Clang 16.0.0 Release Notes

Written by the LLVM Team

Introduction

This document contains the release notes for the Clang C/C++/Objective-C frontend, part of the LLVM Compiler Infrastructure, release 16.0.0. Here we describe the status of Clang in some detail, including major improvements from the previous release and new feature work. For the general LLVM release notes, see the LLVM documentation. For the libc++ release notes, see this page. All LLVM releases may be downloaded from the LLVM releases web site.

For more information about Clang or LLVM, including information about the latest release, please see the Clang Web Site or the LLVM Web Site.

Potentially Breaking Changes

These changes are ones which we think may surprise users when upgrading to Clang 16.0.0 because of the opportunity they pose for disruption to existing code bases.

• Clang’s default C++/ObjC++ standard is now gnu++17 instead of gnu++14. This means Clang will by default accept code using features from C++17 and conforming GNU extensions. Projects incompatible with C++17 can add -std=gnu++14 to their build settings to restore the previous behaviour.

• The -fexperimental-new-pass-manager and -fno-legacy-pass-manager flags have been removed. These flags have been silently ignored since Clang 15.

• Clang’s resource directory path previously included the full clang version. It now includes only the major version. The new resource directory is $prefix/lib/clang/$CLANG_MAJOR_VERSION and can be queried using clang -print-resource-dir, just like before.

• The -Wimplicit-function-declaration and -Wimplicit-int warnings now default to an error in C99, C11, and C17. As of C2x, support for implicit function declarations and implicit int has been removed, and the warning options will have no effect. Specifying -Wimplicit-int in C89 mode will now issue warnings instead of being a noop.

  NOTE: We recommend that projects using configure scripts verify that the results do not change before/after setting -Werror=implicit-function-declarations or -Wimplicit-int to avoid incompatibility with Clang 16.

• -Wincompatible-function-pointer-types now defaults to an error in all C language modes. It may be downgraded to a warning with -Wno-error=incompatible-function-pointer-types or disabled entirely with -Wno-incompatible-function-pointer-types.

  NOTE: We recommend that projects using configure scripts verify that the results do not change before/after setting -Werror=incompatible-function-pointer-types to avoid incompatibility with Clang 16.

  void func(const int *i);
  void other(void) {
    void (*fp)(int *) = func; // Previously a warning, now a downgradable error.
  }
  

• To match GCC, __ppc64__ is no longer defined on PowerPC64 targets. Use __powerpc64__ instead.

• -p is rejected for all targets which are not AIX or OpenBSD. -p led to an -Wunused-command-line-argument warning in previous releases.

C/C++ Language Potentially Breaking Changes

• Clang now disallows types whose sizes aren’t a multiple of their alignments to be used as the element type of arrays.

  typedef char int8_a16 __attribute__((aligned(16)));
  int8_a16 array[4]; // Now diagnosed as the element size not being a multiple of the array alignment.
  

• When compiling for Windows in MSVC compatibility mode (for example by using clang-cl), the compiler will now propagate dllimport/export declspecs in explicit specializations of class template member functions (#54717):

  template <typename> struct __declspec(dllexport) S {
    void f();
  };
  template<> void S<int>::f() {}  // clang-cl will now dllexport this.
  

  This matches what MSVC does, so it improves compatibility, but it can also cause errors for code which clang-cl would previously accept, for example:

  template <typename> struct __declspec(dllexport) S {
    void f();
  };
  template<> void S<int>::f() = delete;  // Error: cannot delete dllexport function.
  

  template <typename> struct __declspec(dllimport) S {
    void f();
  };
  template<> void S<int>::f() {};  // Error: cannot define dllimport function.
  

  These errors also match MSVC’s behavior.

• Clang now diagnoses indirection of void * in C++ mode as a warning which defaults to an error. This is compatible with ISO C++, GCC, ICC, and MSVC. This is also now a SFINAE error so constraint checking and SFINAE checking can be compatible with other compilers. It is expected that this will be upgraded to an error-only diagnostic in the next Clang release.

  void func(void *p) {
    *p; // Now diagnosed as a warning-as-error.
  }
  

• Clang now diagnoses use of a bit-field as an instruction operand in Microsoft style inline asm blocks as an error. Previously, a bit-field operand yielded the address of the allocation unit the bit-field was stored within; reads or writes therefore had the potential to read or write nearby bit-fields. (#57791)

  typedef struct S {
    unsigned bf:1;
  } S;
  void f(S s) {
    __asm {
      mov eax, s.bf // Now diagnosed as an error.
      mov s.bf, eax // Now diagnosed as an error.
    }
  }
  

• Clang now diagnoses if structs/unions with the same name are different in different used modules. Behavior in C and Objective-C language modes now is the same as in C++.

C++ Specific Potentially Breaking Changes

• Clang now prohibits non-inline externally-visible definitions in C++ standard header units as per [module.import/6]. In Clang 15, these definitions were allowed. Note that such definitions are ODR violations if the header is included more than once.

