.. If you want to modify sections/contents permanently, you should modify both ReleaseNotes.rst and ReleaseNotesTemplate.txt. =========================================== Clang |release| |ReleaseNotesTitle| =========================================== .. contents:: :local: :depth: 2 Written by the `LLVM Team `_ .. only:: PreRelease .. warning:: These are in-progress notes for the upcoming Clang |version| release. Release notes for previous releases can be found on `the Releases Page `_. Introduction ============ This document contains the release notes for the Clang C/C++/Objective-C frontend, part of the LLVM Compiler Infrastructure, release |release|. 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 ============================ - Clang will now emit a warning if the auto-detected GCC installation directory (i.e. the one with the largest version number) does not contain libstdc++ include directories although a "complete" GCC installation directory containing the include directories is available. It is planned to change the auto-detection to prefer the "complete" directory in the future. The warning will disappear if the libstdc++ include directories are either installed or removed for all GCC installation directories considered by the auto-detection; see the output of ``clang -v`` for a list of those directories. If the GCC installations cannot be modified and maintaining the current choice of the auto-detection is desired, the GCC installation directory can be selected explicitly using the ``--gcc-install-dir`` command line argument. This will silence the warning. It can also be disabled using the ``-Wno-gcc-install-dir-libstdcxx`` command line flag. - Scalar deleting destructor support has been aligned with MSVC when targeting the MSVC ABI. Clang previously implemented support for ``::delete`` by calling the complete object destructor and then the appropriate global delete operator (as is done for the Itanium ABI). The scalar deleting destructor is now called to destroy the object and deallocate its storage. This is an ABI change that can result in memory corruption when a program built for the MSVC ABI has portions compiled with clang 21 or earlier and portions compiled with a version of clang 22 (or MSVC). Consider a class ``X`` that declares a virtual destructor and an ``operator delete`` member with the destructor defined in library ``A`` and a call to `::delete`` in library ``B``. If library ``A`` is compiled with clang 21 and library ``B`` is compiled with clang 22, the ``::delete`` call might dispatch to the scalar deleting destructor emitted in library ``A`` which will erroneously call the member ``operator delete`` instead of the expected global delete operator. The old behavior is retained under ``-fclang-abi-compat=21`` flag. - Clang warning suppressions file, ``--warning-suppression-mappings=``, now will use the last matching entry instead of the longest one. - Trailing null statements in GNU statement expressions are no longer ignored by Clang; they now result in a void type. Clang previously matched GCC's behavior, which was recently clarified to be incorrect. .. code-block:: c++ // The resulting type is 'void', not 'int' void foo(void) { return ({ 1;; }); } - Downstream projects that previously linked only against ``clangDriver`` may now (also) need to link against the new ``clangOptions`` library, since options-related code has been moved out of the Driver into a separate library. - The ``clangFrontend`` library no longer depends on ``clangDriver``, which may break downstream projects that relied on this transitive dependency. - Clang now supports MSVC vector deleting destructors when targeting Windows. This means that vtables of classes with virtual destructors will contain a pointer to vector deleting destructor (instead of scalar deleting destructor) which in fact is a different symbol with different name and linkage. This may cause runtime failures if two binaries using the same class defining a virtual destructor are compiled with different versions of clang. C/C++ Language Potentially Breaking Changes ------------------------------------------- - The ``__has_builtin`` function now only considers the currently active target when being used with target offloading. - The ``-Wincompatible-pointer-types`` diagnostic now defaults to an error; it can still be downgraded to a warning by passing ``-Wno-error=incompatible-pointer-types``. (#GH74605) C++ Specific Potentially Breaking Changes ----------------------------------------- - For C++20 modules, the Reduced BMI mode will be the default option. This may introduce regressions if your build system supports two-phase compilation model but haven't support reduced BMI or it is a compiler bug or a bug in users code. - Clang now correctly diagnoses during constant expression evaluation undefined behavior due to member pointer access to a member which is not a direct or indirect member of the most-derived object of the accessed object but is instead located directly in a sibling class to one of the classes along the inheritance hierarchy of the most-derived object as ill-formed. Other scenarios in which the member is not member of the most derived object were already diagnosed previously. (#GH150709) .. code-block:: c++ struct A {}; struct B : A {}; struct C : A { constexpr int foo() const { return 1; } }; constexpr A a; constexpr B b; constexpr C c; constexpr auto mp = static_cast(&C::foo); static_assert((a.*mp)() == 1); // continues to be rejected static_assert((b.*mp)() == 1); // newly rejected static_assert((c.*mp)() == 1); // accepted - ``VarTemplateSpecializationDecl::getTemplateArgsAsWritten()`` method now returns ``nullptr`` for implicitly instantiated declarations. - ``__builtin_is_replaceable``, ``trivially_relocable_if_eligible``, and ``replaceable_if_eligible`` have been removed as `P2786 `_ have been removed from C++2c. ``__builtin_is_cpp_trivially_relocatable`` and ``__builtin_trivially_relocate`` have been kept back, with the `P2786 `_ semantics, except there is no longer a way to explicitly specify a type is relocatable. ABI Changes in This Version --------------------------- - Fix AArch64 argument passing for C++ empty classes with large explicitly specified alignment. AST Potentially Breaking Changes -------------------------------- - Abbreviated function templates and generic lambdas now have a valid begin source location. The begin source location of abbreviated function templates is the begin source location of the templated function. The begin source location of generic lambdas is the begin source location of the lambda introducer ``[...]