Clang 18.1.8 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 18.1.8. 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 18.1.8 because of the opportunity they pose for disruption to existing code bases.

• Fix a bug in reversed argument for templated operators. This breaks code in C++20 which was previously accepted in C++17. Clang did not properly diagnose such casese in C++20 before this change. Eg:

  struct P {};
  template<class S> bool operator==(const P&, const S&);
  
  struct A : public P {};
  struct B : public P {};
  
  // This equality is now ambiguous in C++20.
  bool ambiguous(A a, B b) { return a == b; }
  
  template<class S> bool operator!=(const P&, const S&);
  // Ok. Found a matching operator!=.
  bool fine(A a, B b) { return a == b; }
  

  To reduce such widespread breakages, as an extension, Clang accepts this code with an existing warning -Wambiguous-reversed-operator warning. Fixes #53954.

• The CMake variable GCC_INSTALL_PREFIX (which sets the default --gcc-toolchain=) is deprecated and will be removed. Specify --gcc-install-dir= or --gcc-triple= in a configuration file as a replacement. (#77537)

C/C++ Language Potentially Breaking Changes

• The default extension name for PCH generation (-c -xc-header and -c -xc++-header) is now .pch instead of .gch.

• -include a.h probing a.h.gch will now ignore a.h.gch if it is not a clang pch file or a directory containing any clang pch file.

• Fixed a bug that caused __has_cpp_attribute and __has_c_attribute return incorrect values for some C++-11-style attributes. Below is a complete list of behavior changes.

  Test	Old value	New value
  __has_cpp_attribute(unused)	201603	0
  __has_cpp_attribute(gnu::unused)	201603	1
  __has_c_attribute(unused)	202106	0
  __has_cpp_attribute(clang::fallthrough)	201603	1
  __has_cpp_attribute(gnu::fallthrough)	201603	1
  __has_c_attribute(gnu::fallthrough)	201910	1
  __has_cpp_attribute(warn_unused_result)	201907	0
  __has_cpp_attribute(clang::warn_unused_result)	201907	1
  __has_cpp_attribute(gnu::warn_unused_result)	201907	1
  __has_c_attribute(warn_unused_result)	202003	0
  __has_c_attribute(gnu::warn_unused_result)	202003	1

• Fixed a bug in finding matching operator!= while adding reversed operator== as outlined in “The Equality Operator You Are Looking For” (P2468). Fixes (#68901).

C++ Specific Potentially Breaking Changes

• The name mangling rules for function templates has been changed to take into account the possibility that functions could be overloaded on their template parameter lists or requires-clauses. This causes mangled names to change for function templates in the following cases:

  • When a template parameter in a function template depends on a previous template parameter, such as template<typename T, T V> void f().

  • When the function has any constraints, whether from constrained template

    parameters or requires-clauses.

  • When the template parameter list includes a deduced type – either

    auto, decltype(auto), or a deduced class template specialization type.

  • When a template template parameter is given a template template argument

    that has a different template parameter list.

  This fixes a number of issues where valid programs would be rejected due to mangling collisions, or would in some cases be silently miscompiled. Clang will use the old manglings if -fclang-abi-compat=17 or lower is specified. (#48216), (#49884), and (#61273)

• The ClassScopeFunctionSpecializationDecl AST node has been removed. Dependent class scope explicit function template specializations now use DependentFunctionTemplateSpecializationInfo to store candidate primary templates and explicit template arguments. This should not impact users of Clang as a compiler, but it may break assumptions in Clang-based tools iterating over the AST.

• The warning -Wenum-constexpr-conversion is now also enabled by default on system headers and macros. It will be turned into a hard (non-downgradable) error in the next Clang release.

• The flag -fdelayed-template-parsing won’t be enabled by default with C++20 when targetting MSVC to match the behavior of MSVC. (MSVC Docs)

• Remove the hardcoded path to the imported modules for C++20 named modules. Now we require all the dependent modules to specified from the command line. See (#62707).

• Forbid import XXX; in C++ to find module XXX comes from explicit clang modules. See (#64755).

ABI Changes in This Version

• Following the SystemV ABI for x86-64, __int128 arguments will no longer be split between a register and a stack slot.

• Fixed Microsoft calling convention for returning certain classes with a templated constructor. If a class has a templated constructor, it should be returned indirectly even if it meets all the other requirements for returning a class in a register. This affects some uses of std::pair. (#GH86384).

AST Dumping Potentially Breaking Changes

• When dumping a sugared type, Clang will no longer print the desugared type if its textual representation is the same as the sugared one. This applies to both text dumps of the form 'foo':'foo' which will now be dumped as just 'foo', and JSON dumps of the form:

  "type": {
    "qualType": "foo",
    "desugaredQualType": "foo"
  }
  

  which will now be dumped as just:

  "type": {
    "qualType": "foo"
  }
  

Clang Frontend Potentially Breaking Changes

• Target OS macros extension A new Clang extension (see here) is enabled for Darwin (Apple platform) targets. Clang now defines TARGET_OS_* macros for these targets, which could break existing code bases with improper checks for the TARGET_OS_ macros. For example, existing checks might fail to include the TargetConditionals.h header from Apple SDKs and therefore leaving the macros undefined and guarded code unexercised.

  Affected code should be checked to see if it’s still intended for the specific target and fixed accordingly.

  The extension can be turned off by the option -fno-define-target-os-macros as a workaround.

What’s New in Clang 18.1.8?

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++17 Feature Support

• Clang now exposes __GCC_DESTRUCTIVE_SIZE and __GCC_CONSTRUCTIVE_SIZE predefined macros to support standard library implementations of std::hardware_destructive_interference_size and std::hardware_constructive_interference_size, respectively. These macros are predefined in all C and C++ language modes. The values the macros expand to are not stable between releases of Clang and do not need to match the values produced by GCC, so these macros should not be used from header files because they may not be stable across multiple TUs (the values may vary based on compiler version as well as CPU tuning). #GH60174

C++20 Feature Support

• Implemented P1907R1 which extends allowed non-type template argument kinds with e.g. floating point values and pointers and references to subobjects. This feature is still experimental. Accordingly, __cpp_nontype_template_args was not updated. However, its support can be tested with __has_extension(cxx_generalized_nttp).

• Clang won’t perform ODR checks for decls in the global module fragment any more to ease the implementation and improve the user’s using experience. This follows the MSVC’s behavior. Users interested in testing the more strict behavior can use the flag ‘-Xclang -fno-skip-odr-check-in-gmf’. (#79240).

C++23 Feature Support

• Implemented P0847R7: Deducing this. Some related core issues were also implemented (CWG2553, CWG2554, CWG2653, CWG2687). Because the support for this feature is still experimental, the feature test macro __cpp_explicit_this_parameter was not set in this version. However, its support can be tested with __has_extension(cxx_explicit_this_parameter).