• Clang will now correctly diagnose as ill-formed a constant expression where an enum without a fixed underlying type is set to a value outside the range of the enumeration’s values.

  enum E { Zero, One, Two, Three, Four };
  constexpr E Val1 = (E)3;  // Ok
  constexpr E Val2 = (E)7;  // Ok
  constexpr E Val3 = (E)8;  // Now diagnosed as out of the range [0, 7]
  constexpr E Val4 = (E)-1; // Now diagnosed as out of the range [0, 7]
  

  Due to the extended period of time this bug was present in major C++ implementations (including Clang), this error has the ability to be downgraded into a warning (via: -Wno-error=enum-constexpr-conversion) to provide a transition period for users. This diagnostic is expected to turn into an error-only diagnostic in the next Clang release. (#50055)

• As a side effect of implementing DR692/DR1395/DR1432, Clang now rejects some overloaded function templates as ambiguous when one of the candidates has a trailing parameter pack.

  template <typename T> void g(T, T = T());
  template <typename T, typename... U> void g(T, U...);
  void h() {
    // This is rejected due to ambiguity between the pack and the
    // default argument. Only parameters with arguments are considered during
    // partial ordering of function templates.
    g(42);
  }
  

ABI Changes in This Version

• GCC doesn’t pack non-POD members in packed structs unless the packed attribute is also specified on the member. Clang historically did perform such packing. Clang now matches the gcc behavior (except on Darwin, PS4 and AIX). You can switch back to the old ABI behavior with the flag: -fclang-abi-compat=15.0.

• GCC allows POD types to have defaulted special members. Clang historically classified such types as non-POD (for the purposes of Itanium ABI). Clang now matches the gcc behavior (except on Darwin, PS4, AIX and z/OS). You can switch back to the old ABI behavior with the flag: -fclang-abi-compat=15.0.

What’s New in Clang 16.0.0?

Some of the major new features and improvements to Clang are listed here. Generic improvements to Clang as a whole or to its underlying infrastructure are described first, followed by language-specific sections with improvements to Clang’s support for those languages.

C++ Language Changes

C++20 Feature Support

Newly implemented papers:

• Implemented P0848: Conditionally Trivial Special Member Functions. Note: The handling of deleted functions is not yet compliant, as Clang does not implement DR1496 and DR1734.

• Implemented P1091R3 and P1381R1 to support capturing structured bindings in lambdas (#52720, #54300, #54301, #49430).

• Implemented P2468R2: The Equality Operator You Are Looking For.

• Implemented P2113R0: Proposed resolution for 2019 comment CA 112 ([temp.func.order]p6.2.1 is not implemented, matching GCC).

• Implemented P0634R3: Down with typename!, which removes the requirement for the typename keyword in certain contexts.

• Implemented P0857R0: Fixing small-ish functionality gaps in constraints, which specifies constrained lambdas and constrained template template-parameters.

• Implemented P0960R3 and P1975R0, which allow parenthesized aggregate-initialization.

Notable Bug Fixes to C++20 Features:

• Clang now correctly delays the instantiation of function constraints until the time of checking, which should now allow the libstdc++ ranges implementation to work for at least trivial examples. (#44178)

• Consider explicitly defaulted constexpr/consteval special member function template instantiation to be constexpr/consteval even though a call to such a function cannot appear in a constant expression. (C++14 [dcl.constexpr]p6 (CWG DR647/CWG DR1358))

• Correctly defer dependent immediate function invocations until template instantiation. (#55601)

• Correctly set expression evaluation context as ‘immediate function context’ in consteval functions. (#51182)

• Skip rebuilding lambda expressions in arguments of immediate invocations. (#56183, #51695, #50455, #54872, #54587)

• Clang implements DR2621, correcting a defect in using enum handling. The name is found via ordinary lookup so typedefs are found.

• Implemented CWG2635 as a Defect Report, which prohibits structured bindings from being constrained.

• When using header modules, inclusion of a private header and violations of the use-declaration rules are now diagnosed even when the includer is a textual header. This change can be temporarily reversed with -Xclang -fno-modules-validate-textual-header-includes, but this flag will be removed in a future Clang release.

C++2b Feature Support

• Implemented P2324R2: Support label at end of compound statement.

• Implemented P1169R4: static operator() and P2589R1: static operator[].

• Implemented P2513R3: char8_t Compatibility and Portability Fix. This change was applied to C++20 as a Defect Report.

• Implemented P2647R1: Permitting static constexpr variables in constexpr functions.

• Implemented CWG2640: Allow more characters in an n-char sequence.

Resolutions to C++ Defect Reports

• Implemented DR692, DR1395, and DR1432. The fix for DR1432 is speculative since the issue is still open and has no proposed resolution at this time. A speculative fix for DR1432 is needed to prevent regressions that would otherwise occur due to DR692.

• Implemented DR2358, which allows init captures in lambdas in default arguments.

• Implemented DR2654 which undeprecates all compound assignments operations on volatile qualified variables.

• Implemented DR2631. Invalid consteval calls in default arguments and default member initializers are diagnosed when and if the default is used. This changes the value of std::source_location::current() used in default parameters calls compared to previous versions of Clang. (#56379)

C Language Changes

• Adjusted -Wformat warnings according to WG14 N2562. Clang will now consider default argument promotions in printf, and remove unnecessary warnings. Especially int argument with specifier %hhd and %hd.

