Clang 11.0.0 Release Notes

Written by the LLVM Team

Introduction

This document contains the release notes for the Clang C/C++/Objective-C frontend, part of the LLVM Compiler Infrastructure, release 11.0.0. Here we describe the status of Clang in some detail, including major improvements from the previous release and new feature work. For the general LLVM release notes, see the LLVM documentation. 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.

What’s New in Clang 11.0.0?

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

Recovery AST

clang’s AST now improves support for representing broken C++ code. This improves the quality of subsequent diagnostics after an error is encountered. It also exposes more information to tools like clang-tidy and clangd that consume clang’s AST, allowing them to be more accurate on broken code.

A RecoveryExpr is introduced in clang’s AST, marking an expression containing semantic errors. This preserves the source range and subexpressions of the broken expression in the AST (rather than discarding the whole expression).

For the following invalid code:

int NoArg(); // Line 1
int x = NoArg(42); // oops!

clang-10 produces the minimal placeholder:

// VarDecl <line:2:1, col:5> col:5 x 'int'

clang-11 produces a richer AST:

// VarDecl <line:2:1, col:16> col:5 x 'int' cinit
// `-RecoveryExpr <col:9, col:16> '<dependent type>' contains-errors lvalue
//    `-UnresolvedLookupExpr <col:9> '<overloaded function>' lvalue (ADL) = 'NoArg'
//    `-IntegerLiteral <col:15> 'int' 42

Note that error-dependent types and values may now occur outside a template context. Tools may need to adjust assumptions about dependent code.

This feature is on by default for C++ code, and can be explicitly controlled with -Xclang -f[no-]recovery-ast.

Improvements to Clang’s diagnostics

• -Wpointer-to-int-cast is a new warning group. This group warns about C-style casts of pointers to a integer type too small to hold all possible values.

• -Wuninitialized-const-reference is a new warning controlled by -Wuninitialized. It warns on cases where uninitialized variables are passed as const reference arguments to a function.

• -Wimplicit-const-int-float-conversion (enabled by default) is a new option controlled by -Wimplicit-int-float-conversion. It warns on implicit conversion from a floating constant to an integer type.

Non-comprehensive list of changes in this release

• For the ARM target, C-language intrinsics are now provided for the full Arm v8.1-M MVE instruction set. <arm_mve.h> supports the complete API defined in the Arm C Language Extensions.

• For the ARM target, C-language intrinsics <arm_cde.h> for the CDE instruction set are now provided.

• clang adds support for a set of extended integer types (_ExtInt(N)) that permit non-power of 2 integers, exposing the LLVM integer types. Since a major motivating use case for these types is to limit ‘bit’ usage, these types don’t automatically promote to ‘int’ when operations are done between two ExtInt(N) types, instead math occurs at the size of the largest ExtInt(N) type.

• Users of UBSan, PGO, and coverage on Windows will now need to add clang’s library resource directory to their library search path. These features all use runtime libraries, and Clang provides these libraries in its resource directory. For example, if LLVM is installed in C:\Program Files\LLVM, then the profile runtime library will appear at C:\Program Files\LLVM\lib\clang\11.0.0\lib\windows\clang_rt.profile-x86_64.lib. To ensure that the linker can find the appropriate library, users should pass /LIBPATH:C:\Program Files\LLVM\lib\clang\11.0.0\lib\windows to the linker. If the user links the program with the clang or clang-cl drivers, the driver will pass this flag for them.

• Clang’s profile files generated through -fprofile-instr-generate are using a fixed hashing algorithm that prevents some collision when loading out-of-date profile informations. Clang can still read old profile files.

• Clang adds support for the following macros that enable the C-intrinsics from the Arm C language extensions for SVE (version 00bet5, see section 2.1 for the list of intrinsics associated to each macro):

  Preprocessor macro	Target feature
  __ARM_FEATURE_SVE	+sve
  __ARM_FEATURE_SVE_BF16	+sve+bf16
  __ARM_FEATURE_SVE_MATMUL_FP32	+sve+f32mm
  __ARM_FEATURE_SVE_MATMUL_FP64	+sve+f64mm
  __ARM_FEATURE_SVE_MATMUL_INT8	+sve+i8mm
  __ARM_FEATURE_SVE2	+sve2
  __ARM_FEATURE_SVE2_AES	+sve2-aes
  __ARM_FEATURE_SVE2_BITPERM	+sve2-bitperm
  __ARM_FEATURE_SVE2_SHA3	+sve2-sha3
  __ARM_FEATURE_SVE2_SM4	+sve2-sm4

  The macros enable users to write C/C++ Vector Length Agnostic (VLA) loops, that can be executed on any CPU that implements the underlying instructions supported by the C intrinsics, independently of the hardware vector register size.

