returned to the caller, and other
793 low-level details of calling a function.
797 def PcsDocs : Documentation {
798 let Category = DocCatCallingConvs;
800 On ARM targets, this attribute can be used to select calling conventions
801 similar to ``stdcall`` on x86. Valid parameter values are "aapcs" and
806 def RegparmDocs : Documentation {
807 let Category = DocCatCallingConvs;
809 On 32-bit x86 targets, the regparm attribute causes the compiler to pass
810 the first three integer parameters in EAX, EDX, and ECX instead of on the
811 stack. This attribute has no effect on variadic functions, and all parameters
812 are passed via the stack as normal.
816 def SysVABIDocs : Documentation {
817 let Category = DocCatCallingConvs;
819 On Windows x86_64 targets, this attribute changes the calling convention of a
820 function to match the default convention used on Sys V targets such as Linux,
821 Mac, and BSD. This attribute has no effect on other targets.
825 def MSABIDocs : Documentation {
826 let Category = DocCatCallingConvs;
828 On non-Windows x86_64 targets, this attribute changes the calling convention of
829 a function to match the default convention used on Windows x86_64. This
830 attribute has no effect on Windows targets or non-x86_64 targets.
834 def StdCallDocs : Documentation {
835 let Category = DocCatCallingConvs;
837 On 32-bit x86 targets, this attribute changes the calling convention of a
838 function to clear parameters off of the stack on return. This convention does
839 not support variadic calls or unprototyped functions in C, and has no effect on
840 x86_64 targets. This calling convention is used widely by the Windows API and
841 COM applications. See the documentation for `__stdcall`_ on MSDN.
843 .. _`__stdcall`: http://msdn.microsoft.com/en-us/library/zxk0tw93.aspx
847 def FastCallDocs : Documentation {
848 let Category = DocCatCallingConvs;
850 On 32-bit x86 targets, this attribute changes the calling convention of a
851 function to use ECX and EDX as register parameters and clear parameters off of
852 the stack on return. This convention does not support variadic calls or
853 unprototyped functions in C, and has no effect on x86_64 targets. This calling
854 convention is supported primarily for compatibility with existing code. Users
855 seeking register parameters should use the ``regparm`` attribute, which does
856 not require callee-cleanup. See the documentation for `__fastcall`_ on MSDN.
858 .. _`__fastcall`: http://msdn.microsoft.com/en-us/library/6xa169sk.aspx
862 def ThisCallDocs : Documentation {
863 let Category = DocCatCallingConvs;
865 On 32-bit x86 targets, this attribute changes the calling convention of a
866 function to use ECX for the first parameter (typically the implicit ``this``
867 parameter of C++ methods) and clear parameters off of the stack on return. This
868 convention does not support variadic calls or unprototyped functions in C, and
869 has no effect on x86_64 targets. See the documentation for `__thiscall`_ on
872 .. _`__thiscall`: http://msdn.microsoft.com/en-us/library/ek8tkfbw.aspx
876 def VectorCallDocs : Documentation {
877 let Category = DocCatCallingConvs;
879 On 32-bit x86 *and* x86_64 targets, this attribute changes the calling
880 convention of a function to pass vector parameters in SSE registers.
882 On 32-bit x86 targets, this calling convention is similar to ``__fastcall``.
883 The first two integer parameters are passed in ECX and EDX. Subsequent integer
884 parameters are passed in memory, and callee clears the stack. On x86_64
885 targets, the callee does *not* clear the stack, and integer parameters are
886 passed in RCX, RDX, R8, and R9 as is done for the default Windows x64 calling
889 On both 32-bit x86 and x86_64 targets, vector and floating point arguments are
890 passed in XMM0-XMM5. Homogenous vector aggregates of up to four elements are
891 passed in sequential SSE registers if enough are available. If AVX is enabled,
892 256 bit vectors are passed in YMM0-YMM5. Any vector or aggregate type that
893 cannot be passed in registers for any reason is passed by reference, which
894 allows the caller to align the parameter memory.
896 See the documentation for `__vectorcall`_ on MSDN for more details.
898 .. _`__vectorcall`: http://msdn.microsoft.com/en-us/library/dn375768.aspx
902 def DocCatConsumed : DocumentationCategory<"Consumed Annotation Checking"> {
904 Clang supports additional attributes for checking basic resource management
905 properties, specifically for unique objects that have a single owning reference.