``. AST Dumping Potentially Breaking Changes ---------------------------------------- - How nested name specifiers are dumped and printed changes, keeping track of clang AST changes. - Pretty-printing of atomic builtins ``__atomic_test_and_set`` and ``__atomic_clear`` in ``-ast-print`` output. These previously displayed an extra ```` argument, e.g.: ``__atomic_test_and_set(p, , 0)`` Now they are printed as: ``__atomic_test_and_set(p, 0)`` - Pretty-printing of templates with inherited (i.e. specified in a previous redeclaration) default arguments has been fixed. - Default arguments of template template parameters are pretty-printed now. - Pretty-printing of ``asm`` attributes are now always the first attribute on the right side of the declaration. Before we had, e.g.: ``__attribute__(("visibility")) asm("string")`` Now we have: ``asm("string") __attribute__(("visibility"))`` Which is accepted by both clang and gcc parsers. Clang Frontend Potentially Breaking Changes ------------------------------------------- - Members of anonymous unions/structs are now injected as ``IndirectFieldDecl`` into the enclosing record even if their names conflict with other names in the scope. These ``IndirectFieldDecl`` are marked invalid. Clang Python Bindings Potentially Breaking Changes -------------------------------------------------- - Return ``None`` instead of null cursors from ``Token.cursor`` - TypeKind ``ELABORATED`` is not used anymore, per clang AST changes removing ElaboratedTypes. The value becomes unused, and all the existing users should expect the former underlying type to be reported instead. - Remove ``AccessSpecifier.NONE`` kind. No libclang interfaces ever returned this kind. - Allow setting the path to the libclang library via environment variables: ``LIBCLANG_LIBRARY_PATH`` to specifiy the path to the containing folder, or ``LIBCLANG_LIBRARY_FILE`` to specify the path to the library file - ``TranslationUnit.reparse`` will now throw an exception when an error occurs. Previously, errors were silently ignored. What's New in Clang |release|? ============================== C++ Language Changes -------------------- - A new family of builtins ``__builtin_*_synthesizes_from_spaceship`` has been added. These can be queried to know whether the ``<`` (``lt``), ``>`` (``gt``), ``<=`` (``le``), or ``>=`` (``ge``) operators are synthesized from a ``<=>``. This makes it possible to optimize certain facilities by using the ``<=>`` operation directly instead of doing multiple comparisons. C++2c Feature Support ^^^^^^^^^^^^^^^^^^^^^ - Started the implementation of `P2686R5 `_ Constexpr structured bindings. At this time, references to constexpr and decomposition of *tuple-like* types are not supported (only arrays and aggregates are). C++23 Feature Support ^^^^^^^^^^^^^^^^^^^^^ C++20 Feature Support ^^^^^^^^^^^^^^^^^^^^^ - Clang now normalizes constraints before checking whether they are satisfied, as mandated by the standard. As a result, Clang no longer incorrectly diagnoses substitution failures in template arguments only used in concept-ids, and produces better diagnostics for satisfaction failure. (#GH61811) (#GH135190) C++17 Feature Support ^^^^^^^^^^^^^^^^^^^^^ Resolutions to C++ Defect Reports ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ C Language Changes ------------------ - Clang now supports the :ref:`__builtin_stack_address ` () builtin. The semantics match those of GCC's builtin with the same name. - Implemented the ``defer`` draft Technical Specification (`WG14 N3734 `_); it is enabled in C mode by passing ``-fdefer-ts``. Note, the details of this feature are subject to change given that the Technical Specification is not yet ratified. C2y Feature Support ^^^^^^^^^^^^^^^^^^^ - No longer triggering ``-Wstatic-in-inline`` in C2y mode; use of a static function or variable within an extern inline function is no longer a constraint per `WG14 N3622 `_. - Clang now supports `N3355 `_ Named Loops. - Clang's implementation of ``__COUNTER__`` was updated to conform to `WG14 N3457 `_. This includes adding pedantic warnings for the feature being an extension in other language modes as well as an error when the counter is expanded more than 2147483647 times. C23 Feature Support ^^^^^^^^^^^^^^^^^^^ - Added ``FLT_SNAN``, ``DBL_SNAN``, and ``LDBL_SNAN`` to Clang's ```` header in C23 and later modes. This implements `WG14 N2710 `_. - Fixed accepting as compatible unnamed tag types with the same fields within the same translation unit but from different types. - ``-MG`` now silences the "file not found" errors with ``#embed`` when scanning for dependencies and encountering an unknown file. #GH165632 Non-comprehensive list of changes in this release ------------------------------------------------- - Added ``__scoped_atomic_uinc_wrap`` and ``__scoped_atomic_udec_wrap``. - Removed OpenCL header-only feature macros (previously unconditionally enabled on SPIR-V and only selectively disabled via ``-D__undef_``). All OpenCL extensions and features are now centralized in OpenCLExtensions.def, allowing consistent control via ``getSupportedOpenCLOpts`` and ``-cl-ext``. - Added ``__builtin_elementwise_ldexp``. - Added ``__builtin_elementwise_fshl`` and ``__builtin_elementwise_fshr``. - ``__builtin_elementwise_abs`` can now be used in constant expression. - Added ``__builtin_elementwise_minnumnum`` and ``__builtin_elementwise_maxnumnum``. - Trapping UBSan (e.g. ``-fsanitize=undefined -fsanitize-trap=undefined``) now emits a string describing the reason for trapping into the generated debug info. This feature allows debuggers (e.g. LLDB) to display the reason for trapping if the trap is reached. The string is currently encoded in the debug info as an artificial frame that claims to be inlined at the trap location. The function used for the artificial frame is an artificial function whose name encodes the reason for trapping. The encoding used is currently the same as ``__builtin_verbose_trap`` but might change in the future. This feature is enabled by default but can be disabled by compiling with ``-fno-sanitize-debug-trap-reasons``. The feature has a ``basic`` and ``detailed`` mode (the default). The ``basic`` mode emits a hard-coded string per trap kind (e.g. ``Integer addition overflowed``) and the ``detailed`` mode emits a more descriptive string describing each individual trap (e.g. ``signed integer addition overflow in 'a + b'``). The ``detailed`` mode produces larger debug info than ``basic`` but is more helpful for debugging. The ``-fsanitize-debug-trap-reasons=`` flag can be used to switch between