• Added a separate warning to warn the use of attributes on lambdas as a C++23 extension in previous language versions: -Wc++23-lambda-attributes.

C++2c Feature Support

• Implemented P2169R4: A nice placeholder with no name. This allows using _ as a variable name multiple times in the same scope and is supported in all C++ language modes as an extension. An extension warning is produced when multiple variables are introduced by _ in the same scope. Unused warnings are no longer produced for variables named _. Currently, inspecting placeholders variables in a debugger when more than one are declared in the same scope is not supported.

  struct S {
    int _, _; // Was invalid, now OK
  };
  void func() {
    int _, _; // Was invalid, now OK
  }
  void other() {
    int _; // Previously diagnosed under -Wunused, no longer diagnosed
  }
  

• Attributes now expect unevaluated strings in attributes parameters that are string literals. This is applied to both C++ standard attributes, and other attributes supported by Clang. This completes the implementation of P2361R6 Unevaluated Strings

• Implemented P2864R2 Remove Deprecated Arithmetic Conversion on Enumerations From C++26.

• Implemented P2361R6 Template parameter initialization. This change is applied as a DR in all language modes.

Resolutions to C++ Defect Reports

• Implemented CWG2598 and CWG2096, making unions (that have either no members or at least one literal member) literal types. (#77924).

C Language Changes

• structs, unions, and arrays that are const may now be used as constant expressions. This change is more consistent with the behavior of GCC.

• Enums will now be represented in TBAA metadata using their actual underlying integer type. Previously they were treated as chars, which meant they could alias with all other types.

• Clang now supports the C-only attribute counted_by. When applied to a struct’s flexible array member, it points to the struct field that holds the number of elements in the flexible array member. This information can improve the results of the array bound sanitizer and the __builtin_dynamic_object_size builtin.

C23 Feature Support

• Clang now accepts -std=c23 and -std=gnu23 as language standard modes, and the __STDC_VERSION__ macro now expands to 202311L instead of its previous placeholder value. Clang continues to accept -std=c2x and -std=gnu2x as aliases for C23 and GNU C23, respectively.

• Clang now supports requires c23 for module maps.

• Clang now supports N3007 Type inference for object definitions.

• Clang now supports <stdckdint.h> which defines several macros for performing checked integer arithmetic. It is also exposed in pre-C23 modes.

• Completed the implementation of N2508. We previously implemented allowing a label at the end of a compound statement, and now we’ve implemented allowing a label to be followed by a declaration instead of a statement.

• Implemented N2940 which removes support for trigraphs in C23 and later. In earlier language modes, trigraphs remain enabled by default in conforming modes (e.g. -std=c17) and disabled by default in GNU and Microsoft modes (e.g., -std=gnu17 or -fms-compatibility). If needed, you can enable trigraphs by passing -ftrigraphs.

Non-comprehensive list of changes in this release

• Clang now has a __builtin_vectorelements() function that determines the number of elements in a vector. For fixed-sized vectors, e.g., defined via __attribute__((vector_size(N))) or ARM NEON’s vector types (e.g., uint16x8_t), this returns the constant number of elements at compile-time. For scalable vectors, e.g., SVE or RISC-V V, the number of elements is not known at compile-time and is determined at runtime.

• The __datasizeof keyword has been added. It is similar to sizeof except that it returns the size of a type ignoring tail padding.

• __builtin_classify_type() now classifies _BitInt values as the return value 18 and vector types as return value 19, to match GCC 14’s behavior.

• The default value of _MSC_VER was raised from 1920 to 1933.

• Since MSVC 19.33 added undocumented attribute [[msvc::constexpr]], this release adds the attribute as well.

• Added #pragma clang fp reciprocal.

• The version of Unicode used by Clang (primarily to parse identifiers) has been updated to 15.1.

• Clang now defines macro __LLVM_INSTR_PROFILE_GENERATE when compiling with PGO instrumentation profile generation, and __LLVM_INSTR_PROFILE_USE when compiling with PGO profile use.

New Compiler Flags

• -fverify-intermediate-code and its complement -fno-verify-intermediate-code. Enables or disables verification of the generated LLVM IR. Users can pass this to turn on extra verification to catch certain types of compiler bugs at the cost of extra compile time. Since enabling the verifier adds a non-trivial cost of a few percent impact on build times, it’s disabled by default, unless your LLVM distribution itself is compiled with runtime checks enabled.

• -fkeep-system-includes modifies the behavior of the -E option, preserving #include directives for “system” headers instead of copying the preprocessed text to the output. This can greatly reduce the size of the preprocessed output, which can be helpful when trying to reduce a test case.

• -fassume-nothrow-exception-dtor is added to assume that the destructor of a thrown exception object will not throw. The generated code for catch handlers will be smaller. A throw expression of a type with a potentially-throwing destructor will lead to an error.

• -fopenacc was added as a part of the effort to support OpenACC in Clang.

• -fcx-limited-range enables the naive mathematical formulas for complex division and multiplication with no NaN checking of results. The default is -fno-cx-limited-range, but this option is enabled by -ffast-math.

• -fcx-fortran-rules enables the naive mathematical formulas for complex multiplication and enables application of Smith’s algorithm for complex division. See SMITH, R. L. Algorithm 116: Complex division. Commun. ACM 5, 8 (1962). The default is -fno-cx-fortran-rules.

• -fvisibility-global-new-delete=<value> gives more freedom to users to control how and if Clang forces a visibility for the replaceable new and delete declarations. The option takes 4 values: force-hidden, force-protected, force-default and source; force-default is the default. Option values with prefix force- assign such declarations an implicit visibility attribute with the corresponding visibility. An option value of source implies that no implicit attribute is added. Without the attribute the replaceable global new and delete operators behave normally (like other functions) with respect to visibility attributes, pragmas and options (e.g --fvisibility=).

• Full register names can be used when printing assembly via -mregnames. This option now matches the one used by GCC.

• -fdefine-target-os-macros and its complement -fno-define-target-os-macros. Enables or disables the Clang extension to provide built-in definitions of a list of TARGET_OS_* macros based on the target triple.

  The extension is enabled by default for Darwin (Apple platform) targets.

Deprecated Compiler Flags

Modified Compiler Flags

• -Woverriding-t-option is renamed to -Woverriding-option.

• -Winterrupt-service-routine is renamed to -Wexcessive-regsave as a generalization

• -frewrite-includes now guards the original #include directives with __CLANG_REWRITTEN_INCLUDES, and __CLANG_REWRITTEN_SYSTEM_INCLUDES as appropriate.