C2x Feature Support

• Implemented WG14 N2662, so the [[maybe_unused]] attribute may be applied to a label to silence an -Wunused-label warning.

• Implemented WG14 N2508, so labels can placed everywhere inside a compound statement.

• Implemented WG14 N2927, the Not-so-magic typeof operator. Also implemented WG14 N2930, renaming remove_quals, so the typeof_unqual operator is also supported. Both of these operators are supported only in C2x mode. The typeof operator specifies the type of the given parenthesized expression operand or type name, including all qualifiers. The typeof_unqual operator is similar to typeof except that all qualifiers are removed, including atomic type qualification and type attributes which behave like a qualifier, such as an address space attribute.

  __attribute__((address_space(1))) const _Atomic int Val;
  typeof(Val) OtherVal; // type is '__attribute__((address_space(1))) const _Atomic int'
  typeof_unqual(Val) OtherValUnqual; // type is 'int'
  

• Implemented WG14 N3042, Introduce the nullptr constant. This introduces a new type nullptr_t, declared in <stddef.h> which represents the type of the null pointer named constant, nullptr. This constant is implicitly convertible to any pointer type and represents a type-safe null value.

  Note, there are some known incompatibilities with this same feature in C++. The following examples were discovered during implementation and are subject to change depending on how national body comments are resolved by WG14 (C status is based on standard requirements, not necessarily implementation behavior):

  nullptr_t null_val;
  (nullptr_t)nullptr;       // Rejected in C, accepted in C++, Clang accepts
  (void)(1 ? nullptr : 0);  // Rejected in C, accepted in C++, Clang rejects
  (void)(1 ? null_val : 0); // Rejected in C, accepted in C++, Clang rejects
  bool b1 = nullptr;        // Accepted in C, rejected in C++, Clang rejects
  b1 = null_val;            // Accepted in C, rejected in C++, Clang rejects
  null_val = 0;             // Rejected in C, accepted in C++, Clang rejects
  
  void func(nullptr_t);
  func(0);                  // Rejected in C, accepted in C++, Clang rejects
  

• Implemented WG14 N2975, Relax requirements for va_start. In C2x mode, functions can now be declared fully variadic and the va_start macro no longer requires passing a second argument (though it accepts a second argument for backwards compatibility). If a second argument is passed, it is neither expanded nor evaluated in C2x mode.

  void func(...) {  // Invalid in C17 and earlier, valid in C2x and later.
    va_list list;
    va_start(list); // Invalid in C17 and earlier, valid in C2x and later.
    va_end(list);
  }
  

• Diagnose type definitions in the type argument of __builtin_offsetof as a conforming C extension according to WG14 N2350. Also documents the builtin appropriately. Note, a type definition in C++ continues to be rejected.

OpenCL Kernel Language Changes

• Improved diagnostics for C++ templates used in kernel arguments.

• Removed redundant pragma for the arm-integer-dot-product extension.

• Improved support of enqueued block for AMDGPU.

• Added nounwind attribute implicitly to all functions.

• Improved builtin functions support:

  • Allow disabling default header-based feature/extension support by passing -D__undef_<feature>.

  • Fixed conditional definition of the depth image and read_write image3d builtins.

Non-comprehensive list of changes in this release

• It’s now possible to set the crash diagnostics directory through the environment variable CLANG_CRASH_DIAGNOSTICS_DIR. The -fcrash-diagnostics-dir flag takes precedence.

• Unicode support has been updated to support Unicode 15.0. New unicode codepoints are supported as appropriate in diagnostics, C and C++ identifiers, and escape sequences.

• In identifiers, Clang allows a restricted set of additional mathematical symbols as an extension. These symbols correspond to a proposed Unicode Mathematical notation profile for default identifiers. (#54732)

• Clang now supports loading multiple configuration files. The files from default configuration paths are loaded first, unless --no-default-config option is used. All files explicitly specified using --config= option are loaded afterwards.

• When loading default configuration files, clang now unconditionally uses the real target triple (respecting options such as --target= and -m32) rather than the executable prefix. The respective configuration files are also loaded when clang is called via an executable without a prefix (e.g. plain clang).

• Default configuration paths were partially changed. Clang now attempts to load <triple>-<driver>.cfg first, and falls back to loading both <driver>.cfg and <triple>.cfg if the former is not found. Triple is the target triple and driver first tries the canonical name for the driver (respecting --driver-mode=), and then the name found in the executable.

• If the environment variable SOURCE_DATE_EPOCH is set, it specifies a UNIX timestamp to be used in replacement of the current date and time in the __DATE__, __TIME__, and __TIMESTAMP__ macros. See https://reproducible-builds.org/docs/source-date-epoch/.

• Clang now supports __has_constexpr_builtin function-like macro that evaluates to 1 if the builtin is supported and can be constant evaluated. It can be used to writing conditionally constexpr code that uses builtins.

• The time profiler (using -ftime-trace option) now traces various constant evaluation events.

• Clang can now generate a PCH when using -fdelayed-template-parsing for code with templates containing loop hint pragmas, OpenMP pragmas, and #pragma unused.