the different modes or disable the feature entirely. Note due to trap merging in optimized builds (i.e. in each function all traps of the same kind get merged into the same trap instruction) the trap reasons might be removed. To prevent this build without optimizations (i.e. use `-O0` or use the `optnone` function attribute) or use the `fno-sanitize-merge=` flag in optimized builds. - ``__builtin_elementwise_max`` and ``__builtin_elementwise_min`` functions for integer types can now be used in constant expressions. - A vector of booleans is now a valid condition for the ternary ``?:`` operator. This binds to a simple vector select operation. - Added ``__builtin_masked_load``, ``__builtin_masked_expand_load``, ``__builtin_masked_store``, ``__builtin_masked_compress_store`` for conditional memory loads from vectors. Binds to the LLVM intrinsics of the same name. - Added ``__builtin_masked_gather`` and ``__builtin_masked_scatter`` for conditional gathering and scattering operations on vectors. Binds to the LLVM intrinsics of the same name. - The ``__builtin_popcountg``, ``__builtin_ctzg``, and ``__builtin_clzg`` functions now accept fixed-size boolean vectors. - Use of ``__has_feature`` to detect the ``ptrauth_qualifier`` and ``ptrauth_intrinsics`` features has been deprecated, and is restricted to the arm64e target only. The correct method to check for these features is to test for the ``__PTRAUTH__`` macro. - Added a new builtin, ``__builtin_dedup_pack``, to remove duplicate types from a parameter pack. This feature is particularly useful in template metaprogramming for normalizing type lists. The builtin produces a new, unexpanded parameter pack that can be used in contexts like template argument lists or base specifiers. .. code-block:: c++ template struct TypeList; // The resulting type is TypeList using MyTypeList = TypeList<__builtin_dedup_pack...>; Currently, the use of ``__builtin_dedup_pack`` is limited to template arguments and base specifiers, it also must be used within a template context. - ``__builtin_assume_dereferenceable`` now accepts non-constant size operands. - Fixed a crash when the second argument to ``__builtin_assume_aligned`` was not constant (#GH161314) - Introduce support for :doc:`allocation tokens ` to enable allocator-level heap organization strategies. A feature to instrument all allocation functions with a token ID can be enabled via the ``-fsanitize=alloc-token`` flag. - A new generic byte swap builtin function ``__builtin_bswapg`` that extends the existing __builtin_bswap{16,32,64} function family to support all standard integer types. - A builtin ``__builtin_infer_alloc_token(, ...)`` is provided to allow compile-time querying of allocation token IDs, where the builtin arguments mirror those normally passed to an allocation function. - Clang now rejects the invalid use of ``constexpr`` with ``auto`` and an explicit type in C. (#GH163090) New Compiler Flags ------------------ - New option ``-fno-sanitize-debug-trap-reasons`` added to disable emitting trap reasons into the debug info when compiling with trapping UBSan (e.g. ``-fsanitize-trap=undefined``). - New option ``-fsanitize-debug-trap-reasons=`` added to control emitting trap reasons into the debug info when compiling with trapping UBSan (e.g. ``-fsanitize-trap=undefined``). - New options for enabling allocation token instrumentation: ``-fsanitize=alloc-token``, ``-falloc-token-max=``, ``-fsanitize-alloc-token-fast-abi``, ``-fsanitize-alloc-token-extended``. - The ``-resource-dir`` option is now displayed in the list of options shown by ``--help``. - New option ``-fmatrix-memory-layout`` added to control the memory layout of Clang matrix types. (e.g. ``-fmatrix-memory-layout=column-major`` or ``-fmatrix-memory-layout=row-major``). Lanai Support ^^^^^^^^^^^^^^ - The option ``-mcmodel={small,medium,large}`` is supported again. Deprecated Compiler Flags ------------------------- Modified Compiler Flags ----------------------- - The `-gkey-instructions` compiler flag is now enabled by default when DWARF is emitted for plain C/C++ and optimizations are enabled. (#GH149509) - The `-fconstexpr-steps` compiler flag now accepts value `0` to opt out of this limit. (#GH160440) - The `-fdevirtualize-speculatively` compiler flag is now supported to enable speculative devirtualization of virtual function calls, it's disabled by default. (#GH159685) Removed Compiler Flags ------------------------- Attribute Changes in Clang -------------------------- - The definition of a function declaration with ``[[clang::cfi_unchecked_callee]]`` inherits this attribute, allowing the attribute to only be attached to the declaration. Prior, this would be treated as an error where the definition and declaration would have differing types. - Instrumentation added by ``-fsanitize=function`` will also be omitted for indirect calls to function pointers and function declarations marked with ``[[clang::cfi_unchecked_callee]]``. - New format attributes ``gnu_printf``, ``gnu_scanf``, ``gnu_strftime`` and ``gnu_strfmon`` are added as aliases for ``printf``, ``scanf``, ``strftime`` and ``strfmon``. (#GH16219) - New function attribute `malloc_span` is added. It has semantics similar to that of the `malloc` attribute, but `malloc_span` applies not to functions returning pointers, but to functions returning span-like structures (i.e. those that contain a pointer field and a size integer field or two pointers). - Added new attribute ``modular_format`` to allow dynamically selecting at link time which aspects of a statically linked libc's printf (et al) implementation are required. This can reduce code size without requiring e.g. multilibs for printf features. Requires cooperation with the libc implementation. - On targets with Itanium C++ ABI, Clang now supports ``[[gnu:gcc_struct]]`` with the behavior similar to one existing in GCC. In particular, whenever ``-mms-bitfields`` command line option is provided (or if Microsoft-compatible structure layout is default on the target), ``[[gnu::gcc_struct]]`` requests the compiler to follow Itanium rules for the layout of an annotated structure. Improvements to Clang's diagnostics ----------------------------------- - Diagnostics messages now refer to ``structured binding`` instead of ``decomposition``, to align with `P0615R0 `_ changing the term. (#GH157880) - Clang now suppresses runtime behavior warnings for unreachable code in file-scope variable initializers, matching the behavior for functions. This prevents false positives for operations in unreachable branches of constant expressions. - Added a separate