• Introducing a new default calling convention for -fdefault-calling-conv: rtdcall. This new default CC only works for M68k and will use the new m68k_rtdcc CC on every functions that are not variadic. The -mrtd driver/frontend flag has the same effect when targeting M68k.

• -fvisibility-global-new-delete-hidden is now a deprecated spelling of -fvisibility-global-new-delete=force-hidden (-fvisibility-global-new-delete= is new in this release).

• -fprofile-update is enabled for -fprofile-generate.

Removed Compiler Flags

• -enable-trivial-auto-var-init-zero-knowing-it-will-be-removed-from-clang has been removed. It has not been needed to enable -ftrivial-auto-var-init=zero since Clang 16.

Attribute Changes in Clang

• On X86, a warning is now emitted if a function with __attribute__((no_caller_saved_registers)) calls a function without __attribute__((no_caller_saved_registers)), and is not compiled with -mgeneral-regs-only

• On X86, a function with __attribute__((interrupt)) can now call a function without __attribute__((no_caller_saved_registers)) provided that it is compiled with -mgeneral-regs-only

• When a non-variadic function is decorated with the format attribute, Clang now checks that the format string would match the function’s parameters’ types after default argument promotion. As a result, it’s no longer an automatic diagnostic to use parameters of types that the format style supports but that are never the result of default argument promotion, such as float. (#59824)

• Clang now supports [[clang::preferred_type(type-name)]] as an attribute which can be applied to a bit-field. This attribute helps to map a bit-field back to a particular type that may be better-suited to representing the bit- field but cannot be used for other reasons and will impact the debug information generated for the bit-field. This is most useful when mapping a bit-field of basic integer type back to a bool or an enumeration type, e.g.,

  enum E { Apple, Orange, Pear };
  struct S {
    [[clang::preferred_type(E)]] unsigned FruitKind : 2;
  };
  

  When viewing S::FruitKind in a debugger, it will behave as if the member was declared as type E rather than unsigned.

• Clang now warns you that the _Alignas attribute on declaration specifiers is ignored, changed from the former incorrect suggestion to move it past declaration specifiers. (#58637)

• Clang now introduced [[clang::coro_only_destroy_when_complete]] attribute to reduce the size of the destroy functions for coroutines which are known to be destroyed after having reached the final suspend point.

• Clang now introduced [[clang::coro_return_type]] and [[clang::coro_wrapper]] attributes. A function returning a type marked with [[clang::coro_return_type]] should be a coroutine. A non-coroutine function marked with [[clang::coro_wrapper]] is still allowed to return the such a type. This is helpful for analyzers to recognize coroutines from the function signatures.

• Clang now supports [[clang::code_align(N)]] as an attribute which can be applied to a loop and specifies the byte alignment for a loop. This attribute accepts a positive integer constant initialization expression indicating the number of bytes for the minimum alignment boundary. Its value must be a power of 2, between 1 and 4096(inclusive).

  void Array(int *array, size_t n) {
    [[clang::code_align(64)]] for (int i = 0; i < n; ++i) array[i] = 0;
  }
  
  template<int A>
  void func() {
    [[clang::code_align(A)]] for(;;) { }
  }
  

• Clang now introduced [[clang::coro_lifetimebound]] attribute. All parameters of a function are considered to be lifetime bound if the function returns a type annotated with [[clang::coro_lifetimebound]] and [[clang::coro_return_type]]. This analysis can be disabled for a function by annotating the function with [[clang::coro_disable_lifetimebound]].

Improvements to Clang’s diagnostics

• Clang constexpr evaluator now prints template arguments when displaying template-specialization function calls.

• Clang contexpr evaluator now displays notes as well as an error when a constructor of a base class is not called in the constructor of its derived class.

• Clang no longer emits -Wmissing-variable-declarations for variables declared with the register storage class.

• Clang’s -Wswitch-default flag now diagnoses whenever a switch statement does not have a default label.

• Clang’s -Wtautological-negation-compare flag now diagnoses logical tautologies like x && !x and !x || x in expressions. This also makes -Winfinite-recursion diagnose more cases. (#56035).

• Clang constexpr evaluator now diagnoses compound assignment operators against uninitialized variables as a read of uninitialized object. (#51536)

• Clang’s -Wformat-truncation now diagnoses snprintf call that is known to result in string truncation. (#64871). Existing warnings that similarly warn about the overflow in sprintf now falls under its own warning group `-Wformat-overflow so that it can be disabled separately from Wfortify-source. These two new warning groups have subgroups -Wformat-truncation-non-kprintf and -Wformat-overflow-non-kprintf, respectively. These subgroups are used when the format string contains %p format specifier. Because Linux kernel’s codebase has format extensions for %p, kernel developers are encouraged to disable these two subgroups by setting -Wno-format-truncation-non-kprintf and -Wno-format-overflow-non-kprintf in order to avoid false positives on the kernel codebase. Also clang no longer emits false positive warnings about the output length of %g format specifier and about %o, %x, %X with # flag.

• Clang now emits -Wcast-qual for functional-style cast expressions.

• Clang no longer emits irrelevant notes about unsatisfied constraint expressions on the left-hand side of || when the right-hand side constraint is satisfied. (#54678).

• Clang now prints its ‘note’ diagnostic in cyan instead of black, to be more compatible with terminals with dark background colors. This is also more consistent with GCC.

• Clang now displays an improved diagnostic and a note when a defaulted special member is marked constexpr in a class with a virtual base class (#64843).

• -Wfixed-enum-extension and -Wmicrosoft-fixed-enum diagnostics are no longer emitted when building as C23, since C23 standardizes support for enums with a fixed underlying type.

• When describing the failure of static assertion of == expression, clang prints the integer representation of the value as well as its character representation when the user-provided expression is of character type. If the character is non-printable, clang now shows the escpaed character. Clang also prints multi-byte characters if the user-provided expression is of multi-byte character type.