  For example, the __ARM_FEATURE_SVE macro is enabled when targeting AArch64 code generation by setting -march=armv8-a+sve on the command line.

  Example of VLA addition of two arrays with SVE ACLE.

  // Compile with:
  // `clang++ -march=armv8a+sve ...` (for c++)
  // `clang -stc=c11 -march=armv8a+sve ...` (for c)
  #include <arm_sve.h>
  
  void VLA_add_arrays(double *x, double *y, double *out, unsigned N) {
    for (unsigned i = 0; i < N; i += svcntd()) {
      svbool_t Pg = svwhilelt_b64(i, N);
      svfloat64_t vx = svld1(Pg, &x[i]);
      svfloat64_t vy = svld1(Pg, &y[i]);
      svfloat64_t vout = svadd_x(Pg, vx, vy);
      svst1(Pg, &out[i], vout);
    }
  }
  

  Please note that support for lazy binding of SVE function calls is incomplete. When you interface user code with SVE functions that are provided through shared libraries, avoid using lazy binding. If you use lazy binding, the results could be corrupted.

• -O maps to -O1 instead of -O2. (D79916)

• In a -flto={full,thin} link, -Os, -Oz and -Og can be used now. -Os and -Oz map to the -O2 pipe line while -Og maps to the -O1 pipeline. (D79919)

• --coverage (gcov) defaults to gcov [4.8,8) compatible format now.

• On x86, -fpic/-fPIC -fno-semantic-interposition assumes a global definition of default visibility non-interposable and allows interprocedural optimizations. In produced assembly -Lfunc$local local aliases are created for global symbols of default visibility.

New Compiler Flags

• -fstack-clash-protection will provide a protection against the stack clash attack for x86, s390x and ppc64 architectures through automatic probing of each page of allocated stack.

• -ffp-exception-behavior={ignore,maytrap,strict} allows the user to specify the floating-point exception behavior. The default setting is ignore.

• -ffp-model={precise,strict,fast} provides the user an umbrella option to simplify access to the many single purpose floating point options. The default setting is precise.

• The default module cache has moved from /tmp to a per-user cache directory. By default, this is ~/.cache but on some platforms or installations, this might be elsewhere. The -fmodules-cache-path=… flag continues to work.

• -fpch-instantiate-templates tries to instantiate templates already while generating a precompiled header. Such templates do not need to be instantiated every time the precompiled header is used, which saves compile time. This may result in an error during the precompiled header generation if the source header file is not self-contained. This option is enabled by default for clang-cl.

• -fpch-codegen and -fpch-debuginfo generate shared code and/or debuginfo for contents of a precompiled header in a separate object file. This object file needs to be linked in, but its contents do not need to be generated for other objects using the precompiled header. This should usually save compile time. If not using clang-cl, the separate object file needs to be created explicitly from the precompiled header. Example of use:

  $ clang++ -x c++-header header.h -o header.pch -fpch-codegen -fpch-debuginfo
  $ clang++ -c header.pch -o shared.o
  $ clang++ -c source.cpp -o source.o -include-pch header.pch
  $ clang++ -o binary source.o shared.o
  

  • Using -fpch-instantiate-templates when generating the precompiled header usually increases the amount of code/debuginfo that can be shared.

  • In some cases, especially when building with optimizations enabled, using -fpch-codegen may generate so much code in the shared object that compiling it may be a net loss in build time.

  • Since headers may bring in private symbols of other libraries, it may be sometimes necessary to discard unused symbols (such as by adding -Wl,–gc-sections on ELF platforms to the linking command, and possibly adding -fdata-sections -ffunction-sections to the command generating the shared object).

• -fsanitize-coverage-allowlist and -fsanitize-coverage-blocklist are added.

• -mtls-size={12,24,32,48} allows selecting the size of the TLS (thread-local storage) in the local exec TLS model of AArch64, which is the default TLS model for non-PIC objects. Each value represents 4KB, 16MB (default), 4GB, and 256TB (needs -mcmodel=large). This allows large/many thread local variables or a compact/fast code in an executable.