New Compiler Flags

• Implemented -fcoro-aligned-allocation flag. This flag implements Option 2 of P2014R0 aligned allocation of coroutine frames (P2014R0). With this flag, the coroutines will try to lookup aligned allocation function all the time. The compiler will emit an error if it fails to find aligned allocation function. So if the user code implemented self defined allocation function for coroutines, the existing code will be broken. A little divergence with P2014R0 is that clang will lookup ::operator new(size_­t, std::aligned_val_t, nothrow_­t) if there is get_­return_­object_­on_­allocation_­failure. We feel this is more consistent with the intention.

• Added --no-default-config to disable automatically loading configuration files using default paths.

• Added the new level, 3, to the -fstrict-flex-arrays= flag. The new level is the strict, standards-conforming mode for flexible array members. It recognizes only incomplete arrays as flexible array members (which is how the feature is defined by the C standard).

  struct foo {
    int a;
    int b[]; // Flexible array member.
  };
  
  struct bar {
    int a;
    int b[0]; // NOT a flexible array member.
  };
  

• Added -fmodule-output to enable the one-phase compilation model for standard C++ modules. See Standard C++ Modules for more information.

• Added -Rpass-analysis=stack-frame-layout which will emit new diagnostic information about the layout of stack frames through the remarks infrastructure. Since it uses remarks the diagnostic information is available both on the CLI, and in a machine readable format.

Deprecated Compiler Flags

• -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang has been deprecated. The flag will be removed in Clang 18. -ftrivial-auto-var-init=zero is now available unconditionally, to be compatible with GCC.

• -fcoroutines-ts has been deprecated. The flag will be removed in Clang 17. Please use -std=c++20 or higher to use standard C++ coroutines instead.

• -fmodules-ts has been deprecated. The flag will be removed in Clang 17. Please use -std=c++20 or higher to use standard C++ modules instead.

Modified Compiler Flags

• Clang now permits specifying --config= multiple times, to load multiple configuration files.

• The option -rtlib=platform can now be used for all targets to override a different option value (either one set earlier on the command line, or a built-in hardcoded default). (Previously the MSVC and Darwin targets didn’t allow this parameter combination.)

Removed Compiler Flags

• Clang now no longer supports -cc1 -fconcepts-ts. This flag has been deprecated and encouraged use of -std=c++20 since Clang 10, so we’re now removing it.

Attribute Changes in Clang

• Added support for __attribute__((guard(nocf))) and C++-style [[clang::guard(nocf)]], which is equivalent to __declspec(guard(nocf)) when using the MSVC environment. This is to support enabling Windows Control Flow Guard checks with the ability to disable them for specific functions when using the MinGW environment. This attribute is only available for Windows targets.

• Introduced a new function attribute __attribute__((nouwtable)) to suppress LLVM IR uwtable function attribute.

• Updated the value returned by __has_c_attribute(nodiscard) to 202003L based on the final date specified by the C2x committee draft. We already supported the ability to specify a message in the attribute, so there were no changes to the attribute behavior.

• Updated the value returned by __has_c_attribute(fallthrough) to 201910L based on the final date specified by the C2x committee draft. We previously used 201904L (the date the proposal was seen by the committee) by mistake. There were no other changes to the attribute behavior.

• Introduced a new record declaration attribute __attribute__((enforce_read_only_placement)) to support analysis of instances of a given type focused on read-only program memory placement. It emits a warning if something in the code provably prevents an instance from a read-only memory placement.

• Introduced new attribute __attribute__((target_version("cpu_features"))) and expanded the functionality of the existing attribute __attribute__((target_clones("cpu_features1","cpu_features2",...))) to support Function Multi Versioning on AArch64 target. It detects at runtime which function versions are supported by CPU and calls the one with highest priority. Refer to clang attributes for more details.

Improvements to Clang’s diagnostics

• Disabled a FIX-IT suggested for a case of bad conversion in system headers.

• Clang will now check compile-time determinable string literals as format strings. (#55805)

• -Wformat now recognizes %b for the printf/scanf family of functions and %B for the printf family of functions. (#56885)

• Introduced -Wsingle-bit-bitfield-constant-conversion, grouped under -Wbitfield-constant-conversion, which diagnoses implicit truncation when 1 is assigned to a 1-bit signed integer bitfield. To reduce potential false positives, this diagnostic will not diagnose use of the true macro (from <stdbool.h>) in C language mode despite the macro being defined to expand to 1. (#53253)

• Clang will now print more information about failed static assertions. In particular, simple static assertion expressions are evaluated to their compile-time value and printed out if the assertion fails.

• Diagnostics about uninitialized constexpr variables have been improved to mention the missing constant initializer.

• Correctly diagnose a future keyword if it exists as a keyword in the higher language version and specifies in which version it will be a keyword. This supports both C and C++.

• no_sanitize("...") on a global variable for known but not relevant sanitizers is now just a warning. It now says that this will be ignored instead of incorrectly saying no_sanitize only applies to functions and methods.

• No longer mention reinterpet_cast in the invalid constant expression diagnostic note when in C mode.