  Example Code:

  static_assert("A\n"[1] == U'🌍');
  

  BEFORE:

  source:1:15: error: static assertion failed due to requirement '"A\n"[1] == U'\U0001f30d''
  1 | static_assert("A\n"[1] == U'🌍');
    |               ^~~~~~~~~~~~~~~~~
  source:1:24: note: expression evaluates to ''
  ' == 127757'
  1 | static_assert("A\n"[1] == U'🌍');
    |               ~~~~~~~~~^~~~~~~~
  

  AFTER:

  source:1:15: error: static assertion failed due to requirement '"A\n"[1] == U'\U0001f30d''
  1 | static_assert("A\n"[1] == U'🌍');
    |               ^~~~~~~~~~~~~~~~~
  source:1:24: note: expression evaluates to ''\n' (0x0A, 10) == U'🌍' (0x1F30D, 127757)'
  1 | static_assert("A\n"[1] == U'🌍');
    |               ~~~~~~~~~^~~~~~~~
  

• Clang now always diagnoses when using non-standard layout types in offsetof . (#64619)

• Clang now diagnoses redefined defaulted constructor when redefined defaulted constructor with different exception specs. (#69094)

• Clang now diagnoses use of variable-length arrays in C++ by default (and under -Wall in GNU++ mode). This is an extension supported by Clang and GCC, but is very easy to accidentally use without realizing it’s a nonportable construct that has different semantics from a constant-sized array. (#62836)

• Clang changed the order in which it displays candidate functions on overloading failures. Previously, Clang used definition of ordering from the C++ Standard. The order defined in the Standard is partial and is not suited for sorting. Instead, Clang now uses a strict order that still attempts to push more relevant functions to the top by comparing their corresponding conversions. In some cases, this results in better order. E.g., for the following code

  struct Foo {
    operator int();
    operator const char*();
  };
  
  void test() { Foo() - Foo(); }
  

  Clang now produces a list with two most relevant builtin operators at the top, i.e. operator-(int, int) and operator-(const char*, const char*). Previously operator-(const char*, const char*) was the first element, but operator-(int, int) was only the 13th element in the output. However, new implementation does not take into account some aspects of C++ semantics, e.g. which function template is more specialized. This can sometimes lead to worse ordering.

• When describing a warning/error in a function-style type conversion Clang underlines only until the end of the expression we convert from. Now Clang underlines until the closing parenthesis.

  Before:

  warning: cast from 'long (*)(const int &)' to 'decltype(fun_ptr)' (aka 'long (*)(int &)') converts to incompatible function type [-Wcast-function-type-strict]
  24 | return decltype(fun_ptr)( f_ptr /*comment*/);
     |        ^~~~~~~~~~~~~~~~~~~~~~~~
  

  After:

  warning: cast from 'long (*)(const int &)' to 'decltype(fun_ptr)' (aka 'long (*)(int &)') converts to incompatible function type [-Wcast-function-type-strict]
  24 | return decltype(fun_ptr)( f_ptr /*comment*/);
     |        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  

• -Wzero-as-null-pointer-constant diagnostic is no longer emitted when using __null (or, more commonly, NULL when the platform defines it as __null) to be more consistent with GCC.

• Clang will warn on deprecated specializations used in system headers when their instantiation is caused by user code.

• Clang will now print static_assert failure details for arithmetic binary operators. Example:

  static_assert(1 << 4 == 15);
  

  will now print:

  error: static assertion failed due to requirement '1 << 4 == 15'
     48 | static_assert(1 << 4 == 15);
        |               ^~~~~~~~~~~~
  note: expression evaluates to '16 == 15'
     48 | static_assert(1 << 4 == 15);
        |               ~~~~~~~^~~~~
  

• Clang now diagnoses definitions of friend function specializations, e.g. friend void f<>(int) {}.

• Clang now diagnoses narrowing conversions involving const references. (#63151).

• Clang now diagnoses unexpanded packs within the template argument lists of function template specializations.

• The warning -Wnan-infinity-disabled is now emitted when INFINITY or NAN are used in arithmetic operations or function arguments in floating-point mode where INFINITY or NAN don’t have the expected values.

• Clang now diagnoses attempts to bind a bitfield to an NTTP of a reference type as erroneous converted constant expression and not as a reference to subobject.

• Clang now diagnoses auto and decltype(auto) in declarations of conversion function template (CWG1878)

• Clang now diagnoses the requirement that non-template friend declarations with requires clauses and template friend declarations with a constraint that depends on a template parameter from an enclosing template must be a definition.

• Clang now diagnoses incorrect usage of const and pure attributes, so -Wignored-attributes diagnoses more cases.

• Clang now emits more descriptive diagnostics for ‘unusual’ expressions (e.g. incomplete index expressions on matrix types or builtin functions without an argument list) as placement-args to new-expressions.

  Before:

  error: no matching function for call to 'operator new'
     13 |     new (__builtin_memset) S {};
        |     ^   ~~~~~~~~~~~~~~~~~~
  
  note: candidate function not viable: no known conversion from '<builtin fn type>' to 'int' for 2nd argument
      5 |     void* operator new(__SIZE_TYPE__, int);
        |           ^
  

  After:

  error: builtin functions must be directly called
     13 |     new (__builtin_memset) S {};
        |          ^
  

• Clang now diagnoses import before module declarations but not in global module fragment. (#67627).

• Clang now diagnoses include headers with angle in module purviews, which is not usually intended. (#68615)

• Clang now won’t mention invisible namespace when diagnose invisible declarations inside namespace. The original diagnostic message is confusing. (#73893)

Improvements to Clang’s time-trace

• Two time-trace scope variables are added. A time trace scope variable of ParseDeclarationOrFunctionDefinition with the function’s source location is added to record the time spent parsing the function’s declaration or definition. Another time trace scope variable of ParseFunctionDefinition is also added to record the name of the defined function.

Bug Fixes in This Version

• Fixed an issue where a class template specialization whose declaration is instantiated in one module and whose definition is instantiated in another module may end up with members associated with the wrong declaration of the class, which can result in miscompiles in some cases.

• Fix crash on use of a variadic overloaded operator. (#42535)

• Fix a hang on valid C code passing a function type as an argument to typeof to form a function declaration. (#64713)

• Clang now reports missing-field-initializers warning for missing designated initializers in C++. (#56628)

• Clang now respects -fwrapv and -ftrapv for __builtin_abs and abs builtins. (#45129, #45794)

• Fixed an issue where accesses to the local variables of a coroutine during await_suspend could be misoptimized, including accesses to the awaiter object itself. (#56301) The current solution may bring performance regressions if the awaiters have non-static data members. See #64945 for details.

• Clang now prints unnamed members in diagnostic messages instead of giving an empty ‘’. Fixes (#63759)

• Fix crash in __builtin_strncmp and related builtins when the size value exceeded the maximum value representable by int64_t. Fixes (#64876)

• Fixed an assertion if a function has cleanups and fatal erors. (#48974)

• Clang now emits an error if it is not possible to deduce array size for a variable with incomplete array type. (#37257)

• Clang’s -Wunused-private-field no longer warns on fields whose type is declared with [[maybe_unused]]. (#61334)

• For function multi-versioning using the target, target_clones, or target_version attributes, remove comdat for internal linkage functions. (#65114)

• Clang now reports -Wformat for bool value and char specifier confusion in scanf. Fixes (#64987)

• Support MSVC predefined macro expressions in constant expressions and in local structs.