• -menable-experimental-extension` can be used to enable experimental or unratified RISC-V extensions, allowing them to be targeted by specifying the extension name and precise version number in the -march string. For these experimental extensions, there is no expectation of ongoing support - the compiler support will continue to change until the specification is finalised.

Modified Compiler Flags

• -fno-common has been enabled as the default for all targets. Therefore, C code that uses tentative definitions as definitions of a variable in multiple translation units will trigger multiple-definition linker errors. Generally, this occurs when the use of the extern keyword is neglected in the declaration of a variable in a header file. In some cases, no specific translation unit provides a definition of the variable. The previous behavior can be restored by specifying -fcommon.

• -Wasm-ignored-qualifier (ex. asm const (“”)) has been removed and replaced with an error (this matches a recent change in GCC-9).

• -Wasm-file-asm-volatile (ex. asm volatile (“”) at global scope) has been removed and replaced with an error (this matches GCC’s behavior).

• Duplicate qualifiers on asm statements (ex. asm volatile volatile (“”)) no longer produces a warning via -Wduplicate-decl-specifier, but now an error (this matches GCC’s behavior).

• The deprecated argument -f[no-]sanitize-recover has changed to mean -f[no-]sanitize-recover=all instead of -f[no-]sanitize-recover=undefined,integer and is no longer deprecated.

• The argument to -f[no-]sanitize-trap=... is now optional and defaults to all.

• -fno-char8_t now disables the char8_t keyword, not just the use of char8_t as the character type of u8 literals. This restores the Clang 8 behavior that regressed in Clang 9 and 10.

• -print-targets has been added to print the registered targets.

• -mcpu is now supported for RISC-V, and recognises the generic-rv32, rocket-rv32, sifive-e31, generic-rv64, rocket-rv64, and sifive-u54 target CPUs.

• -fwhole-program-vtables (along with -flto*) now prepares all classes for possible whole program visibility if specified during the LTO link. (D71913)

New Pragmas in Clang

• The clang max_tokens_here pragma can be used together with -Wmax-tokens to emit a warning when the number of preprocessor tokens exceeds a limit. Such limits can be helpful in limiting code growth and slow compiles due to large header files.

Attribute Changes in Clang

• Attributes can now be specified by clang plugins. See the Clang Plugins documentation for details.

Windows Support

• Don’t warn about ms_struct may not produce Microsoft-compatible layouts for classes with base classes or virtual functions if the option is enabled globally, as opposed to enabled on a specific class/struct or on a specific section in the source files. This avoids needing to couple -mms-bitfields with -Wno-incompatible-ms-struct if building C++ code.

• Enable -mms-bitfields by default for MinGW targets, matching a similar change in GCC 4.7.

C Language Changes in Clang

• The default C language standard used when -std= is not specified has been upgraded from gnu11 to gnu17.

• Clang now supports the GNU C extension asm inline; it won’t do anything yet, but it will be parsed.

C++ Language Changes in Clang

• Clang now implements a restriction on giving non-C-compatible anonymous structs a typedef name for linkage purposes, as described in C++ committee paper P1766R1 <http://wg21.link/p1766r1>. This paper was adopted by the C++ committee as a Defect Report resolution, so it is applied retroactively to all C++ standard versions. This affects code such as:

  typedef struct {
    int f() { return 0; }
  } S;
  

  Previous versions of Clang rejected some constructs of this form (specifically, where the linkage of the type happened to be computed before the parser reached the typedef name); those cases are still rejected in Clang 11. In addition, cases that previous versions of Clang did not reject now produce an extension warning. This warning can be disabled with the warning flag -Wno-non-c-typedef-for-linkage.

  Affected code should be updated to provide a tag name for the anonymous struct:

  struct S {
    int f() { return 0; }
  };
  

  If the code is shared with a C compilation (for example, if the parts that are not C-compatible are guarded with #ifdef __cplusplus), the typedef declaration should be retained, but a tag name should still be provided:

  typedef struct S {
    int f() { return 0; }
  } S;
  

OpenCL Kernel Language Changes in Clang

• Added extensions from cl_khr_subgroup_extensions to clang and the internal header.

• Added rocm device libs linking for AMDGPU.

• Added diagnostic for OpenCL 2.0 blocks used in function arguments.

• Fixed MS mangling for OpenCL 2.0 pipe type specifier.

• Improved command line options for fast relaxed math.