• Clang will now give a more suitale diagnostic for declaration of block scope identifiers that have external/internal linkage that has an initializer. (#57478)

• A new analysis pass will helps preserve sugar when combining deductions, in an order agnostic way. This will be in effect when deducing template arguments, when deducing function return type from multiple return statements, for the conditional operator, and for most binary operations. Type sugar is combined in a way that strips the sugar which is different between terms, and preserves those which are common.

• Correctly diagnose the use of an integer literal without a suffix whose underlying type is long long or unsigned long long as an extension in C89 mode . Clang previously only diagnosed if the literal had an explicit LL suffix.

• Clang now correctly diagnoses index that refers past the last possible element of FAM-like arrays.

• Clang now correctly emits a warning when dereferencing a void pointer in C mode. (#53631)

• Clang will now diagnose an overload set where a candidate has a constraint that refers to an expression with a previous error as nothing viable, so that it doesn’t generate strange cascading errors, particularly in cases where a subsuming constraint fails, which would result in a less-specific overload to be selected.

• Add a fix-it hint for the -Wdefaulted-function-deleted warning to explicitly delete the function.

• Fixed an accidental duplicate diagnostic involving the declaration of a function definition without a prototype which is preceded by a static declaration of the function with a prototype. (#58181)

• Copy-elided initialization of lock scopes is now handled differently in -Wthread-safety-analysis: annotations on the move constructor are no longer taken into account, in favor of annotations on the function returning the lock scope by value. This could result in new warnings if code depended on the previous undocumented behavior. As a side effect of this change, constructor calls outside of initializer expressions are no longer ignored, which can result in new warnings (or make existing warnings disappear).

• The wording of diagnostics regarding arithmetic on fixed-sized arrays and pointers is improved to include the type of the array and whether it’s cast to another type. This should improve comprehension for why an index is out-of-bounds.

• Clang now correctly points to the problematic parameter for the -Wnonnull warning. (#58273)

• Introduced -Wcast-function-type-strict and -Wincompatible-function-pointer-types-strict to warn about function type mismatches in casts and assignments that may result in runtime indirect call Control-Flow Integrity (CFI) failures. The -Wcast-function-type-strict diagnostic is grouped under -Wcast-function-type as it identifies a more strict set of potentially problematic function type casts.

• Clang will now disambiguate NTTP types when printing diagnostic that contain NTTP types. (#57562)

• Better error recovery for pack expansion of expressions. (#58673)

• Better diagnostics when the user has missed auto in a declaration. (#49129)

• Clang now diagnoses use of invalid or reserved module names in a module export declaration. Both are diagnosed as an error, but the diagnostic is suppressed for use of reserved names in a system header.

• -Winteger-overflow will diagnose overflow in more cases. (#58944)

• Clang has an internal limit of 2GB of preprocessed source code per compilation, including source reachable through imported AST files such as PCH or modules. When Clang hits this limit, it now produces notes mentioning which header and source files are consuming large amounts of this space. #pragma clang __debug sloc_usage can also be used to request this report.

• Clang no longer permits the keyword ‘bool’ in a concept declaration as a concepts-ts compatibility extension.

• Clang now diagnoses overflow undefined behavior in a constant expression while evaluating a compound assignment with remainder as operand.

• Add -Wreturn-local-addr, a GCC alias for -Wreturn-stack-address.

• Clang now suppresses -Wlogical-op-parentheses on (x && a || b) and (a || b && x) only when x is a string literal.

• Clang will now reject the GNU extension address of label in coroutines explicitly. (#56436)

• Clang now automatically adds [[clang::lifetimebound]] to the parameters of std::move, std::forward et al, this enables Clang to diagnose more cases where the returned reference outlives the object.

• Fix clang not properly diagnosing the failing subexpression when chained binary operators are used in a static_assert expression.

• -Wtautological-compare missed warnings for tautological comparisons involving a negative integer literal. (#42918)

• -Wcomma is no longer emitted for void returning functions. (#57151)

Bug Fixes in This Version

Bug Fixes to Compiler Builtins

• stdatomic.h will use the internal declarations when targeting pre-C++-23 on Windows platforms as the MSVC support requires newer C++ standard.

• Correct _Static_assert to accept the same set of extended integer constant expressions as is accpted in other contexts that accept them. (#57687)

• Fix crashes caused when builtin C++ language extension type traits were instantiated by templates with an unexpected number of arguments. (#57008)

• Fix __builtin_assume_aligned crash when the 1st arg is array type. (#57169)

• Fixes to builtin template emulation of regular templates. (#42102, #51928)

• Fix incorrect handling of inline builtins with asm labels.

• The builtin type trait __is_aggregate now returns true for arrays of incomplete types in accordance with the suggested fix for LWG3823

Bug Fixes to Attribute Support

• Fixes an accepts-invalid bug in C when using a _Noreturn function specifier on something other than a function declaration. (#56800)

• No longer assert/miscompile when trying to make a vectorized _BitInt type using the ext_vector_type attribute (the vector_size attribute was already properly diagnosing this case).

• Invalid destructor names are no longer accepted on template classes. (#56772)

• Improve compile-times with large dynamic array allocations with trivial constructors. (#56774)

• Fix a crash when a btf_type_tag attribute is applied to the pointee of a function pointer.

