Major New Features¶

• Clang supports use of a profile remapping file, which permits profile data captured for one version of a program to be applied when building another version where symbols have changed (for example, due to renaming a class or namespace). See the Clang Compiler User’s Manual for details.

Non-comprehensive list of changes in this release¶

• The experimental feature Pretokenized Headers (PTH) was removed in its entirely from Clang. The feature did not properly work with about 1/3 of the possible tokens available and was unmaintained.
• The internals of libc++ include directory detection on MacOS have changed. Instead of running a search based on the -resource-dir flag, the search is now based on the path of the compiler in the filesystem. The default behaviour should not change. However, if you override -resource-dir manually and rely on the old behaviour you will need to add appropriate compiler flags for finding the corresponding libc++ include directory.

New Compiler Flags¶

• -fprofile-filter-files=[regexes] and -fprofile-exclude-files=[regexes].

  Clang has now options to filter or exclude some files when instrumenting for gcov-based profiling. See the Clang Compiler User’s Manual for details.

• When using a custom stack alignment, the stackrealign attribute is now implicitly set on the main function.

• …

Deprecated Compiler Flags¶

The following options are deprecated and ignored. They will be removed in future versions of Clang.

• …

Modified Compiler Flags¶

• As of clang 8, alignof and _Alignof return the ABI alignment of a type, as opposed to the preferred alignment. __alignof still returns the preferred alignment. -fclang-abi-compat=7 (and previous) will make alignof and _Alignof return preferred alignment again.

New Pragmas in Clang¶

• Clang now supports adding multiple #pragma clang attribute attributes into a scope of pushed attributes.

Attribute Changes in Clang¶

• …

Windows Support¶

• clang-cl now supports the use of the precompiled header options /Yc and /Yu without the filename argument. When these options are used without the filename, a #pragma hdrstop inside the source marks the end of the precompiled code.
• clang-cl has a new command-line option, /Zc:dllexportInlines-, similar to -fvisibility-inlines-hidden on non-Windows, that makes class-level dllexport and dllimport attributes not apply to inline member functions. This can significantly reduce compile and link times. See the User’s Manual for more info.
• For MinGW, -municode now correctly defines UNICODE during preprocessing.
• For MinGW, clang now produces vtables and RTTI for dllexported classes without key functions. This fixes building Qt in debug mode.
• Allow using Address Sanitizer and Undefined Behaviour Sanitizer on MinGW.
• …

C Language Changes in Clang¶

• …

…

C++ Language Changes in Clang¶

• …

Objective-C Language Changes in Clang¶

…

OpenCL C Language Changes in Clang¶

…

ABI Changes in Clang¶

• _Alignof and alignof now return the ABI alignment of a type, as opposed to the preferred alignment.
  • This is more in keeping with the language of the standards, as well as being compatible with gcc
  • __alignof and __alignof__ still return the preferred alignment of a type
  • This shouldn’t break any ABI except for things that explicitly ask for alignas(alignof(T)).
  • If you have interfaces that break with this change, you may wish to switch to alignas(__alignof(T)), instead of using the -fclang-abi-compat switch.

OpenMP Support in Clang¶

• OpenMP 5.0 features
  • Support relational-op != (not-equal) as one of the canonical forms of random access iterator.
  • Added support for mapping of the lambdas in target regions.
  • Added parsing/sema analysis for the requires directive.
  • Support nested declare target directives.
  • Make the this pointer implicitly mapped as map(this[:1]).
  • Added the close map-type-modifier.
• Various bugfixes and improvements.

New features supported for Cuda devices:

• Added support for the reductions across the teams.
• Extended number of constructs that can be executed in SPMD mode.
• Fixed support for lastprivate/reduction variables in SPMD constructs.
• New collapse clause scheme to avoid expensive remainder operations.
• New default schedule for distribute and parallel constructs.
• Simplified code generation for distribute and parallel in SPMD mode.
• Flag (-fopenmp_optimistic_collapse) for user to limit collapsed loop counter width when safe to do so.
• General performance improvement.

CUDA Support in Clang¶

Internal API Changes¶

These are major API changes that have happened since the 7.0.0 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.

• …

AST Matchers¶

• …

clang-format¶

• …

libclang¶

…

Static Analyzer¶

• …

…

Undefined Behavior Sanitizer (UBSan)¶

• The Implicit Conversion Sanitizer (-fsanitize=implicit-conversion) group was extended. One more type of issues is caught - implicit integer sign change. (-fsanitize=implicit-integer-sign-change). This makes the Implicit Conversion Sanitizer feature-complete, with only missing piece being bitfield handling. While there is a -Wsign-conversion diagnostic group that catches this kind of issues, it is both noisy, and does not catch all the cases.

  bool consume(unsigned int val);
  
  void test(int val) {
    (void)consume(val); // If the value was negative, it is now large positive.
    (void)consume((unsigned int)val); // OK, the conversion is explicit.
  }
  

  Like some other -fsanitize=integer checks, these issues are not undefined behaviour. But they are not always intentional, and are somewhat hard to track down. This group is not enabled by -fsanitize=undefined, but the -fsanitize=implicit-integer-sign-change check is enabled by -fsanitize=integer. (as is -fsanitize=implicit-integer-truncation check)

• The Implicit Conversion Sanitizer (-fsanitize=implicit-conversion) has learned to sanitize compound assignment operators.

• alignment check has learned to sanitize the assume_aligned-like attributes:

  typedef char **__attribute__((align_value(1024))) aligned_char;
  struct ac_struct {
    aligned_char a;
  };
  char **load_from_ac_struct(struct ac_struct *x) {
    return x->a; // <- check that loaded 'a' is aligned
  }
  
  char **passthrough(__attribute__((align_value(1024))) char **x) {
    return x; // <- check the pointer passed as function argument
  }
  
  char **__attribute__((alloc_align(2)))
  alloc_align(int size, unsigned long alignment);
  
  char **caller(int size) {
    return alloc_align(size, 1024); // <- check returned pointer
  }
  
  char **__attribute__((assume_aligned(1024))) get_ptr();
  
  char **caller2() {
    return get_ptr(); // <- check returned pointer
  }
  
  void *caller3(char **x) {
    return __builtin_assume_aligned(x, 1024);  // <- check returned pointer
  }
  
  void *caller4(char **x, unsigned long offset) {
    return __builtin_assume_aligned(x, 1024, offset);  // <- check returned pointer accounting for the offest
  }
  
  void process(char *data, int width) {
      #pragma omp for simd aligned(data : 1024) // <- aligned clause will be checked.
      for (int x = 0; x < width; x++)
      data[x] *= data[x];
  }
  

Python Binding Changes¶

The following methods have been added:

• …