Clang 18.0.0git (In-Progress) Release Notes¶

Written by the LLVM Team

Warning

These are in-progress notes for the upcoming Clang 18 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 18.0.0git. 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 18.0.0git because of the opportunity they pose for disruption to existing code bases.

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)

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.

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"
}
```

What’s New in Clang 18.0.0git?¶

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

C++ Language Changes ¶

C++20 Feature Support¶

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.

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.

Resolutions to C++ Defect Reports¶

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.
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.

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.
Added #pragma clang fp reciprocal.

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 an 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.

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.

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]].

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 -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.
The fix-it emitted by -Wformat for scoped enumerations now take the enumeration’s underlying type into account instead of suggesting a type just based on the format string specifier being used.
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);
      |               ~~~~~~~^~~~~

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)
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).

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)

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).
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 <https://github.com/llvm/llvm-project/issues/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 <https://github.com/llvm/llvm-project/issues/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 https://github.com/llvm/llvm-project/issues/66612)
Clang now disambiguates NTTP types when printing diagnostics where the NTTP types are compared with the ‘diff’ method. (#66744 https://github.com/llvm/llvm-project/issues/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 defers the instantiation of explicit specifier until constraint checking completes (except deduction guides). Fixes: (#59827)
Fix crash when parsing nested requirement. Fixes: (#73112)
Clang now immediately instantiates function template specializations at the end of the definition of the corresponding function template when the definition appears after the first point of instantiation. (#73232)

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

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.

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.

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.

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):
- Arm Cortex-A520 (cortex-a520).
- Arm Cortex-A720 (cortex-a720).
- Arm Cortex-X4 (cortex-x4).

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.

LoongArch Support¶

RISC-V Support¶

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

CUDA/HIP Language Changes¶

CUDA Support¶

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.

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.

AST Matchers ¶

Add convertVectorExpr.
Add dependentSizedExtVectorType.
Add macroQualifiedType.

clang-format ¶

Add AllowBreakBeforeNoexceptSpecifier option.
Add AllowShortCompoundRequirementOnASingleLine option.
Change BreakAfterAttributes from Never to Leave in LLVM style.
Add BreakAdjacentStringLiterals option.

libclang ¶

Exposed arguments of clang::annotate.
clang::getCursorKindForDecl now recognizes linkage specifications such as extern "C" and reports them as CXCursor_LinkageSpec.

Static Analyzer ¶

Added a new checker core.BitwiseShift which reports situations where bitwise shift operators produce undefined behavior (because some operand is negative or too large).
Move checker alpha.unix.Errno out of the alpha package to unix.Errno.
Move checker alpha.unix.StdCLibraryFunctions out of the alpha package to unix.StdCLibraryFunctions.
Fix false positive in mutation check when using pointer to member function. (#66204:).
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)
A few crashes have been found and fixed using randomized testing related to the use of _BitInt() in tidy checks and in clang analysis. See #67212, #66782, #65889, #65888, and #65887

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).