• Improved atomic_fetch_min/max functions in the internal header (opencl-c.h).

• Improved size of builtin function table for TableGen-based internal header (enabled by -fdeclare-opencl-builtins) and added new functionality for OpenCL 2.0 atomics, pipes, enqueue kernel, cl_khr_subgroups, cl_arm_integer_dot_product.

Changes related to C++ for OpenCL

• Added addrspace_cast operator.

• Improved address space deduction in templates.

• Improved diagnostics of address spaces in nested pointer conversions.

ABI Changes in Clang

• For RISC-V, an ABI bug was fixed when passing complex single-precision floats in RV64 with the hard float ABI. The bug could only be triggered for function calls that exhaust the available FPRs.

OpenMP Support in Clang

New features for OpenMP 5.0 were implemented.

• OpenMP 5.0 is the default version supported by the compiler. User can switch to OpenMP 4.5 using -fopenmp-version=45 option.

• Added support for declare variant directive.

• Improved support of math functions and complex types for NVPTX target.

• Added support for parallel execution of target regions for NVPTX target.

• Added support for scan directives and inscan modifier in reduction clauses.

• Added support for iterator construct.

• Added support for depobj construct.

• Added support for detach clauses in task-based directives.

• Added support for array shaping operations.

• Added support for cancellation constructs in taskloop directives.

• Nonmonotonic modifier is allowed with all schedule kinds.

• Added support for task and default modifiers in reduction clauses.

• Added support for strides in array sections.

• Added support for use_device_addr clause.

• Added support for uses_allocators clause.

• Added support for defaultmap clause.

• Added basic support for hint clause in atomic directives.

• Added basic support for affinity clause.

• Added basic support for ancestor modifier in device clause.

• Added support for default(firstprivate) clause. This clause is the part of upcoming OpenMP 5.1 and can be enabled using -fopenmp-version=51 option.

• Bug fixes and optimizations.

Internal API Changes

These are major API changes that have happened since the 10.0.0 release of Clang. If upgrading an external codebase that uses Clang as a library, this section should help get you past the largest hurdles of upgrading.

• RecursiveASTVisitor no longer calls separate methods to visit specific operator kinds. Previously, RecursiveASTVisitor treated unary, binary, and compound assignment operators as if they were subclasses of the corresponding AST node. For example, the binary operator plus was treated as if it was a BinAdd subclass of the BinaryOperator class: during AST traversal of a BinaryOperator AST node that had a BO_Add opcode, RecursiveASTVisitor was calling the TraverseBinAdd method instead of TraverseBinaryOperator. This feature was contributing a non-trivial amount of complexity to the implementation of RecursiveASTVisitor, it was used only in a minor way in Clang, was not tested, and as a result it was buggy. Furthermore, this feature was creating a non-uniformity in the API. Since this feature was not documented, it was quite difficult to figure out how to use RecursiveASTVisitor to visit operators.

  To update your code to the new uniform API, move the code from separate visitation methods into methods that correspond to the actual AST node and perform case analysis based on the operator opcode as needed:

  • TraverseUnary*() => TraverseUnaryOperator()

  • WalkUpFromUnary*() => WalkUpFromUnaryOperator()

  • VisitUnary*() => VisiUnaryOperator()

  • TraverseBin*() => TraverseBinaryOperator()

  • WalkUpFromBin*() => WalkUpFromBinaryOperator()

  • VisitBin*() => VisiBinaryOperator()

  • TraverseBin*Assign() => TraverseCompoundAssignOperator()

  • WalkUpFromBin*Assign() => WalkUpFromCompoundAssignOperator()

  • VisitBin*Assign() => VisiCompoundAssignOperator()

Build System Changes

These are major changes to the build system that have happened since the 10.0.0 release of Clang. Users of the build system should adjust accordingly.

• clang-tidy and clang-include-fixer are no longer compiled into libclang by default. You can set LIBCLANG_INCLUDE_CLANG_TOOLS_EXTRA=ON to undo that, but it’s expected that that setting will go away eventually. If this is something you need, please reach out to the mailing list to discuss possible ways forward.

clang-format

• Option IndentExternBlock has been added to optionally apply indenting inside extern "C" and extern "C++" blocks.

• IndentExternBlock option accepts AfterExternBlock to use the old behavior, as well as Indent and NoIndent options, which map to true and false, respectively.