• Correctly parse non-ascii identifiers that appear immediately after a line splicing (#65156)

• Clang no longer considers the loss of __unaligned qualifier from objects as an invalid conversion during method function overload resolution.

• Fix lack of comparison of declRefExpr in ASTStructuralEquivalence (#66047)

• Fix parser crash when dealing with ill-formed objective C++ header code. Fixes (#64836)

• Fix crash in implicit conversions from initialize list to arrays of unknown bound for C++20. Fixes (#62945)

• Clang now allows an _Atomic qualified integer in a switch statement. Fixes (#65557)

• Fixes crash when trying to obtain the common sugared type of decltype(instantiation-dependent-expr). Fixes (#67603)

• Fixes a crash caused by a multidimensional array being captured by a lambda (#67722).

• Fixes a crash when instantiating a lambda with requires clause. (#64462)

• Fixes a regression where the UserDefinedLiteral was not properly preserved while evaluating consteval functions. (#63898).

• Fix a crash when evaluating value-dependent structured binding variables at compile time. Fixes (#67690)

• Fixes a clang-17 regression where LLVM_UNREACHABLE_OPTIMIZE=OFF cannot be used with Release mode builds. (#68237).

• Fix crash in evaluating constexpr value for invalid template function. Fixes (#68542)

• Clang will correctly evaluate noexcept expression for template functions of template classes. Fixes (#68543, #42496, #77071, #77411)

• Fixed an issue when a shift count larger than __INT64_MAX__, in a right shift operation, could result in missing warnings about shift count >= width of type or internal compiler error.

• Fixed an issue with computing the common type for the LHS and RHS of a ?: operator in C. No longer issuing a confusing diagnostic along the lines of “incompatible operand types (‘foo’ and ‘foo’)” with extensions such as matrix types. Fixes (#69008)

• Clang no longer permits using the _BitInt types as an underlying type for an enumeration as specified in the C23 Standard. Fixes (#69619)

• Fixed an issue when a shift count specified by a small constant _BitInt(), in a left shift operation, could result in a faulty warnings about shift count >= width of type.

• Clang now accepts anonymous members initialized with designated initializers inside templates. Fixes (#65143)

• Fix crash in formatting the real/imaginary part of a complex lvalue. Fixes (#69218)

• No longer use C++ thread_local semantics in C23 when using thread_local instead of _Thread_local. Fixes (#70068) and (#69167)

• Fix crash in evaluating invalid lambda expression which forget capture this. Fixes (#67687)

• Fix crash from constexpr evaluator evaluating uninitialized arrays as rvalue. Fixes (#67317)

• Clang now properly diagnoses use of stand-alone OpenMP directives after a label (including case or default labels).

• Fix compiler memory leak for enums with underlying type larger than 64 bits. Fixes (#78311)

  Before:

  label:
  #pragma omp barrier // ok
  

  After:

  label:
  #pragma omp barrier // error: '#pragma omp barrier' cannot be an immediate substatement
  

• Fixed an issue that a benign assertion might hit when instantiating a pack expansion inside a lambda. (#61460)

• Fix crash during instantiation of some class template specializations within class templates. Fixes (#70375)

• Fix crash during code generation of C++ coroutine initial suspend when the return type of await_resume is not trivially destructible. Fixes (#63803)

• __is_trivially_relocatable no longer returns true for non-object types such as references and functions. Fixes (#67498)

• Fix crash when the object used as a static_assert message has size or data members which are not member functions.

• Support UDLs in static_assert message.

• Fixed false positive error emitted by clang when performing qualified name lookup and the current class instantiation has dependent bases. Fixes (#13826)

• Fix a clang-17 regression where a templated friend with constraints is not properly applied when its parameters reference an enclosing non-template class. Fixes (#71595)

• Fix the name of the ifunc symbol emitted for multiversion functions declared with the target_clones attribute. This addresses a linker error that would otherwise occur when these functions are referenced from other TUs.

• Fixes compile error that double colon operator cannot resolve macro with parentheses. Fixes (#64467)

• Clang’s -Wchar-subscripts no longer warns on chars whose values are known non-negative constants. Fixes (#18763)

• Fix crash due to incorrectly allowing conversion functions in copy elision. Fixes (#39319) and (#60182) and (#62157) and (#64885) and (#65568)

• Fix an issue where clang doesn’t respect detault template arguments that are added in a later redeclaration for CTAD. Fixes (#69987)

• Fix an issue where CTAD fails for explicit type conversion. Fixes (#64347)

• Fix crash when using C++ only tokens like :: in C compiler clang. Fixes (#73559)

• Clang now accepts recursive non-dependent calls to functions with deduced return type. Fixes (#71015)

• Fix assertion failure when initializing union containing struct with flexible array member using empty initializer list. Fixes (#77085)

• Fix assertion crash due to failed scope restoring caused by too-early VarDecl invalidation by invalid initializer Expr. Fixes (#30908)

• Clang now emits correct source location for code-coverage regions in if constexpr and if consteval branches. Untaken branches are now skipped. Fixes (#54419)

• Fix assertion failure when declaring a template friend function with a constrained parameter in a template class that declares a class method or lambda at different depth. Fixes (#75426)

• Fix an issue where clang cannot find conversion function with template parameter when instantiation of template class. Fixes (#77583)

• Fix an issue where CTAD fails for function-type/array-type arguments. Fixes (#51710)

• Fix crashes when using the binding decl from an invalid structured binding. Fixes (#67495) and (#72198)

• Fix assertion failure when call noreturn-attribute function with musttail attribute. Fixes (#76631) - The MS __noop builtin without an argument list is now accepted in the placement-args of new-expressions, matching MSVC’s behaviour.

• Fix an issue that caused MS __decspec(property) accesses as well as Objective-C++ property accesses to not be converted to a function call to the getter in the placement-args of new-expressions. Fixes (#65053)

• Fix an issue with missing symbol definitions when the first coroutine statement appears in a discarded if constexpr branch. Fixes (#78290)

• Fixed assertion failure with deleted overloaded unary operators. Fixes (#78314)

• The XCOFF object file format does not support aliases to symbols having common linkage. Clang now diagnoses the use of an alias for a common symbol when compiling for AIX.

• Clang now doesn’t produce false-positive warning -Wconstant-logical-operand for logical operators in C23. Fixes (#64356).

• Clang’s -Wshadow no longer warns when an init-capture is named the same as a class field unless the lambda can capture this. Fixes (#71976)

Bug Fixes to Compiler Builtins

Bug Fixes to Attribute Support

Bug Fixes to C++ Support

• Clang limits the size of arrays it will try to evaluate at compile time to avoid memory exhaustion. This limit can be modified by -fconstexpr-steps. (#63562)

• Fix a crash caused by some named unicode escape sequences designating a Unicode character whose name contains a -. (Fixes #64161)

• Fix cases where we ignore ambiguous name lookup when looking up members. (#22413), (#29942), (#35574) and (#27224).