• In C mode, when e1 has __attribute__((noreturn)) but e2 doesn’t, (c ? e1 : e2) is no longer considered noreturn, fixing a miscompilation. (#59792)

• GNU attributes being applied prior to standard attributes would be handled improperly, which was corrected to match the behaviour exhibited by GCC. (#58229)

• Fix issue using __attribute__((format)) on non-variadic functions that expect more than one formatted argument.

• Clang will now no longer treat a C ‘overloadable’ function without a prototype as a variadic function with the attribute. This should make further diagnostics more clear.

Bug Fixes to C++ Support

• Address the thread identification problems in coroutines. (#47177, #47179)

• Reject non-type template arguments formed by casting a non-zero integer to a pointer in pre-C++17 modes, instead of treating them as null pointers.

• Fix template arguments of pointer and reference not taking the type as part of their identity. (#47136)

• Fix handling of unexpanded packs in template argument expressions. (#58679)

• Fix bug with using enum that could lead to enumerators being treated as if they were part of an overload set. (#58067, #59014, #54746)

• Fix bug where constant evaluation treated a pointer to member that points to a weak member as never being null. Such comparisons are now treated as non-constant.

• Fix issue that the standard C++ modules importer will call global constructor/destructor for the global variables in the importing modules. (#59765)

• Reject in-class defaulting of previously declared comparison operators. (#51227)

• Do not hide templated base members introduced via using-decl in derived class (useful specially for constrained members). (#50886)

Concepts Specific Fixes:

• Class member variables are now in scope when parsing a requires clause. (#55216)

• Required parameter pack to be provided at the end of the concept parameter list. (#48182)

• Correctly handle access-checks in requires expression. (#53364, #53334)

• Fixed an issue with concept requirement evaluation, where we incorrectly allowed implicit conversions to bool for a requirement. (#54524)

• Fix a crash when emitting a concept-related diagnostic. (#57415)

• Fix a crash when trying to form a recovery expression on a call inside a constraint, which re-evaluated the same constraint. (#53213, #45736)

• Fix an issue when performing constraints partial ordering on non-template functions. (#56154)

• Fix a number of recursively-instantiated constraint issues, which would possibly result in a stack overflow. (#44304, #50891)

• Finished implementing C++ DR2565, which results in a requirement becoming not satisfied in the event of an instantiation failures in a requires expression’s parameter list. We previously handled this correctly in a constraint evaluation context, but not in a requires clause evaluated as a boolean.

Consteval Specific Fixes:

• Fix a crash when generating code coverage information for an if consteval statement. (#57377)

• Fixed a crash-on-valid with consteval evaluation of a list-initialized constructor for a temporary object. (#55871)

• Fixes an assert crash caused by looking up missing vtable information on consteval virtual functions. (#55065)

Bug Fixes to AST Handling

• A SubstTemplateTypeParmType can now represent the pack index for a substitution from an expanded pack. (#56099)

• The template arguments of a variable template being accessed as a member will now be represented in the AST.

Miscellaneous Bug Fixes

• Clang configuration files are now read through the virtual file system rather than the physical one, if these are different.

• Fix an issue where -frewrite-includes generated line control directives with incorrect line numbers in some cases when a header file used an end of line character sequence that differed from the primary source file. (#59736)

• Clang 14 predeclared some builtin POSIX library functions in gnu2x mode, and Clang 15 accidentally stopped predeclaring those functions in that language mode. Clang 16 now predeclares those functions again. (#56607)

• Fix the bug of inserting the ZeroInitializationFixit before the template argument list of VarTemplateSpecializationDecl.

• Fix the bug where Clang emits constrained float intrinsics when specifying -ffp-model=strict -ffp-model=fast.

Miscellaneous Clang Crashes Fixed

• Fix a crash when evaluating a multi-dimensional array’s array filler expression is element-dependent. (#50601)

• Fix an assert that triggers a crash during template name lookup when a type is incomplete but was not also a TagType. (#57387)

• Fix a crash when attempting to default a virtual constexpr non-special member function in a derived class. (#57431)

• Fix -Wpre-c++17-compat crashing Clang when compiling C++20 code which contains deduced template specializations. (#57369, #57643, #57793)

• Fix a crash upon stray coloncolon token in C2x mode.

• Fix assert that triggers a crash during some types of list initialization that generate a CXXTemporaryObjectExpr instead of a InitListExpr. (#58302, #58753, #59100)

• Fix a crash where we attempt to define a deleted destructor. (#57516)

• Fix an issue that triggers a crash if we instantiate a hidden friend functions. (#54457)

• Fix an issue that makes Clang crash on lambda template parameters. (#57960)

• Fixed a crash in C++20 mode in Clang and Clangd when compile source with compilation errors. (#53628)

• Fix sanity check when value initializing an empty union so that it takes into account anonymous structs which is a GNU extension. (#58800)

Target Specific Changes

X86 Support

• Support -mindirect-branch-cs-prefix for call and jmp to indirect thunk.

• Fix 32-bit __fastcall and __vectorcall ABI mismatch with MSVC.

• Add ISA of AMX-FP16 which support _tile_dpfp16ps.

• Switch AVX512-BF16 intrinsics types from short to __bf16.

• Add support for PREFETCHI instructions.