906 The following attributes are currently supported, although **the implementation
907 for these annotations is currently in development and are subject to change.**
911 def SetTypestateDocs : Documentation {
912 let Category = DocCatConsumed;
914 Annotate methods that transition an object into a new state with
915 ``__attribute__((set_typestate(new_state)))``. The new state must be
916 unconsumed, consumed, or unknown.
920 def CallableWhenDocs : Documentation {
921 let Category = DocCatConsumed;
923 Use ``__attribute__((callable_when(...)))`` to indicate what states a method
924 may be called in. Valid states are unconsumed, consumed, or unknown. Each
925 argument to this attribute must be a quoted string. E.g.:
927 ``__attribute__((callable_when("unconsumed", "unknown")))``
931 def TestTypestateDocs : Documentation {
932 let Category = DocCatConsumed;
934 Use ``__attribute__((test_typestate(tested_state)))`` to indicate that a method
935 returns true if the object is in the specified state..
939 def ParamTypestateDocs : Documentation {
940 let Category = DocCatConsumed;
942 This attribute specifies expectations about function parameters. Calls to an
943 function with annotated parameters will issue a warning if the corresponding
944 argument isn't in the expected state. The attribute is also used to set the
945 initial state of the parameter when analyzing the function's body.
949 def ReturnTypestateDocs : Documentation {
950 let Category = DocCatConsumed;
952 The ``return_typestate`` attribute can be applied to functions or parameters.
953 When applied to a function the attribute specifies the state of the returned
954 value. The function's body is checked to ensure that it always returns a value
955 in the specified state. On the caller side, values returned by the annotated
956 function are initialized to the given state.
958 When applied to a function parameter it modifies the state of an argument after
959 a call to the function returns. The function's body is checked to ensure that
960 the parameter is in the expected state before returning.
964 def ConsumableDocs : Documentation {
965 let Category = DocCatConsumed;
967 Each ``class`` that uses any of the typestate annotations must first be marked
968 using the ``consumable`` attribute. Failure to do so will result in a warning.
970 This attribute accepts a single parameter that must be one of the following:
971 ``unknown``, ``consumed``, or ``unconsumed``.
975 def NoSanitizeDocs : Documentation {
976 let Category = DocCatFunction;
978 Use the ``no_sanitize`` attribute on a function declaration to specify
979 that a particular instrumentation or set of instrumentations should not be
980 applied to that function. The attribute takes a list of string literals,
981 which have the same meaning as values accepted by the ``-fno-sanitize=``
982 flag. For example, ``__attribute__((no_sanitize("address", "thread")))``
983 specifies that AddressSanitizer and ThreadSanitizer should not be applied
986 See :ref:`Controlling Code Generation <controlling-code-generation>` for a
987 full list of supported sanitizer flags.
991 def NoSanitizeAddressDocs : Documentation {
992 let Category = DocCatFunction;
993 // This function has multiple distinct spellings, and so it requires a custom
994 // heading to be specified. The most common spelling is sufficient.
995 let Heading = "no_sanitize_address (no_address_safety_analysis, gnu::no_address_safety_analysis, gnu::no_sanitize_address)";
997 .. _langext-address_sanitizer:
999 Use ``__attribute__((no_sanitize_address))`` on a function declaration to
1000 specify that address safety instrumentation (e.g. AddressSanitizer) should
1001 not be applied to that function.
1005 def NoSanitizeThreadDocs : Documentation {
1006 let Category = DocCatFunction;
1007 let Heading = "no_sanitize_thread";
1009 .. _langext-thread_sanitizer:
1011 Use ``__attribute__((no_sanitize_thread))`` on a function declaration to
1012 specify that checks for data races on plain (non-atomic) memory accesses should
1013 not be inserted by ThreadSanitizer. The function is still instrumented by the
1014 tool to avoid false positives and provide meaningful stack traces.
1018 def NoSanitizeMemoryDocs : Documentation {
1019 let Category = DocCatFunction;
1020 let Heading = "no_sanitize_memory";
1022 .. _langext-memory_sanitizer:
1024 Use ``__attribute__((no_sanitize_memory))`` on a function declaration to
1025 specify that checks for uninitialized memory should not be inserted
1026 (e.g. by MemorySanitizer). The function may still be instrumented by the tool
1027 to avoid false positives in other places.