• Clang emits an error on substitution failure within lambda body inside a requires-expression. This fixes: (#64138) and (#71684).

• Update FunctionDeclBitfields.NumFunctionDeclBits. This fixes: (#64171).

• Expressions producing nullptr are correctly evaluated by the constant interpreter when appearing as the operand of a binary comparison. (#64923)

• Fix a crash when an immediate invocation is not a constant expression and appear in an implicit cast. (#64949).

• Fix crash when parsing ill-formed lambda trailing return type. Fixes: (#64962) and (#28679).

• Fix a crash caused by substitution failure in expression requirements. (#64172) and (#64723).

• Fix crash when parsing the requires clause of some generic lambdas. (#64689)

• Fix crash when the trailing return type of a generic and dependent lambda refers to an init-capture. (#65067 and #63675)

• Clang now properly handles out of line template specializations when there is a non-template inner-class between the function and the class template. (#65810)

• Fix a crash when calling a non-constant immediate function in the initializer of a static data member. (#65985).

• Clang now properly converts static lambda call operator to function pointers on win32. (#62594)

• Fixed some cases where the source location for an instantiated specialization of a function template or a member function of a class template was assigned the location of a non-defining declaration rather than the location of the definition the specialization was instantiated from. (#26057)

• Fix a crash when a default member initializer of a base aggregate makes an invalid call to an immediate function. (#66324)

• Fix crash for a lambda attribute with a statement expression that contains a return. (#48527)

• Clang now no longer asserts when an UnresolvedLookupExpr is used as an expression requirement. (#66612)

• Clang now disambiguates NTTP types when printing diagnostics where the NTTP types are compared with the ‘diff’ method. (#66744)

• Fix crash caused by a spaceship operator returning a comparision category by reference. Fixes: (#64162)

• Fix a crash when calling a consteval function in an expression used as the size of an array. (#65520)

• Clang no longer tries to capture non-odr-used variables that appear in the enclosing expression of a lambda expression with a noexcept specifier. (#67492)

• Fix crash when fold expression was used in the initialization of default argument. Fixes: (#67395)

• Fixed a bug causing destructors of constant-evaluated structured bindings initialized by array elements to be called in the wrong evaluation context.

• Fix crash where ill-formed code was being treated as a deduction guide and we now produce a diagnostic. Fixes: (#65522)

• Fixed a bug where clang incorrectly considered implicitly generated deduction guides from a non-templated constructor and a templated constructor as ambiguous, rather than prefer the non-templated constructor as specified in [standard.group]p3.

• Fixed a crash caused by incorrect handling of dependence on variable templates with non-type template parameters of reference type. Fixes: (#65153)

• Clang now properly compares constraints on an out of line class template declaration definition. Fixes: (#61763)

• Fix a bug where implicit deduction guides are not correctly generated for nested template classes. Fixes: (#46200) (#57812)

• Diagnose use of a variable-length array in a coroutine. The design of coroutines is such that it is not possible to support VLA use. Fixes: (#65858)

• Fix bug where we were overriding zero-initialization of class members when default initializing a base class in a constant expression context. Fixes: (#69890)

• Fix crash when template class static member imported to other translation unit. Fixes: (#68769)

• Clang now rejects incomplete types for __builtin_dump_struct. Fixes: (#63506)

• Fixed a crash for C++98/03 while checking an ill-formed _Static_assert expression. Fixes: (#72025)

• Clang now defers the instantiation of explicit specifier until constraint checking completes (except deduction guides). Fixes: (#59827)

• Fix crash when parsing nested requirement. Fixes: (#73112)

• Fixed a crash caused by using return type requirement in a lambda. Fixes: (#63808) (#64607) (#64086)

• Fixed a crash where we lost uninstantiated constraints on placeholder NTTP packs. Fixes: (#63837)

• Fixed a regression where clang forgets how to substitute into constraints on template-template parameters. Fixes: (#57410) and (#76604)

• Fix a bug where clang would produce inconsistent values when std::source_location::current() was used in a function template. Fixes (#78128)

• Clang now allows parenthesized initialization of arrays in operator new[]. Fixes: (#68198)

• Fixes CTAD for aggregates on nested template classes. Fixes: (#77599)

• Fix crash when importing the same module with an dynamic initializer twice in different visibility. Fixes (#67893)

• Remove recorded #pragma once state for headers included in named modules. Fixes (#77995)

• Set the __cpp_auto_cast feature test macro in C++23 mode.

• Fix crash for inconsistent deducing state of function return types in importing modules. Fixes (#78830) Fixes (#60085)

• Fixed a bug where variables referenced by requires-clauses inside nested generic lambdas were not properly injected into the constraint scope. (#73418)

• Fix incorrect code generation caused by the object argument of static operator() and static operator[] calls not being evaluated. Fixes (#67976)

• Fix crash when using an immediate-escalated function at global scope. (#82258)

• Correctly immediate-escalate lambda conversion functions. (#82258)

• Fix a crash when an unresolved overload set is encountered on the RHS of a .* operator. (#53815)

• Fixed a regression in CTAD that a friend declaration that befriends itself may cause incorrect constraint substitution. (#86769)

Bug Fixes to AST Handling

• Fixed an import failure of recursive friend class template. Issue 64169

• Remove unnecessary RecordLayout computation when importing UnaryOperator. The computed RecordLayout is incorrect if fields are not completely imported and should not be cached. Issue 64170

• Fixed hasAnyBase not binding nodes in its submatcher. (#65421)

• Fixed a bug where RecursiveASTVisitor fails to visit the initializer of a bitfield. Issue 64916

• Fixed a bug where range-loop-analysis checks for trivial copyability, rather than trivial copy-constructibility Issue 47355

• Fixed a bug where Template Instantiation failed to handle Lambda Expressions with certain types of Attributes. (#76521)

Miscellaneous Bug Fixes

Miscellaneous Clang Crashes Fixed

• Fixed a crash when parsing top-level ObjC blocks that aren’t properly terminated. Clang should now also recover better when an @end is missing between blocks. Issue 64065

• Fixed a crash when check array access on zero-length element. Issue 64564

• Fixed a crash when an ObjC ivar has an invalid type. See (#68001)

• Fixed a crash in C when redefined struct is another nested redefinition. Issue 41302

• Fixed a crash when -ast-dump=json was used for code using class template deduction guides.