• Support ISA of CMPCCXADD. * Support intrinsic of _cmpccxadd_epi32. * Support intrinsic of _cmpccxadd_epi64.

• Add support for RAO-INT instructions. * Support intrinsic of _aadd_i32/64 * Support intrinsic of _aand_i32/64 * Support intrinsic of _aor_i32/64 * Support intrinsic of _axor_i32/64

• Support ISA of AVX-IFMA. * Support intrinsic of _mm(256)_madd52hi_avx_epu64. * Support intrinsic of _mm(256)_madd52lo_avx_epu64.

• Support ISA of AVX-VNNI-INT8. * Support intrinsic of _mm(256)_dpbssd(s)_epi32. * Support intrinsic of _mm(256)_dpbsud(s)_epi32. * Support intrinsic of _mm(256)_dpbuud(s)_epi32.

• Support ISA of AVX-NE-CONVERT. * Support intrinsic of _mm(256)_bcstnebf16_ps. * Support intrinsic of _mm(256)_bcstnesh_ps. * Support intrinsic of _mm(256)_cvtneebf16_ps. * Support intrinsic of _mm(256)_cvtneeph_ps. * Support intrinsic of _mm(256)_cvtneobf16_ps. * Support intrinsic of _mm(256)_cvtneoph_ps. * Support intrinsic of _mm(256)_cvtneps_avx_pbh.

• -march=raptorlake, -march=meteorlake and -march=emeraldrapids are now supported.

• -march=sierraforest, -march=graniterapids and -march=grandridge are now supported.

• Lift _BitInt() supported max width from 128 to 8388608.

• Support intrinsics of _mm(256)_reduce_(add|mul|or|and)_epi8/16.

• Support intrinsics of _mm(256)_reduce_(max|min)_ep[i|u]8/16.

Arm and AArch64 Support

• The target(..) function attributes for AArch64 now accept: * "arch=<arch>" strings, that specify the architecture for a function as per the -march option. * "cpu=<cpu>" strings, that specify the cpu for a function as per the -mcpu option. * "tune=<cpu>" strings, that specify the tune cpu for a function as per -mtune. * "+<feature>", "+no<feature>" enables/disables the specific feature, for compatibility with GCC target attributes. * "<feature>", "no-<feature>" enabled/disables the specific feature, for backward compatibility with previous releases.

• -march values for targeting armv2, armv2A, armv3 and armv3M have been removed. Their presence gave the impression that Clang can correctly generate code for them, which it cannot.

• Support has been added for the following processors (-mcpu identifiers in parenthesis): * Arm Cortex-A715 (cortex-a715). * Arm Cortex-X3 (cortex-x3). * Arm Neoverse V2 (neoverse-v2)

• Strict floating point has been enabled for AArch64, which means that -ftrapping-math, -frounding-math, -ffp-model=strict, and -ffp-exception-behaviour=<arg> are now accepted.

Windows Support

• For the MinGW driver, added the options -mguard=none, -mguard=cf and -mguard=cf-nochecks (equivalent to /guard:cf-, /guard:cf and /guard:cf,nochecks in clang-cl) for enabling Control Flow Guard checks and generation of address-taken function table.

• Switched from SHA1 to BLAKE3 for PDB type hashing / -gcodeview-ghash

• Fixed code generation with emulated TLS, when the emulated TLS is enabled by default (with downstream patches; no upstream configurations default to this configuration, but some mingw downstreams change the default in this way).

• Improved detection of MinGW cross sysroots for finding sysroots provided by Linux distributions such as Fedora. Also improved such setups by avoiding to include /usr/include among the include paths when cross compiling with a cross sysroot based in /usr.

• Fix incorrect alignment attribute on the this parameter of certain non-complete destructors when using the Microsoft ABI. Issue 60465.

LoongArch Support

• Clang now supports LoongArch. Along with the backend, clang is able to build a large corpus of Linux applications. Test-suite 100% pass.

• Support basic option -march= which is used to select the target architecture, i.e. the basic set of ISA modules to be enabled. Possible values are loongarch64 and la464.

• Support basic option -mabi= which is used to select the base ABI type. Possible values are lp64d, lp64f, lp64s, ilp32d, ilp32f and ilp32s.

• Support extended options: -msoft-float, -msingle-float, -mdouble-float and mfpu=. See LoongArch toolchain conventions.

RISC-V Support

• sifive-7-rv32 and sifive-7-rv64 are no longer supported for -mcpu. Use sifive-e76, sifive-s76, or sifive-u74 instead.

• Native detections via -mcpu=native and -mtune=native are supported.

• Fix interaction of -mcpu and -march, RISC-V backend will take the architecture extension union of -mcpu and -march before, and now will take architecture extensions from -march if both are given.

• An ABI mismatch between GCC and Clang that related to the sign/zero-extension of integer scalars was fixed.

• An overall simplification of the RISC-V Vector intrinsics are done. The simplification is based on riscv-non-isa/rvv-intrinsic-doc#186.

• Intrinsics of vcompress and vmerge have been adjusted to have interfaces be aligned among vvm, vxm intrinsics. The adjustment is base on riscv-non-isa/rvv-intrinsic-doc#185.