diagnostic group ``-Wfunction-effect-redeclarations``, for the more pedantic diagnostics for function effects (``[[clang::nonblocking]]`` and ``[[clang::nonallocating]]``). Moved the warning for a missing (though implied) attribute on a redeclaration into this group. Added a new warning in this group for the case where the attribute is missing/implicit on an override of a virtual method. - Remove ``-Wperf-constraint-implies-noexcept`` from ``-Wall``. This warning is somewhat nit-picky and attempts to resolve it, by adding ``noexcept``, can create new ways for programs to crash. (#GH167540) - Implemented diagnostics when retrieving the tuple size for types where its specialization of `std::tuple_size` produces an invalid size (either negative or greater than the implementation limit). (#GH159563) - Fixed fix-it hint for fold expressions. Clang now correctly places the suggested right parenthesis when diagnosing malformed fold expressions. (#GH151787) - Added fix-it hint for when scoped enumerations require explicit conversions for binary operations. (#GH24265) - Constant template parameters are now type checked in template definitions, including template template parameters. - Fixed an issue where emitted format-signedness diagnostics were not associated with an appropriate diagnostic id. Besides being incorrect from an API standpoint, this was user visible, e.g.: "format specifies type 'unsigned int' but the argument has type 'int' [-Wformat]" "signedness of format specifier 'u' is incompatible with 'c' [-Wformat]" This was misleading, because even though -Wformat is required in order to emit the diagnostics, the warning flag the user needs to concerned with here is -Wformat-signedness, which is also required and is not enabled by default. With the change you'll now see: "format specifies type 'unsigned int' but the argument has type 'int', which differs in signedness [-Wformat-signedness]" "signedness of format specifier 'u' is incompatible with 'c' [-Wformat-signedness]" and the API-visible diagnostic id will be appropriate. - Clang now produces better diagnostics for template template parameter matching involving 'auto' template parameters. - Fixed false positives in ``-Waddress-of-packed-member`` diagnostics when potential misaligned members get processed before they can get discarded. (#GH144729) - Fix a false positive warning in ``-Wignored-qualifiers`` when the return type is undeduced. (#GH43054) - Clang now emits a diagnostic with the correct message in case of assigning to const reference captured in lambda. (#GH105647) - Fixed false positive in ``-Wmissing-noreturn`` diagnostic when it was requiring the usage of ``[[noreturn]]`` on lambdas before C++23 (#GH154493). - Clang now diagnoses the use of ``#`` and ``##`` preprocessor tokens in attribute argument lists in C++ when ``-pedantic`` is enabled. The operators can be used in macro replacement lists with the usual preprocessor semantics, however, non-preprocessor use of tokens now triggers a pedantic warning in C++. Compilation in C mode is unchanged, and still permits these tokens to be used. (#GH147217) - Clang now diagnoses misplaced array bounds on declarators for template specializations in th same way as it already did for other declarators. (#GH147333) - A new warning ``-Walloc-size`` has been added to detect calls to functions decorated with the ``alloc_size`` attribute don't allocate enough space for the target pointer type. - The :doc:`ThreadSafetyAnalysis` attributes ``ACQUIRED_BEFORE(...)`` and ``ACQUIRED_AFTER(...)`` have been moved to the stable feature set and no longer require ``-Wthread-safety-beta`` to be used. - The :doc:`ThreadSafetyAnalysis` gains basic alias-analysis of capability pointers under ``-Wthread-safety-beta`` (still experimental), which reduces both false positives but also false negatives through more precise analysis. - Clang now looks through parenthesis for ``-Wundefined-reinterpret-cast`` diagnostic. - Fixed a bug where the source location was missing when diagnosing ill-formed placeholder constraints. - The two-element, unary mask variant of ``__builtin_shufflevector`` is now properly being rejected when used at compile-time. It was not implemented and caused assertion failures before (#GH158471). - Closed a loophole in the diagnosis of function pointer conversions changing extended function type information in C mode (#GH41465). Function conversions that were previously incorrectly accepted in case of other irrelevant conditions are now consistently diagnosed, identical to C++ mode. - Fix false-positive unused label diagnostic when a label is used in a named break or continue (#GH166013) - Clang now emits a diagnostic in case `vector_size` or `ext_vector_type` attributes are used with a negative size (#GH165463). - Clang no longer emits ``-Wmissing-noreturn`` for virtual methods where the function body consists of a `throw` expression (#GH167247). - A new warning ``-Wenum-compare-typo`` has been added to detect potential erroneous comparison operators when mixed with bitwise operators in enum value initializers. This can be locally disabled by explicitly casting the initializer value. - Clang now provides correct caret placement when attributes appear before `enum class` (#GH163224). - A new warning ``-Wshadow-header`` has been added to detect when a header file is found in multiple search directories (excluding system paths). - Clang now detects potential missing format and format_matches attributes on function, Objective-C method and block declarations when calling format functions. It is part of the format-nonliteral diagnostic (#GH60718) - Fixed a crash when enabling ``-fdiagnostics-format=sarif`` and the output carries messages like 'In file included from ...' or 'In module ...'. Now the include/import locations are written into `sarif.run.result.relatedLocations`. - Clang now generates a fix-it for C++20 designated initializers when the initializers do not match the declaration order in the structure. Improvements to Clang's time-trace ---------------------------------- Improvements to Coverage Mapping -------------------------------- - [MC/DC] Unary logical not `!