1031 def DocCatTypeSafety : DocumentationCategory<"Type Safety Checking"> {
1033 Clang supports additional attributes to enable checking type safety properties
1034 that can't be enforced by the C type system. Use cases include:
1036 * MPI library implementations, where these attributes enable checking that
1037 the buffer type matches the passed ``MPI_Datatype``;
1038 * for HDF5 library there is a similar use case to MPI;
1039 * checking types of variadic functions' arguments for functions like
1040 ``fcntl()`` and ``ioctl()``.
1042 You can detect support for these attributes with ``__has_attribute()``. For
1047 #if defined(__has_attribute)
1048 # if __has_attribute(argument_with_type_tag) && \
1049 __has_attribute(pointer_with_type_tag) && \
1050 __has_attribute(type_tag_for_datatype)
1051 # define ATTR_MPI_PWT(buffer_idx, type_idx) __attribute__((pointer_with_type_tag(mpi,buffer_idx,type_idx)))
1052 /* ... other macros ... */
1056 #if !defined(ATTR_MPI_PWT)
1057 # define ATTR_MPI_PWT(buffer_idx, type_idx)
1060 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1065 def ArgumentWithTypeTagDocs : Documentation {
1066 let Category = DocCatTypeSafety;
1067 let Heading = "argument_with_type_tag";
1069 Use ``__attribute__((argument_with_type_tag(arg_kind, arg_idx,
1070 type_tag_idx)))`` on a function declaration to specify that the function
1071 accepts a type tag that determines the type of some other argument.
1072 ``arg_kind`` is an identifier that should be used when annotating all
1073 applicable type tags.
1075 This attribute is primarily useful for checking arguments of variadic functions
1076 (``pointer_with_type_tag`` can be used in most non-variadic cases).
1082 int fcntl(int fd, int cmd, ...)
1083 __attribute__(( argument_with_type_tag(fcntl,3,2) ));
1087 def PointerWithTypeTagDocs : Documentation {
1088 let Category = DocCatTypeSafety;
1089 let Heading = "pointer_with_type_tag";
1091 Use ``__attribute__((pointer_with_type_tag(ptr_kind, ptr_idx, type_tag_idx)))``
1092 on a function declaration to specify that the function accepts a type tag that
1093 determines the pointee type of some other pointer argument.
1099 int MPI_Send(void *buf, int count, MPI_Datatype datatype /*, other args omitted */)
1100 __attribute__(( pointer_with_type_tag(mpi,1,3) ));
1104 def TypeTagForDatatypeDocs : Documentation {
1105 let Category = DocCatTypeSafety;
1107 Clang supports annotating type tags of two forms.
1109 * **Type tag that is an expression containing a reference to some declared
1110 identifier.** Use ``__attribute__((type_tag_for_datatype(kind, type)))`` on a
1111 declaration with that identifier:
1115 extern struct mpi_datatype mpi_datatype_int
1116 __attribute__(( type_tag_for_datatype(mpi,int) ));
1117 #define MPI_INT ((MPI_Datatype) &mpi_datatype_int)
1119 * **Type tag that is an integral literal.** Introduce a ``static const``
1120 variable with a corresponding initializer value and attach
1121 ``__attribute__((type_tag_for_datatype(kind, type)))`` on that declaration,
1126 #define MPI_INT ((MPI_Datatype) 42)
1127 static const MPI_Datatype mpi_datatype_int
1128 __attribute__(( type_tag_for_datatype(mpi,int) )) = 42
1130 The attribute also accepts an optional third argument that determines how the
1131 expression is compared to the type tag. There are two supported flags:
1133 * ``layout_compatible`` will cause types to be compared according to
1134 layout-compatibility rules (C++11 [class.mem] p 17, 18). This is
1135 implemented to support annotating types like ``MPI_DOUBLE_INT``.