• Fixed a crash when a lambda marked as static referenced a captured variable in an expression. Issue 74608

• Fixed a crash with modules and a constexpr destructor. Issue 68702

OpenACC Specific Changes

• OpenACC Implementation effort is beginning with semantic analysis and parsing of OpenACC pragmas. The -fopenacc flag was added to enable these new, albeit incomplete changes. The _OPENACC macro is currently defined to 1, as support is too incomplete to update to a standards-required value.

• Added -fexperimental-openacc-macro-override, a command line option to permit overriding the _OPENACC macro to be any digit-only value specified by the user, which permits testing the compiler against existing OpenACC workloads in order to evaluate implementation progress.

Target Specific Changes

AMDGPU Support

• Use pass-by-reference (byref) in stead of pass-by-value (byval) for struct arguments in C ABI. Callee is responsible for allocating stack memory and copying the value of the struct if modified. Note that AMDGPU backend still supports byval for struct arguments.

• The default value for -mcode-object-version is now 5. See AMDHSA Code Object V5 Metadata for more details.

X86 Support

• Added option -m[no-]evex512 to disable ZMM and 64-bit mask instructions for AVX512 features.

• Support ISA of USER_MSR. * Support intrinsic of _urdmsr. * Support intrinsic of _uwrmsr.

• Support ISA of AVX10.1.

• -march=pantherlake and -march=clearwaterforest are now supported.

• Added ABI handling for __float128 to match with GCC.

• Emit warnings for options to enable knl/knm specific ISAs: AVX512PF, AVX512ER and PREFETCHWT1. From next version (LLVM 19), these ISAs’ intrinsic supports will be deprecated: * intrinsic series of _exp2a23_ * intrinsic series of _rsqrt28_ * intrinsic series of _rcp28_ * intrinsic series of _prefetch_i[3|6][2|4]gather_ * intrinsic series of _prefetch_i[3|6][2|4]scatter_

Arm and AArch64 Support

• C++ function name mangling has been changed to align with the specification (https://github.com/ARM-software/abi-aa/blob/main/aapcs64/aapcs64.rst). This affects C++ functions with SVE ACLE parameters. Clang will use the old manglings if -fclang-abi-compat=17 or lower is specified.

• New AArch64 asm constraints have been added for r8-r11(Uci) and r12-r15(Ucj).

• Support has been added for the following processors (-mcpu identifiers in parenthesis):

  For Arm:

  • Cortex-M52 (cortex-m52).

  For AArch64:

  • Cortex-A520 (cortex-a520).

  • Cortex-A720 (cortex-a720).

  • Cortex-X4 (cortex-x4).

• Alpha support has been added for SVE2.1 intrinsics.

• Support has been added for -fstack-clash-protection and -mstack-probe-size command line options.

• Function Multi Versioning has been extended to support Load-Acquire RCpc instructions v3 (rcpc3) as well as Memory Copy and Memory Set Acceleration instructions (mops) when targeting AArch64. The feature identifiers (in parenthesis) can be used with either of the target_version and target_clones attributes.

Android Support

• Android target triples are usually suffixed with a version. Clang searches for target-specific runtime and standard libraries in directories named after the target (e.g. if you’re building with --target=aarch64-none-linux-android21, Clang will look for lib/aarch64-none-linux-android21 under its resource directory to find runtime libraries). If an exact match isn’t found, Clang would previously fall back to a directory without any version (which would be lib/aarch64-none-linux-android in our example). Clang will now look for directories for lower versions and use the newest version it finds instead, e.g. if you have lib/aarch64-none-linux-android21 and lib/aarch64-none-linux-android29, -target aarch64-none-linux-android23 will use the former and -target aarch64-none-linux-android30 will use the latter. Falling back to a versionless directory will now emit a warning, and the fallback will be removed in Clang 19.

Windows Support

• Fixed an assertion failure that occurred due to a failure to propagate MSInheritanceAttr attributes to class template instantiations created for explicit template instantiation declarations.

• The -fno-auto-import option was added for MinGW targets. The option both affects code generation (inhibiting generating indirection via .refptr stubs for potentially auto imported symbols, generating smaller and more efficient code) and linking (making the linker error out on such cases). If the option only is used during code generation but not when linking, linking may succeed but the resulting executables may expose issues at runtime.

• Clang now passes relevant LTO options to the linker (LLD) in MinGW mode.

LoongArch Support

• Added builtins support for all LSX (128-bits SIMD) and LASX (256-bits SIMD) instructions.

• Added builtins support for approximate calculation instructions that were introduced in LoongArch Reference Manual V1.10.

• Made -mcmodel= compatible with LoongArch gcc that accepted normal, medium and extreme.

• The model attribute was now supported for overriding the default code model used to access global variables. The following values were supported: normal, medium and extreme.

  Example Code:

  int var __attribute((model("extreme")));
  

• Default to -fno-direct-access-external-data for non-PIC.

• An ABI mismatch with gcc/g++ about empty structs/unions passing was fixed.

• _mcount was generated instead of mcount.

RISC-V Support

• Unaligned memory accesses can be toggled by -m[no-]unaligned-access or the aliases -m[no-]strict-align.

• CodeGen of RV32E/RV64E was supported experimentally.

• CodeGen of ilp32e/lp64e was supported experimentally.

• Default ABI with F but without D was changed to ilp32f for RV32 and to lp64f for RV64.

• __attribute__((rvv_vector_bits(N))) is now supported for RVV vbool*_t types.

• -mtls-dialect=desc is now supported to enable TLS descriptors (TLSDESC).

CUDA/HIP Language Changes

CUDA Support

• Clang now supports CUDA SDK up to 12.3

• Added support for sm_90a

PowerPC Support

• Added nmmintrin.h to intrinsics headers.

• Added __builtin_ppc_fence as barrier of code motion, and __builtin_ppc_mffsl for corresponding instruction.

• Supported __attribute__((target("tune=cpu"))).

• Emit float-abi module flag on 64-bit ELFv2 PowerPC targets if long double type is used in current module.

AIX Support

• Introduced the -maix-small-local-exec-tls option to produce a faster access sequence for local-exec TLS variables where the offset from the TLS base is encoded as an immediate operand. This access sequence is not used for TLS variables larger than 32KB, and is currently only supported on 64-bit mode.

• Inline assembler supports VSR register in pure digits.

• Enabled ThinLTO support. Requires AIX 7.2 TL5 SP7 or newer, or AIX 7.3 TL2 or newer. Similar to the LTO support on AIX, ThinLTO is implemented with the libLTO.so plugin.

SystemZ Support

• Properly support 16 byte atomic int/fp types and ops. Atomic __int128 (and long double) variables are now aligned to 16 bytes by default (like gcc 14).

WebAssembly Support

AVR Support

DWARF Support in Clang

Floating Point Support in Clang

• Add __builtin_elementwise_log builtin for floating point types only.

• Add __builtin_elementwise_log10 builtin for floating point types only.