  Indent:                       NoIndent:
   #ifdef __cplusplus          #ifdef __cplusplus
   extern "C" {                extern "C++" {
   #endif                      #endif
  
        void f(void);          void f(void);
  
   #ifdef __cplusplus          #ifdef __cplusplus
   }                           }
   #endif                      #endif
  

• Option IndentCaseBlocks has been added to support treating the block following a switch case label as a scope block which gets indented itself. It helps avoid having the closing bracket align with the switch statement’s closing bracket (when IndentCaseLabels is false).

  switch (fool) {                vs.     switch (fool) {
  case 1:                                case 1: {
    {                                      bar();
       bar();                            } break;
    }                                    default: {
    break;                                 plop();
  default:                               }
    {                                    }
      plop();
    }
  }
  

• Option ObjCBreakBeforeNestedBlockParam has been added to optionally apply linebreaks for function arguments declarations before nested blocks.

• Option InsertTrailingCommas can be set to TCS_Wrapped to insert trailing commas in container literals (arrays and objects) that wrap across multiple lines. It is currently only available for JavaScript and disabled by default (TCS_None).

• Option BraceWrapping.BeforeLambdaBody has been added to manage lambda line break inside function parameter call in Allman style.

  true:
  connect(
    []()
    {
      foo();
      bar();
    });
  
  false:
  connect([]() {
      foo();
      bar();
    });
  

• Option AlignConsecutiveBitFields has been added to align bit field declarations across multiple adjacent lines

  true:
    bool aaa  : 1;
    bool a    : 1;
    bool bb   : 1;
  
  false:
    bool aaa : 1;
    bool a : 1;
    bool bb : 1;
  

• Option BraceWrapping.BeforeWhile has been added to allow wrapping before the `while in a do..while loop. By default the value is (false)

  In previous releases IndentBraces implied BraceWrapping.BeforeWhile. If using a Custom BraceWrapping style you may need to now set BraceWrapping.BeforeWhile to (true) to be explicit.

  true:
  do {
    foo();
  }
  while(1);
  
  false:
  do {
    foo();
  } while(1);
  

Static Analyzer

• Improved the analyzer’s understanding of inherited C++ constructors.

• Improved the analyzer’s understanding of dynamic class method dispatching in Objective-C.

• Greatly improved the analyzer’s constraint solver by better understanding when constraints are imposed on multiple symbolic values that are known to be equal or known to be non-equal. It will now also efficiently reject impossible if-branches between known comparison expressions.

• Added on-demand parsing capability to Cross Translation Unit (CTU) analysis.

• Numerous fixes and improvements for the HTML output.

• New checker: alpha.core.C11Lock and alpha.fuchsia.Lock checks for their respective locking APIs.

• New checker: alpha.security.cert.pos.34c finds calls to putenv where a pointer to an autmoatic variable is passed as an argument.

• New checker: webkit.NoUncountedMemberChecker to enforce the existence of virtual destructors for all uncounted types used as base classes.

• New checker: webkit.RefCntblBaseVirtualDtor checks that only ref-counted types are used as class members, not raw pointers and references to uncounted types.

• New checker: alpha.cplusplus.SmartPtr check for dereference of null smart pointers.

• Moved PlacementNewChecker out of alpha, making it enabled by default.

• Added support for multi-dimensional variadic arrays in core.VLASize.

• Added a check for insufficient storage in array placement new calls, as well as support for alignment variants to cplusplus.PlacementNew.

• While still in alpha, alpha.unix.PthreadLock, the iterator and container modeling infrastructure, alpha.unix.Stream, and taint analysis were improved greatly. An ongoing, currently off-by-default improvement was made on the pre-condition modeling of several functions defined in the POSIX standard.

• Improved the warning messages of several checkers.

• Fixed a few remaining cases of checkers emitting reports under incorrect checker names, and employed a few restrictions to more easily identify and avoid such errors.

• Moved several non-reporting checkers (those that model, among other things, standard functions, but emit no diagnostics) to be listed under -analyzer-checker-help-developer instead of -analyzer-checker-help. Manually enabling or disabling checkers found on this list is not supported in production.

• Numerous fixes for crashes, false positives and false negatives in unix.Malloc, osx.cocoa.NSError, and several other checkers.

• Implemented a dockerized testing system to more easily determine the correctness and performance impact of a change to the static analyzer itself. The currently beta-version tool is found in (llvm-project repository)/clang/utils/analyzer/SATest.py.

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 via the mailing list.