1142 struct internal_mpi_double_int { double d; int i; };
1143 extern struct mpi_datatype mpi_datatype_double_int
1144 __attribute__(( type_tag_for_datatype(mpi, struct internal_mpi_double_int, layout_compatible) ));
1146 #define MPI_DOUBLE_INT ((MPI_Datatype) &mpi_datatype_double_int)
1149 struct my_pair { double a; int b; };
1150 struct my_pair *buffer;
1151 MPI_Send(buffer, 1, MPI_DOUBLE_INT /*, ... */); // no warning
1153 struct my_int_pair { int a; int b; }
1154 struct my_int_pair *buffer2;
1155 MPI_Send(buffer2, 1, MPI_DOUBLE_INT /*, ... */); // warning: actual buffer element
1156 // type 'struct my_int_pair'
1157 // doesn't match specified MPI_Datatype
1159 * ``must_be_null`` specifies that the expression should be a null pointer
1160 constant, for example:
1165 extern struct mpi_datatype mpi_datatype_null
1166 __attribute__(( type_tag_for_datatype(mpi, void, must_be_null) ));
1168 #define MPI_DATATYPE_NULL ((MPI_Datatype) &mpi_datatype_null)
1171 MPI_Send(buffer, 1, MPI_DATATYPE_NULL /*, ... */); // warning: MPI_DATATYPE_NULL
1172 // was specified but buffer
1173 // is not a null pointer
1177 def FlattenDocs : Documentation {
1178 let Category = DocCatFunction;
1180 The ``flatten`` attribute causes calls within the attributed function to
1181 be inlined unless it is impossible to do so, for example if the body of the
1182 callee is unavailable or if the callee has the ``noinline`` attribute.
1186 def FormatDocs : Documentation {
1187 let Category = DocCatFunction;
1190 Clang supports the ``format`` attribute, which indicates that the function
1191 accepts a ``printf`` or ``scanf``-like format string and corresponding
1192 arguments or a ``va_list`` that contains these arguments.
1194 Please see `GCC documentation about format attribute
1195 <http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html>`_ to find details
1196 about attribute syntax.
1198 Clang implements two kinds of checks with this attribute.
1200 #. Clang checks that the function with the ``format`` attribute is called with
1201 a format string that uses format specifiers that are allowed, and that
1202 arguments match the format string. This is the ``-Wformat`` warning, it is
1205 #. Clang checks that the format string argument is a literal string. This is
1206 the ``-Wformat-nonliteral`` warning, it is off by default.
1208 Clang implements this mostly the same way as GCC, but there is a difference
1209 for functions that accept a ``va_list`` argument (for example, ``vprintf``).
1210 GCC does not emit ``-Wformat-nonliteral`` warning for calls to such
1211 functions. Clang does not warn if the format string comes from a function
1212 parameter, where the function is annotated with a compatible attribute,
1213 otherwise it warns. For example:
1217 __attribute__((__format__ (__scanf__, 1, 3)))
1218 void foo(const char* s, char *buf, ...) {
1222 vprintf(s, ap); // warning: format string is not a string literal
1225 In this case we warn because ``s`` contains a format string for a
1226 ``scanf``-like function, but it is passed to a ``printf``-like function.
1228 If the attribute is removed, clang still warns, because the format string is
1229 not a string literal.
1235 __attribute__((__format__ (__printf__, 1, 3)))
1236 void foo(const char* s, char *buf, ...) {
1240 vprintf(s, ap); // warning
1243 In this case Clang does not warn because the format string ``s`` and
1244 the corresponding arguments are annotated. If the arguments are
1245 incorrect, the caller of ``foo`` will receive a warning.
1249 def AlignValueDocs : Documentation {
1250 let Category = DocCatType;
1252 The align_value attribute can be added to the typedef of a pointer type or the
1253 declaration of a variable of pointer or reference type. It specifies that the
1254 pointer will point to, or the reference will bind to, only objects with at
1255 least the provided alignment. This alignment value must be some positive power
1260 typedef double * aligned_double_ptr __attribute__((align_value(64)));
1261 void foo(double & x __attribute__((align_value(128)),
1262 aligned_double_ptr y) { ... }
1264 If the pointer value does not have the specified alignment at runtime, the
1265 behavior of the program is undefined.
1269 def FlagEnumDocs : Documentation {
1270 let Category = DocCatType;
1272 This attribute can be added to an enumerator to signal to the compiler that it
1273 is intended to be used as a flag type. This will cause the compiler to assume
1274 that the range of the type includes all of the values that you can get by
1275 manipulating bits of the enumerator when issuing warnings.
1279 def MSInheritanceDocs : Documentation {
1280 let Category = DocCatType;
1281 let Heading = "__single_inhertiance, __multiple_inheritance, __virtual_inheritance";
1283 This collection of keywords is enabled under ``-fms-extensions`` and controls
1284 the pointer-to-member representation used on ``*-*-win32`` targets.
1286 The ``*-*-win32`` targets utilize a pointer-to-member representation which
1287 varies in size and alignment depending on the definition of the underlying
1290 However, this is problematic when a forward declaration is only available and
1291 no definition has been made yet. In such cases, Clang is forced to utilize the
1292 most general representation that is available to it.