` among binary operators is recognized as a part of the expression. (#GH124563) Bug Fixes in This Version ------------------------- - Fix a crash when marco name is empty in ``#pragma push_macro("")`` or ``#pragma pop_macro("")``. (#GH149762). - Fix a crash in variable length array (e.g. ``int a[*]``) function parameter type being used in ``_Countof`` expression. (#GH152826). - ``-Wunreachable-code`` now diagnoses tautological or contradictory comparisons such as ``x != 0 || x != 1.0`` and ``x == 0 && x == 1.0`` on targets that treat ``_Float16``/``__fp16`` as native scalar types. Previously the warning was silently lost because the operands differed only by an implicit cast chain. (#GH149967). - Fix crash in ``__builtin_function_start`` by checking for invalid first parameter. (#GH113323). - Fixed a crash with incompatible pointer to integer conversions in designated initializers involving string literals. (#GH154046) - Fix crash on CTAD for alias template. (#GH131342), (#GH131408) - Clang now emits a frontend error when a function marked with the `flatten` attribute calls another function that requires target features not enabled in the caller. This prevents a fatal error in the backend. - Fixed scope of typedefs present inside a template class. (#GH91451) - Builtin elementwise operators now accept vector arguments that have different qualifiers on their elements. For example, vector of 4 ``const float`` values and vector of 4 ``float`` values. (#GH155405) - Fixed inconsistent shadow warnings for lambda capture of structured bindings. Previously, ``[val = val]`` (regular parameter) produced no warnings with ``-Wshadow`` while ``[a = a]`` (where ``a`` is from ``auto [a, b] = std::make_pair(1, 2)``) incorrectly produced warnings. Both cases now consistently show no warnings with ``-Wshadow`` and show uncaptured-local warnings with ``-Wshadow-all``. (#GH68605) - Fixed a failed assertion with a negative limit parameter value inside of ``__has_embed``. (#GH157842) - Fixed an assertion when an improper use of the ``malloc`` attribute targeting a function without arguments caused us to try to access a non-existent argument. (#GH159080) - Fixed a failed assertion with empty filename arguments in ``__has_embed``. (#GH159898) - Fixed a failed assertion with empty filename in ``#embed`` directive. (#GH162951) - Fixed a crash triggered by unterminated ``__has_embed``. (#GH162953) - Accept empty enumerations in MSVC-compatible C mode. (#GH114402) - Fix a bug leading to incorrect code generation with complex number compound assignment and bitfield values, which also caused a crash with UBsan. (#GH166798) - Fixed false-positive shadow diagnostics for lambdas in explicit object member functions. (#GH163731) - Fix an assertion failure when a ``target_clones`` attribute is only on the forward declaration of a multiversioned function. (#GH165517) (#GH129483) - Fix a crash caused by invalid format string in printf-like functions with ``-Wunsafe-buffer-usage-in-libc-call`` option enabled. (#GH170496) - Fixed a crash when parsing ``#embed`` parameters with unmatched closing brackets. (#GH152829) - Fixed a crash when compiling ``__real__`` or ``__imag__`` unary operator on scalar value with type promotion. (#GH160583) - Fixed a crash when parsing invalid nested name specifier sequences containing a single colon. (#GH167905) - Fixed a crash when parsing malformed #pragma clang loop vectorize_width(4,8,16) by diagnosing invalid comma-separated argument lists. (#GH166325) - Clang now treats enumeration constants of fixed-underlying enums as the enumerated type. (#GH172118) - Fixed a failed assertion in the preprocessor when ``__has_embed`` parameters are missing parentheses. (#GH175088) Bug Fixes to Compiler Builtins ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - Fix an ambiguous reference to the builtin `type_info` (available when using `-fms-compatibility`) with modules. (#GH38400) Bug Fixes to Attribute Support ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - ``[[nodiscard]]`` is now respected on Objective-C and Objective-C++ methods (#GH141504) and on types returned from indirect calls (#GH142453). - Fixes some late parsed attributes, when applied to function definitions, not being parsed in function try blocks, and some situations where parsing of the function body is skipped, such as error recovery, code completion, and msvc-compatible delayed template parsing. (#GH153551) - Using ``[[gnu::cleanup(some_func)]]`` where some_func is annotated with ``[[gnu::error("some error")]]`` now correctly triggers an error. (#GH146520) - Fix a crash when the function name is empty in the `swift_name` attribute. (#GH157075) - Fixes crashes or missing diagnostics with the `device_kernel` attribute. (#GH161905) - Fix handling of parameter indexes when an attribute is applied to a C++23 explicit object member function. - Fixed several false positives and false negatives in function effect (`nonblocking`) analysis. (#GH166078) (#GH166101) (#GH166110) - Fix ``cleanup`` attribute by delaying type checks until after the type is deduced. (#GH129631) - Fix a crash when instantiating a function template with ``constructor`` or ``destructor`` attributes without a priority argument. (#GH169072) - Fix an assertion when using ``target_clones`` attribute with empty argument list. (#GH173684) - Fixed a crash when applying the ``sentinel`` attribute to block variables. (#GH173820) Bug Fixes to C++ Support ^^^^^^^^^^^^^^^^^^^^^^^^ - Diagnose binding a reference to ``*nullptr`` during constant evaluation. (#GH48665) - Suppress ``-Wdeprecated-declarations`` in implicitly generated functions. (#GH147293) - Fix a crash when deleting a pointer to an incomplete array (#GH150359). - Fixed a mismatched lambda scope bug when propagating up ``consteval`` within nested lambdas. (#GH145776) - Disallow immediate escalation in destructors. (#GH109096) - Fix an assertion failure when expression in assumption attribute (``[[assume(expr)]]``) creates temporary objects. - Fix the dynamic_cast to final class optimization to correctly handle casts that are guaranteed to fail (#GH137518). - Fix bug rejecting partial specialization of variable templates with auto NTTPs (#GH118190). - Fix a crash if errors "member of anonymous [...] redeclares" and "initializing multiple members of union" coincide (#GH149985). - Fix a crash when using ``explicit(bool)`` in pre-C++11 language modes. (#GH152729) - Fix the parsing of variadic member functions when the ellipis immediately follows a default argument.(#GH153445) - Fix a crash when using an explicit object parameter in a non-member function with an invalid return type.