• Add __builtin_elementwise_log2 builtin for floating point types only.

• Add __builtin_elementwise_exp builtin for floating point types only.

• Add __builtin_elementwise_exp2 builtin for floating point types only.

• Add __builtin_set_flt_rounds builtin for X86, x86_64, Arm and AArch64 only.

• Add __builtin_elementwise_pow builtin for floating point types only.

• Add __builtin_elementwise_bitreverse builtin for integer types only.

• Add __builtin_elementwise_sqrt builtin for floating point types only.

• __builtin_isfpclass builtin now supports vector types.

• #pragma float_control(precise,on) enables precise floating-point semantics. If math-errno is disabled in the current TU, clang will re-enable math-errno in the presense of #pragma float_control(precise,on).

• Add __builtin_exp10, __builtin_exp10f, __builtin_exp10f16, __builtin_exp10l and __builtin_exp10f128 builtins.

• Add __builtin_iszero, __builtin_issignaling and __builtin_issubnormal.

• Add support for C99’s #pragma STDC CX_LIMITED_RANGE feature. This enables the naive mathematical formulas for complex multiplication and division, which are faster but do not correctly handle overflow and infinities.

AST Matchers

• Add convertVectorExpr.

• Add dependentSizedExtVectorType.

• Add macroQualifiedType.

• Add CXXFoldExpr related matchers: cxxFoldExpr, callee, hasInit, hasPattern, isRightFold, isLeftFold, isUnaryFold, isBinaryFold, hasOperator, hasLHS, hasRHS, hasEitherOperand.

clang-format

• Add AllowBreakBeforeNoexceptSpecifier option.

• Add AllowShortCompoundRequirementOnASingleLine option.

• Change BreakAfterAttributes from Never to Leave in LLVM style.

• Add BreakAdjacentStringLiterals option.

• Add ObjCPropertyAttributeOrder which can be used to sort ObjC property attributes (like nonatomic, strong, nullable).

• Add PenaltyBreakScopeResolution option.

• Add .clang-format-ignore files.

• Add AlignFunctionPointers sub-option for AlignConsecutiveDeclarations.

• Add SkipMacroDefinitionBody option.

libclang

• Exposed arguments of clang::annotate.

• clang::getCursorKindForDecl now recognizes linkage specifications such as extern "C" and reports them as CXCursor_LinkageSpec.

• Changed the libclang library on AIX to export only the necessary symbols to prevent issues of resolving to the wrong duplicate symbol.

Static Analyzer

New features

• Implemented the [[clang::suppress]] attribute for suppressing diagnostics of static analysis tools, such as the Clang Static Analyzer. Documentation.

• Support “Deducing this” (P0847R7). (Worked out of the box) (af4751738db8)

• Added a new checker core.BitwiseShift which reports situations where bitwise shift operators produce undefined behavior (because some operand is negative or too large). Documentation.

• Added a new experimental checker alpha.core.StdVariant to detect variant accesses via wrong alternatives. Documentation. (#66481)

• Added a new experimental checker alpha.cplusplus.ArrayDelete to detect destructions of arrays of polymorphic objects that are destructed as their base class (CERT EXP51-CPP). Documentation. (0e246bb67573)

• Added a new checker configuration option InvalidatingGetEnv=[true,false] to security.cert.env.InvalidPtr. It’s not set by default. If set, getenv calls won’t invalidate previously returned pointers. Documentation. (#67663)

Crash and bug fixes

• Fixed a crash caused by builtin_bit_cast. (#69922)

• Fixed a core.StackAddressEscape crash on temporary object fields. (#66221)

• A few crashes have been found and fixed using randomized testing related to the use of _BitInt() in tidy checks and in clang analysis. (#67212, #66782, #65889, #65888, #65887)

• Fixed note links of the HTML output. (#64054)

• Allow widening rage-based for loops. (#70190)

• Fixed uninitialized base class with initializer list when ctor is not declared in the base class. (#70464, #59493, #54533)

• Fixed an alpha.unix.cstring crash on variadic functions. (#74269)

• Fix false positive in mutation check when using pointer to member function. (#66204)

• Fixed a crash in security.cert.env.InvalidPtr checker when accidentally matched user-defined strerror and similar library functions. (#88181)

Improvements

• Improved the unix.StdCLibraryFunctions checker by modeling more functions like send, recv, readlink, fflush, mkdtemp, getcwd and errno behavior. (52ac71f92d38, #77040, #76671, #71373, #76557, #71392)

• Fixed a false negative for when accessing a nonnull property (ObjC). (1dceba3a3684)

• security.insecureAPI.DeprecatedOrUnsafeBufferHandling now considers fprintf calls unsafe. Documentation.

• Improved the diagnostics of the optin.core.EnumCastOutOfRange checker. It will display the name and the declaration of the enumeration along with the concrete value being cast to the enum. (#74503)

• Improved the alpha.security.ArrayBoundV2 checker for detecting buffer accesses prior the buffer; and also reworked the diagnostic messages. (3e014038b373, #70056, #72107)

• Improved the alpha.unix.cstring.OutOfBounds checking both ends of the buffers in more cases. (c3a87ddad62a, 0954dc3fb921)

• Improved the alpha.unix.Stream checker by modeling more functions fputs, fputc, fgets, fgetc, fdopen, ungetc, fflush, getdelim, getline and no not recognize alternative fopen and tmpfile implementations. (#78693, #76776, #74296, #73335, #72627, #71518, #72016, #70540, #73638, #77331)

• The alpha.security.taint.TaintPropagation checker no longer propagates taint on strlen and strnlen calls, unless these are marked explicitly propagators in the user-provided taint configuration file. This removal empirically reduces the number of false positive reports. Read the PR for the details. (#66086)

• Other taint-related improvements. (#66358, #66074, #66358)

• Checkers can query constraint bounds to improve diagnostic messages. (#74141)

• Improved the generated initializers for modules. Now the calls to initialize functions from imported module units can be omitted if the initialize function is known to be empty. (#56794)

• Clang now allow to export declarations within linkage-specification. (#71347)

Moved checkers

• Move checker alpha.unix.Errno out of the alpha package to unix.Errno. Documentation.

• Move checker alpha.unix.StdCLibraryFunctions out of the alpha package to unix.StdCLibraryFunctions. Documentation.

• Move checker alpha.security.cert.env.InvalidPtr out of the alpha package to security.cert.env.InvalidPtr. Documentation.

• Move checker alpha.cplusplus.EnumCastOutOfRange out of the alpha package to optin.core.EnumCastOutOfRange. Documentation.

Sanitizers

• -fsanitize=signed-integer-overflow now instruments __builtin_abs and abs builtins.

Python Binding Changes

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).