1294 These keywords make it possible to use a pointer-to-member representation other
1295 than the most general one regardless of whether or not the definition will ever
1296 be present in the current translation unit.
1298 This family of keywords belong between the ``class-key`` and ``class-name``:
1302 struct __single_inheritance S;
1306 This keyword can be applied to class templates but only has an effect when used
1307 on full specializations:
1311 template <typename T, typename U> struct __single_inheritance A; // warning: inheritance model ignored on primary template
1312 template <typename T> struct __multiple_inheritance A<T, T>; // warning: inheritance model ignored on partial specialization
1313 template <> struct __single_inheritance A<int, float>;
1315 Note that choosing an inheritance model less general than strictly necessary is
1320 struct __multiple_inheritance S; // error: inheritance model does not match definition
1326 def MSNoVTableDocs : Documentation {
1327 let Category = DocCatType;
1329 This attribute can be added to a class declaration or definition to signal to
1330 the compiler that constructors and destructors will not reference the virtual
1335 def OptnoneDocs : Documentation {
1336 let Category = DocCatFunction;
1338 The ``optnone`` attribute suppresses essentially all optimizations
1339 on a function or method, regardless of the optimization level applied to
1340 the compilation unit as a whole. This is particularly useful when you
1341 need to debug a particular function, but it is infeasible to build the
1342 entire application without optimization. Avoiding optimization on the
1343 specified function can improve the quality of the debugging information
1346 This attribute is incompatible with the ``always_inline`` and ``minsize``
1351 def LoopHintDocs : Documentation {
1352 let Category = DocCatStmt;
1353 let Heading = "#pragma clang loop";
1355 The ``#pragma clang loop`` directive allows loop optimization hints to be
1356 specified for the subsequent loop. The directive allows vectorization,
1357 interleaving, and unrolling to be enabled or disabled. Vector width as well
1358 as interleave and unrolling count can be manually specified. See
1359 `language extensions
1360 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
1365 def UnrollHintDocs : Documentation {
1366 let Category = DocCatStmt;
1367 let Heading = "#pragma unroll, #pragma nounroll";
1369 Loop unrolling optimization hints can be specified with ``#pragma unroll`` and
1370 ``#pragma nounroll``. The pragma is placed immediately before a for, while,
1371 do-while, or c++11 range-based for loop.
1373 Specifying ``#pragma unroll`` without a parameter directs the loop unroller to
1374 attempt to fully unroll the loop if the trip count is known at compile time:
1383 Specifying the optional parameter, ``#pragma unroll _value_``, directs the
1384 unroller to unroll the loop ``_value_`` times. The parameter may optionally be
1385 enclosed in parentheses:
1399 Specifying ``#pragma nounroll`` indicates that the loop should not be unrolled:
1408 ``#pragma unroll`` and ``#pragma unroll _value_`` have identical semantics to
1409 ``#pragma clang loop unroll(full)`` and
1410 ``#pragma clang loop unroll_count(_value_)`` respectively. ``#pragma nounroll``
1411 is equivalent to ``#pragma clang loop unroll(disable)``. See
1412 `language extensions
1413 <http://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations>`_
1414 for further details including limitations of the unroll hints.
1418 def DocOpenCLAddressSpaces : DocumentationCategory<"OpenCL Address Spaces"> {
1420 The address space qualifier may be used to specify the region of memory that is
1421 used to allocate the object. OpenCL supports the following address spaces:
1422 __generic(generic), __global(global), __local(local), __private(private),
1423 __constant(constant).
1427 __constant int c = ...;
1429 __generic int* foo(global int* g) {
1436 More details can be found in the OpenCL C language Spec v2.0, Section 6.5.
1440 def OpenCLAddressSpaceGenericDocs : Documentation {
1441 let Category = DocOpenCLAddressSpaces;
1442 let Heading = "__generic(generic)";
1444 The generic address space attribute is only available with OpenCL v2.0 and later.
1445 It can be used with pointer types. Variables in global and local scope and
1446 function parameters in non-kernel functions can have the generic address space
1447 type attribute. It is intended to be a placeholder for any other address space
1448 except for '__constant' in OpenCL code which can be used with multiple address
1453 def OpenCLAddressSpaceConstantDocs : Documentation {
1454 let Category = DocOpenCLAddressSpaces;
1455 let Heading = "__constant(constant)";
1457 The constant address space attribute signals that an object is located in
1458 a constant (non-modifiable) memory region. It is available to all work items.