(#GH173943) - Fixed a bug that caused ``this`` captured by value in a lambda with a dependent explicit object parameter to not be instantiated properly. (#GH154054) - Fixed a bug where our ``member-like constrained friend`` checking caused an incorrect analysis of lambda captures. (#GH156225) - Fixed a crash when implicit conversions from initialize list to arrays of unknown bound during constant evaluation. (#GH151716) - Support the dynamic_cast to final class optimization with pointer authentication enabled. (#GH152601) - Fix the check for narrowing int-to-float conversions, so that they are detected in cases where converting the float back to an integer is undefined behaviour (#GH157067). - Stop rejecting C++11-style attributes on the first argument of constructors in older standards. (#GH156809). - Fix a crash when applying binary or ternary operators to two same function types with different spellings, where at least one of the function parameters has an attribute which affects the function type. - Fix an assertion failure when a ``constexpr`` variable is only referenced through ``__builtin_addressof``, and related issues with builtin arguments. (#GH154034) - Fix an assertion failure when taking the address on a non-type template parameter argument of object type. (#GH151531) - Suppress ``-Wdouble-promotion`` when explicitly asked for with C++ list initialization (#GH33409). - Fix the result of `__builtin_is_implicit_lifetime` for types with a user-provided constructor. (#GH160610) - Correctly deduce return types in ``decltype`` expressions. (#GH160497) (#GH56652) (#GH116319) (#GH161196) - Fixed a crash in the pre-C++23 warning for attributes before a lambda declarator (#GH161070). - Fix a crash when attempting to deduce a deduction guide from a non deducible template template parameter. (#130604) - Fix for clang incorrectly rejecting the default construction of a union with nontrivial member when another member has an initializer. (#GH81774) - Fixed a template depth issue when parsing lambdas inside a type constraint. (#GH162092) - Fix the support of zero-length arrays in SFINAE context. (#GH170040) - Diagnose unresolved overload sets in non-dependent compound requirements. (#GH51246) (#GH97753) - Fix a crash when extracting unavailable member type from alias in template deduction. (#GH165560) - Fix incorrect diagnostics for lambdas with init-captures inside braced initializers. (#GH163498) - Fixed an issue where templates prevented nested anonymous records from checking the deletion of special members. (#GH167217) - Fixed serialization of pack indexing types, where we failed to expand those packs from a PCH/module. (#GH172464) - Fixed spurious diagnoses of certain nested lambda expressions. (#GH149121) (#GH156579) - Fix the result of ``__is_pointer_interconvertible_base_of`` when arguments are qualified and passed via template parameters. (#GH135273) - Fixed a crash when evaluating nested requirements in requires-expressions that reference invented parameters. (#GH166325) - Fixed a crash when standard comparison categories (e.g. ``std::partial_ordering``) are defined with incorrect static member types. (#GH170015) (#GH56571) - Fixed a crash when parsing the ``enable_if`` attribute on C function declarations with identifier-list parameters. (#GH173826) - Fixed an assertion failure triggered by nested lambdas during capture handling. (#GH172814) - Fixed an assertion failure in vector conversions involving instantiation-dependent template expressions. (#GH173347) - Fixed an issue where recursive instantiation could lead to escape of SFINAE errors. (#GH179118) - Fixed an assertion failure in floating conversion narrowing caused by C++ constant expression checks in C23 mode. (#GH173847) Bug Fixes to AST Handling ^^^^^^^^^^^^^^^^^^^^^^^^^ - Fix incorrect name qualifiers applied to alias CTAD. (#GH136624) - Fixed ElaboratedTypes appearing within NestedNameSpecifier, which was not a legal representation. This is fixed because ElaboratedTypes don't exist anymore. (#GH43179) (#GH68670) (#GH92757) - Fix unrecognized html tag causing undesirable comment lexing (#GH152944) - Fix comment lexing of special command names (#GH152943) - Use `extern` as a hint to continue parsing when recovering from a malformed declaration. Miscellaneous Bug Fixes ^^^^^^^^^^^^^^^^^^^^^^^ - Fixed missing diagnostics of ``diagnose_if`` on templates involved in initialization. (#GH160776) Miscellaneous Clang Crashes Fixed ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - Fixed a crash when attempting to jump over initialization of a variable with variably modified type. (#GH175540) OpenACC Specific Changes ------------------------ Target Specific Changes ----------------------- AMDGPU Support ^^^^^^^^^^^^^^ NVPTX Support ^^^^^^^^^^^^^^ X86 Support ^^^^^^^^^^^ - More SSE, AVX and AVX512 intrinsics, including initializers and general arithmetic can now be used in C++ constant expressions. - Some SSE, AVX and AVX512 intrinsics have been converted to wrap generic __builtin intrinsics. - NOTE: Please avoid use of the __builtin_ia32_* intrinsics - these are not guaranteed to exist in future releases, or match behaviour with previous releases of clang or other compilers. - Remove `m[no-]avx10.x-[256,512]` and `m[no-]evex512` options from Clang driver. - Remove `[no-]evex512` feature request from intrinsics and builtins. - Change features `avx10.x-[256,512]` to `avx10.x`. - `-march=wildcatlake` is now supported. - `-march=novalake` is now supported. Arm and AArch64 Support ^^^^^^^^^^^^^^^^^^^^^^^ - Support has been added for the following processors (command-line identifiers in parentheses): - Ampere Computing Ampere1C (``ampere1c``) - Arm C1-Nano (``c1-nano``) - Arm C1-Pro (``c1-pro``) - Arm C1-Premium (``c1-premium``) - Arm C1-Ultra (``c1-ultra``) - More intrinsics for the following AArch64 instructions: FCVTZ[US], FCVTN[US], FCVTM[US], FCVTP[US], FCVTA[US] - Support level for Function Multi-Versioning (FMV) has been upgraded to Release in ACLE. - Resolver functions can use the PAC and BTI hardening settings. - Users can override function version priority. - Unreachable function versions are diagnosed and ignored. Android Support ^^^^^^^^^^^^^^^ Windows Support ^^^^^^^^^^^^^^^ - clang-cl now supports /arch:AVX10.1 and /arch:AVX10.2. - clang-cl now supports /vlen, /vlen=256 and /vlen=512. - Clang now supports MSVC vector deleting destructors (GH19772). LoongArch Support ^^^^^^^^^^^^^^^^^ - Enable linker relaxation by default for loongarch64. - Introduce LASX and LSX conversion intrinsics. - `__attribute__((target("lasx")))` now implies the `lsx` feature. - DWARF fission is now compatible with linker relaxations, allowing `-gsplit-dwarf` and `-mrelax` to be used together when building for the LoongArch platform. - Add support for LoongArch32, including toolchain options and pre-defined macros. RISC-V Support ^^^^^^^^^^^^^^ - Add support for `__attribute__((interrupt("rnmi")))` to be used with the `Smrnmi` extension. With this