• All RISC-V Vector intrinsics now share a __riscv_ prefix, based on the naming convention defined by riscv-non-isa/riscv-c-api-doc#31.

• Note that the RISC-V Vector C intrinsics are still under development. The RVV C Intrinsic Task Group is working towards a ratified v1.0.

• The rvv-intrinsic-doc provides compatible headers for transition from the previous implemented version to the current (v0.11) version.

CUDA/HIP Language Changes in Clang

• Allow the use of __noinline__ as a keyword (instead of __attribute__((noinline))) in lambda declarations.

CUDA Support

• Clang now supports CUDA SDK up to 11.8

• Added support for targeting sm_{87,89,90} GPUs.

AIX Support

• When using -shared, the clang driver now invokes llvm-nm to create an export list if the user doesn’t specify one via linker flag or pass an alternative export control option.

• Driver work done for -pg to link with the right paths and files.

• Improved support for -bcdtors:mbr and -bcdtors:csect linker flags when linking with -fprofile-generate.

• Enabled LTO support. Requires AIX 7.2 TL5 SP3 or newer, or AIX 7.3. LTO support is implemented with the libLTO.so plugin. To specify a different plugin, use the linker option -bplugin:<path to plugin>. To pass options to the plugin, use the linker option -bplugin_opt:<option>.

• -mcpu option’s values are checked against a list of known CPUs. An error is reported if the specified CPU model is not found.

WebAssembly Support

• The -mcpu=generic configuration now enables sign-ext and mutable-globals. These proposals are standardized and available in all major engines.

DWARF Support in Clang

Previously when emitting DWARFv4 and tuning for GDB, Clang would use DWARF v2’s DW_AT_bit_offset and DW_AT_data_member_location. Clang now uses DWARF v4’s DW_AT_data_bit_offset regardless of tuning.

Support for DW_AT_data_bit_offset was added in GDB 8.0. For earlier versions, you can use the -gdwarf-3 option to emit compatible DWARF.

Floating Point Support in Clang

• The driver option -menable-unsafe-fp-math has been removed. To enable unsafe floating-point optimizations use -funsafe-math-optimizations or -ffast-math instead.

• Add __builtin_elementwise_sin and __builtin_elementwise_cos builtins for floating point types only.

AST Matchers

• Add isInAnoymousNamespace matcher to match declarations in an anonymous namespace.

clang-format

• Add RemoveSemicolon option for removing ; after a non-empty function definition.

• Add RequiresExpressionIndentation option for configuring the alignment of requires-expressions. The default value of this option is OuterScope, which differs in behavior from clang-format 15. To match the default behavior of clang-format 15, use the Keyword value.

• Add IntegerLiteralSeparator option for fixing integer literal separators in C++, C#, Java, and JavaScript.

• Add BreakAfterAttributes option for breaking after a group of C++11 attributes before a function declaration/definition name.

• Add InsertNewlineAtEOF option for inserting a newline at EOF if missing.

• Add LineEnding option to deprecate DeriveLineEnding and UseCRLF.

libclang

• Introduced the new function clang_getUnqualifiedType, which mimics the behavior of QualType::getUnqualifiedType for CXType.

• Introduced the new function clang_getNonReferenceType, which mimics the behavior of QualType::getNonReferenceType for CXType.

• Introduced the new function clang_CXXMethod_isDeleted, which queries whether the method is declared = delete.

• Introduced the new function clang_CXXMethod_isCopyAssignmentOperator, which identifies whether a method cursor is a copy-assignment operator.

• Introduced the new function clang_CXXMethod_isMoveAssignmentOperator, which identifies whether a method cursor is a move-assignment operator.

• clang_Cursor_getNumTemplateArguments, clang_Cursor_getTemplateArgumentKind, clang_Cursor_getTemplateArgumentType, clang_Cursor_getTemplateArgumentValue and clang_Cursor_getTemplateArgumentUnsignedValue now work on struct, class, and partial template specialization cursors in addition to function cursors.

Static Analyzer

• Removed the deprecated -analyzer-store and -analyzer-opt-analyze-nested-blocks analyzer flags. scanbuild was also updated accordingly. Passing these flags will result in a hard error.

• Deprecate the consider-single-element-arrays-as-flexible-array-members analyzer-config option. This option will be still accepted, but a warning will be displayed. This option will be rejected, thus turned into a hard error starting with clang-17. Use -fstrict-flex-array=<N> instead if necessary.

• Trailing array objects of structs with single elements will be considered as flexible-array-members. Use -fstrict-flex-array=<N> to define what should be considered as flexible-array-member if needed.

Sanitizers

• -fsanitize-memory-param-retval is turned on by default. With -fsanitize=memory, passing uninitialized variables to functions and returning uninitialized variables from functions is more aggressively reported. -fno-sanitize-memory-param-retval restores the previous behavior.

Additional Information

A wide variety of additional information is available on the Clang web page. The web page contains versions of the API documentation which are up-to-date with the Git version of the source code. You can access versions of these documents specific to this release by going into the “clang/docs/” directory in the Clang tree.

If you have any questions or comments about Clang, please feel free to contact us on the Discourse forums (Clang Frontend category).