1459 Any type can be annotated with the constant address space attribute. Objects
1460 with the constant address space qualifier can be declared in any scope and must
1461 have an initializer.
1465 def OpenCLAddressSpaceGlobalDocs : Documentation {
1466 let Category = DocOpenCLAddressSpaces;
1467 let Heading = "__global(global)";
1469 The global address space attribute specifies that an object is allocated in
1470 global memory, which is accessible by all work items. The content stored in this
1471 memory area persists between kernel executions. Pointer types to the global
1472 address space are allowed as function parameters or local variables. Starting
1473 with OpenCL v2.0, the global address space can be used with global (program
1474 scope) variables and static local variable as well.
1478 def OpenCLAddressSpaceLocalDocs : Documentation {
1479 let Category = DocOpenCLAddressSpaces;
1480 let Heading = "__local(local)";
1482 The local address space specifies that an object is allocated in the local (work
1483 group) memory area, which is accessible to all work items in the same work
1484 group. The content stored in this memory region is not accessible after
1485 the kernel execution ends. In a kernel function scope, any variable can be in
1486 the local address space. In other scopes, only pointer types to the local address
1487 space are allowed. Local address space variables cannot have an initializer.
1491 def OpenCLAddressSpacePrivateDocs : Documentation {
1492 let Category = DocOpenCLAddressSpaces;
1493 let Heading = "__private(private)";
1495 The private address space specifies that an object is allocated in the private
1496 (work item) memory. Other work items cannot access the same memory area and its
1497 content is destroyed after work item execution ends. Local variables can be
1498 declared in the private address space. Function arguments are always in the
1499 private address space. Kernel function arguments of a pointer or an array type
1500 cannot point to the private address space.
1504 def NullabilityDocs : DocumentationCategory<"Nullability Attributes"> {
1506 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``).
1508 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:
1512 // No meaningful result when 'ptr' is null (here, it happens to be undefined behavior).
1513 int fetch(int * _Nonnull ptr) { return *ptr; }
1515 // 'ptr' may be null.
1516 int fetch_or_zero(int * _Nullable ptr) {
1517 return ptr ? *ptr : 0;
1520 // A nullable pointer to non-null pointers to const characters.
1521 const char *join_strings(const char * _Nonnull * _Nullable strings, unsigned n);
1523 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:
1525 .. code-block:: objective-c
1527 @interface NSView : NSResponder
1528 - (nullable NSView *)ancestorSharedWithView:(nonnull NSView *)aView;
1529 @property (assign, nullable) NSView *superview;
1530 @property (readonly, nonnull) NSArray *subviews;
1535 def TypeNonNullDocs : Documentation {
1536 let Category = NullabilityDocs;
1537 let Heading = "_Nonnull";
1539 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:
1543 int fetch(int * _Nonnull ptr);
1545 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.
1549 def TypeNullableDocs : Documentation {
1550 let Category = NullabilityDocs;
1551 let Heading = "_Nullable";
1553 The ``_Nullable`` nullability qualifier indicates that a value of the ``_Nullable`` pointer type can be null. For example, given:
1557 int fetch_or_zero(int * _Nullable ptr);
1559 a caller of ``fetch_or_zero`` can provide null.
1563 def TypeNullUnspecifiedDocs : Documentation {
1564 let Category = NullabilityDocs;
1565 let Heading = "_Null_unspecified";
1567 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.
1571 def NonNullDocs : Documentation {
1572 let Category = NullabilityDocs;
1573 let Heading = "nonnull";
1575 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:
1579 extern void * my_memcpy (void *dest, const void *src, size_t len)
1580 __attribute__((nonnull (1, 2)));
1582 Here, the ``nonnull`` attribute indicates that parameters 1 and 2
1583 cannot have a null value. Omitting the parenthesized list of parameter indices means that all parameters of pointer type cannot be null:
1587 extern void * my_memcpy (void *dest, const void *src, size_t len)
1588 __attribute__((nonnull));
1590 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:
1594 extern void * my_memcpy (void *dest __attribute__((nonnull)),
1595 const void *src __attribute__((nonnull)), size_t len);
1597 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.
1601 def ReturnsNonNullDocs : Documentation {
1602 let Category = NullabilityDocs;
1603 let Heading = "returns_nonnull";
1605 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:
1609 extern void * malloc (size_t size) __attribute__((returns_nonnull));
1611 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