the `Smrnmi` extension is fully supported. - Add `-march=unset` to clear any previous `-march=` value. This ISA string will be computed from `-mcpu` or the platform default. - `__GCC_CONSTRUCTIVE_SIZE` and `__GCC_DESTRUCTIVE_SIZE` are changed to 64. These values are unstable according to `Clang's documentation `_. - DWARF fission is now compatible with linker relaxations, allowing `-gsplit-dwarf` and `-mrelax` to be used together when building for the RISC-V platform. CUDA/HIP Language Changes ^^^^^^^^^^^^^^^^^^^^^^^^^ - Clang now supports C++17 Class Template Argument Deduction (CTAD) in CUDA/HIP device code by treating deduction guides as if they were ``__host__ __device__``. - Clang avoids ambiguous CTAD in CUDA/HIP by not synthesizing duplicate implicit deduction guides when ``__host__`` and ``__device__`` constructors differ only in CUDA target attributes (same signature and constraints). - Clang diagnoses CUDA/HIP deduction guides that are annotated as host-only, device-only, or ``__global__`` as errors. Explicit ``__host__ __device__`` deduction guides remain accepted for now but are deprecated and will be rejected in a future version of Clang; deduction guides do not participate in code generation and are treated as implicitly host+device. - Clang preserves distinct implicit deduction guides for constructors that differ by constraints, so constraint-based CTAD works in CUDA/HIP device code as in standard C++. CUDA Support ^^^^^^^^^^^^ - Fixed an assertion failure when processing CUDA files with `-fsyntax-only` or device architecture flags. - Support calling `consteval` function between different target. PowerPC Support ^^^^^^^^^^^^^^^ - Prototyped Dense Math Facility builtins. - Prototyped initial support for AMO load builtins. - Add support for ``-fpatchable-function-entry`` on PPC64LE. AIX Support ^^^^^^^^^^^ NetBSD Support ^^^^^^^^^^^^^^ WebAssembly Support ^^^^^^^^^^^^^^^^^^^ - Fix a bug so that ``__has_attribute(musttail)`` is no longer true when WebAssembly's tail-call is not enabled. (#GH163256) - The `wasm32-wasi` target has been renamed to `wasm32-wasip1`. The old option is still recognized, though by default will emit a deprecation warning. AVR Support ^^^^^^^^^^^ DWARF Support in Clang ---------------------- Floating Point Support in Clang ------------------------------- Fixed Point Support in Clang ---------------------------- AST Matchers ------------ - Removed elaboratedType matchers, and related nested name specifier changes, following the corresponding changes in the clang AST. - Ensure ``hasBitWidth`` doesn't crash on bit widths that are dependent on template parameters. - Remove the ``dependentTemplateSpecializationType`` matcher, as the corresponding AST node was removed. This matcher was never very useful, since there was no way to match on its template name. - Add a boolean member ``IgnoreSystemHeaders`` to ``MatchFinderOptions``. This allows it to ignore nodes in system headers when traversing the AST. - ``hasConditionVariableStatement`` now supports ``for`` loop, ``while`` loop and ``switch`` statements. - Fixed detection of explicit parameter lists in ``LambdaExpr``. (#GH168452) - Added ``hasExplicitParameters`` for ``LambdaExpr`` as an output attribute to AST JSON dumps. - Add ``arrayTypeLoc`` matcher for matching ``ArrayTypeLoc``. clang-format ------------ - Add ``SpaceInEmptyBraces`` option and set it to ``Always`` for WebKit style. - Add ``NumericLiteralCase`` option for enforcing character case in numeric literals. - Add ``Leave`` suboption to ``IndentPPDirectives``. - Add ``AllowBreakBeforeQtProperty`` option. - Add ``BreakAfterOpenBracketBracedList'', ``BreakAfterOpenBracketFunction'', ``BreakAfterOpenBracketIf``, ``BreakAfterOpenBracketLoop``, ``BreakAfterOpenBracketSwitch``, ``BreakBeforeCloseBracketBracedList'', ``BreakBeforeCloseBracketFunction``, ``BreakBeforeCloseBracketIf``, ``BreakBeforeCloseBracketLoop``, ``BreakBeforeCloseBracketSwitch`` options. - Deprecate ``AlwaysBreak`` and ``BlockIndent`` suboptions from the ``AlignAfterOpenBracket`` option, and make ``AlignAfterOpenBracket`` a ``bool`` type. - Add ``AlignPPAndNotPP`` suboption to ``AlignTrailingComments``. - Rename ``(Binary|Decimal|Hex)MinDigits`` to ``...MinDigitsInsert`` and add ``(Binary|Decimal|Hex)MaxDigitsSeparator`` suboptions to ``IntegerLiteralSeparator``. libclang -------- Code Completion --------------- Static Analyzer --------------- New checkers or options ^^^^^^^^^^^^^^^^^^^^^^^ - Introduced the ``alpha.core.StoreToImmutable`` checker to catch writes to immutable memory. See the `documentation `__. (#GH150417) - Introduced the ``core.NullPointerArithm`` checker to catch arithmetic on null pointers. See the `documentation `__. (#GH157129) - The ``core.CallAndMessage`` checker gained a new checker option called ``ArgPointeeInitializednessComplete``, enabling suppression of diagnostics of this checker in case at least some parts of the object was initialized. By default it's disabled. (#GH164600) Crash and bug fixes ^^^^^^^^^^^^^^^^^^^ - Fixed a crash in the static analyzer that when the expression in an ``[[assume(expr)]]`` attribute was enclosed in parentheses. (#GH151529) - Fixed an assertion failure of ``[[assume(expr)]]`` when the expression couldn't be folded into a constant by the engine. (#GH151854) - ``__datasizeof`` (C++) and ``_Countof`` (C) no longer cause a failed assertion when given an operand of VLA type. (#GH151711) - Sometimes the ``unix.Malloc`` checker asserted when constructing a report under rare circumstances. (#GH149754) - The ``core.StackAddressEscape`` checker was crashing if a lambda or Objective-C block expression captured itself. (#GH169208) - Fixed an assertion in the ``alpha.unix.cstring`` checker package when non-default address-spaces were used in ``memcpy``. (#GH153498) - The engine sometimes crashed when modeling a parenthesis initializer-list expression. (#GH147686) - `crosscheck-with-z3 `__ config option sometimes crashed. (#GH168034) - The ``alpha.core.StdVariant`` checker sometimes crashed when unknown values or type aliases were involved. (#GH167341) - Hardened the engine for supporting platforms where different address-space pointers might have different sizes. (#GH151370) - The ``core.builtin.BuiltinFunctions`` checker crashed when passing ``_BitInt(N)`` or ``__int128_t`` to ``__builtin_add_overflow`` or similar checked arithmetic builtin functions. (#GH173795) - Fixed a crash introduced in clang-20 when analyzing some "swap" functions. (#GH178797) - Fixed a crash introduced in clang-20 in the ``unix.Malloc`` checker when analyzing functions that had both of the ``ownership_returns`` and ``ownership_takes`` attributes. (#GH183344) Improvements ^^^^^^^^^^^^ - The `expand-macros `__ analyzer config option now formats the macro expansions using LLVM-style clang-format. (#GH154743) - ``[[clang::suppress]]`` now can suppress diagnostics within primary templates. (#GH168954) - Improved the false-positive suppression for ``std::unique_ptr`` and ``std::shared_ptr`` in the ``unix.Malloc`` checker. (#GH60896) - Improved the false-positive suppression for ``unix.Malloc`` on protobuf code. (#GH162124) - Supporting parenthesized initialization. (#GH148875) - Fixed a false-positive of ``cplusplus.PlacementNew`` by assuming that placement-new doesn't allocate. (#GH149240) - The ``unix.Malloc`` checker can now detect use-after-free even if the address does not directly refer to the beginning of the object. (For example, taking the address of a field.) (#GH152446) - Improved the modeling of cstring lengths in the ``alpha.unix.cstring`` checkers. (#GH159795) - The ``strxfrm`` is modeled in the ``alpha.unix.cstring`` checkers. (#GH156507) - Handling of placement-new expressions were improved in the ``alpha.core.PointerArithm`` checker. (#GH155855) - The ``security.ArrayBound`` checker now mentions the element count in underflow reports. (#GH158639) - The engine had a ``PrivateMethodCache`` that was not cleaned up - potentially causing spuriously failing Objective-C unittest failures because that uses the same process address-space but different clang instances inside - filling up the static cache and cause spurious hits, thus crashes. It only affects unittests. (#GH161327) - The internal ``-analyze-function`` option now also accepts Unified Symbol Resolution (USR) names. See the `documentation `__. (#GH161666) - The ``dump-entry-point-stats-to-csv`` analyzer config now includes the Translation Unit path and the analysis entry point USR for each entry. The CSV output got many more improvements. See the `documentation `__. (#GH162839) - The engine became smarter in folding binary operations. (#GH161537) - The `model-path `__ analyzer config option now honors virtual file system overlays (``-ivfsoverlay``). Similarly, the ``optin.taint.TaintPropagation`` checker-specific ``Config`` also accepts a virtual file path for the taint configuration file. See the `documentation `__. (#GH164323, #GH159164) - The dynamic memory modeling got more accurate by keeping extent information for longer. (#GH163562) - The ``unix.BlockInCriticalSection`` checker started honoring ``defer_lock_t``. (#GH166573) - Improved loop unrolling for compile-time upper-bounded loops. See the documentation of `unroll-loops `__. (#GH169400) - Improved the ``cplusplus.Move`` checker by relaxing how it handles opaque function calls, making it more accurate. (#GH169626) Moved checkers ^^^^^^^^^^^^^^ - The functionality of the checkers in the package ``valist.*`` was consolidated as a single new checker called ``security.VAList``. Diagnostic messages were also improved. (#GH156682, #GH157846) Removed checkers ^^^^^^^^^^^^^^^^ - The ``alpha.core.CastSize`` checker was removed. It had a poor false-positives / true-positive ratio, thus rarely used in practice. (#GH156350) Diagnostic changes ^^^^^^^^^^^^^^^^^^ - Harmonized the ``unix.Malloc`` checker diagnostics to use "release" instead of "free". (#GH150935) - ``sarif-html`` outputs no longer report detected issues 3 times. (#GH158103) - ``sarif`` reports now also emit the ``IssueHash`` field. (#GH158159) .. _release-notes-sanitizers: Sanitizers ---------- - Improved documentation for legacy ``no_sanitize`` attributes. - Added ``__builtin_allow_sanitize_check("name")`` that returns true if the specified sanitizer is enabled for the current function (after inlining). This allows for conditional code execution based on sanitizer enablement, respecting ``no_sanitize`` attributes. It currently supports sanitizers: "address", "kernel-address", "hwaddress", "kernel-hwaddress", "memory", "kernel-memory", and "thread". Python Binding Changes ---------------------- - Exposed ``clang_Cursor_isFunctionInlined``. - Exposed ``clang_getCursorLanguage`` via ``Cursor.language``. - Add all missing ``CursorKind``s, ``TypeKind``s and ``ExceptionSpecificationKind``s from ``Index.h`` OpenMP Support -------------- - Added parsing and semantic analysis support for the ``need_device_addr`` modifier in the ``adjust_args`` clause. - Allow array length to be omitted in array section subscript expression. - Fixed non-contiguous strided update in the ``omp target update`` directive with the ``from`` clause. - Added support for threadset clause in task and taskloop directives. - Properly handle array section/assumed-size array privatization in C/C++. - Added support to handle new syntax of the ``uses_allocators`` clause. - Added support for ``variable-category`` modifier in ``default clause``. - Added support for ``defaultmap`` directive implicit-behavior ``storage``. - Added support for ``defaultmap`` directive implicit-behavior ``private``. - Added parsing and semantic analysis support for ``groupprivate`` directive. - Added support for 'omp fuse' directive. - Updated parsing and semantic analysis support for ``nowait`` clause to accept optional argument in OpenMP >= 60. - Added support for ``default`` clause on ``target`` directive. - Added parsing and semantic analysis support for ``need_device_ptr`` modifier to accept an optional fallback argument (``fb_nullify`` or ``fb_preserve``) with OpenMP >= 61. - ``use_device_ptr`` and ``use_device_addr`` now preserve the original host address when lookup fails. Improvements ^^^^^^^^^^^^ - Mapping of expressions that have base-pointers now conforms to the OpenMP's conditional pointer-attachment based on both pointee and poitner being present, and one being new. This also lays the foundation of supporting OpenMP 6.1's attach map-type modifier. - Several improvements were made to the handling of maps on list items involving multiple levels of pointer dereferences, including not mapping intermediate expressions, and grouping the items that share the same base-pointer, as belonging to the same containing structure. - Support of array-sections on ``use_device_addr`` was made more robust, including diagnosing when the array-section's base is not a named-variable. - Handling of ``use_device_addr`` and ``use_device_ptr`` in the presence of other maps with the same base-pointer/variable, was improved. - Preserve the initializer when variable declaration dedution fails for better error recovery. 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) `_.