C and C++: differences, compatibility etc.

Two Camps

Bjarne Stroustrup, the creator of C++, has suggested that the incompatibilities between C and C++ should be reduced as much as possible in order to maximize interoperability between the two languages. Others have argued that since C and C++ are two different languages, compatibility between them is useful but not vital; according to this camp, efforts to reduce incompatibility should not hinder attempts to improve each language in isolation. The official rationale for the 1999 C standard (C99) endorse[d] the principle of maintaining the largest common subset between C and C++ while maintaining a distinction between them and allowing them to evolve separately, and stated that the authors were content to let C++ be the big and ambitious language.

Constructs valid in C but not in C++

C++ enforces stricter typing rules (no implicit violations of the static type system), and initialization requirements (compile-time enforcement that in-scope variables do not have initialization subverted) than C, and so some valid C code is invalid in C++.

One commonly encountered difference is C being more weakly-typed regarding pointers. Specifically, C allows a void* pointer to be assigned to any pointer type without a cast, while C++ does not; this idiom appears often in C code using malloc memory allocation, or in the passing of context pointers to the POSIX pthreads API, and other frameworks involving callbacks. For example, the following is valid in C but not C++:

void *ptr;
/* Implicit conversion from void* to int* */
int *i = ptr;

or similarly:

int *j = malloc(5 * sizeof *j);
  /* Implicit conversion from void* to int* */

In order to make the code compile as both C and C++, one must use an explicit cast, as follows (with some caveats in both languages:

void *ptr;
int *i = (int *) ptr;
int *j = (int *) malloc(5 * sizeof *j);

• C++ has more complicated rules about pointer assignments that add qualifiers as it allows the assignment of int ** to const int *const * but not the unsafe assignment to const int ** while C allows neither of those (although compilers will usually only emit a warning).
• C++ changes some C standard library functions to add additional overloaded functions with const type qualifiers, e.g. strchr returns char* in C, while C++ acts as if there were two overloaded functions const char *strchr(const char *) and a char *strchr(char *). In C23 generic selection will be used to make C's behaviour more similar to C++'s.
• C++ is also more strict in conversions to enums: ints cannot be implicitly converted to enums as in C. Also, enumeration constants (enum enumerators) are always of type int in C, whereas they are distinct types in C++ and may have a size different from that of int.
• In C++ a const variable must be initialized; in C this is not necessary.

• C++ compilers prohibit goto or switch from crossing an initialization, as in the following C99 code:

  void fn(void)
  {
      goto flack;
      int i = 1;
  flack:
      ;
  }

• While syntactically valid, a longjmp() results in undefined behaviour in C++ if the jumped-over stack frames include objects with nontrivial destructors. The C++ implementation is free to define the behaviour such that destructors would be called. However, this would preclude some uses of longjmp() which would otherwise be valid, such as implementation of threads or coroutines switching between separate call stacks with longjmp() — when jumping from the lower to the upper call stack in global address space, destructors would be called for every object in the lower call stack. No such issue exists in C.

• C allows for multiple tentative definitions of a single global variable in a single translation unit, which is invalid as an ODR violation in C++.

  int N;
  int N = 10;

• In C, declaring a new type with the same name as an existing struct, union or enum is valid, but it is invalid in C++, because in C, struct, union, and enum types must be indicated as such whenever the type is referenced whereas in C++, all declarations of such types carry the typedef implicitly.

  enum BOOL {FALSE, TRUE};
  typedef int BOOL;

• Non-prototype ("K&R"-style) function declarations are invalid in C++; they are still valid in C until C23, although they have been deemed obsolescent since C's original standardization in 1990. (The term obsolescent is a defined term in the ISO C standard, meaning a feature that may be considered for withdrawal in future revisions of the standard.) Similarly, implicit function declarations (using functions that have not been declared) are not allowed in C++, and have been invalid in C since 1999.
• In C until C23, a function declaration without parameters, e.g. int foo();, implies that the parameters are unspecified. Therefore, it is legal to call such a function with one or more arguments, e.g. foo(42, "hello world"). In contrast, in C++ a function prototype without arguments means that the function takes no arguments, and calling such a function with arguments is ill-formed. In C, the correct way to declare a function that takes no arguments is by using void, as in int foo(void);, which is also valid in C++. Empty function prototypes are a deprecated feature in C99 (as they were in C89).
• In both C and C++, one can define nested struct types, but the scope is interpreted differently: in C++, a nested struct is defined only within the scope/namespace of the outer struct, whereas in C the inner struct is also defined outside the outer struct.
• C allows struct, union, and enum types to be declared in function prototypes, whereas C++ does not.

C99 and C11 added several additional features to C that have not been incorporated into standard C++ as of C++20, such as complex numbers, variable length arrays (complex numbers and variable length arrays are designated as optional extensions in C11), flexible array members, the restrict keyword, array parameter qualifiers, and compound literals.

• Complex arithmetic using the float complex and double complex primitive data types was added in the C99 standard, via the _Complex keyword and complex convenience macro. In C++, complex arithmetic can be performed using the complex number class, but the two methods are not code-compatible. (The standards since C++11 require binary compatibility, however.)

• Variable length arrays. This feature leads to possibly non-compile time sizeof operator.

  void foo(size_t x, int a[*]);  // VLA declaration
  void foo(size_t x, int a[x])
  {
      printf("%zu\n", sizeof a); // same as sizeof(int*)
      char s[x * 2];
      printf("%zu\n", sizeof s); // will print x*2
  }

• The last member of a C99 structure type with more than one member may be a flexible array member, which takes the syntactic form of an array with unspecified length. This serves a purpose similar to variable-length arrays, but VLAs cannot appear in type definitions, and unlike VLAs, flexible array members have no defined size. ISO C++ has no such feature. Example:

  struct X
  {
      int n, m;
      char bytes[];
  }

• The restrict type qualifier defined in C99 was not included in the C++03 standard, but most mainstream compilers such as the GNU Compiler Collection, Microsoft Visual C++, and Intel C++ Compiler provide similar functionality as an extension.

• Array parameter qualifiers in functions are supported in C but not C++.

  int foo(int a[const]);     // equivalent to int *const a
  int bar(char s[static 5]); // annotates that s is at least 5 chars long
  

• The functionality of compound literals in C is generalized to both built-in and user-defined types by the list initialization syntax of C++11, although with some syntactic and semantic differences.

  struct X a = (struct X){4, 6};  // The equivalent in C++ would be X{4, 6}. The C syntactic form used in C99 is supported as an extension in the GCC and Clang C++ compilers.
  foo(&(struct X){4, 6});         // The object is allocated in the stack and its address can be passed to a function. This is not supported in C++.
  
  if (memcmp(d, (int []){8, 6, 7, 5, 3, 0, 9}, n) == 0) {} // The equivalent in C++ would be using digits = int []; if (memcmp(d, digits{8, 6, 7, 5, 3, 0, 9}, n) == 0) {}

• Designated initializers for arrays are valid only in C:

  char s[20] = { [0] = 'a', [8] = 'g' };  // allowed in C, not in C++

• Functions that do not return can be annotated using a noreturn attribute in C++ whereas C uses a distinct keyword. In C23, the attribute syntax is also supported.

C++ adds numerous additional keywords to support its new features. This renders C code using those keywords for identifiers invalid in C++. For example:

struct template
{
    int new;
    struct template* class;
};

is valid C code, but is rejected by a C++ compiler, since the keywords template, new and class are reserved.

Constructs that behave differently in C and C++

There are a few syntactic constructs that are valid in both C and C++ but produce different results in the two languages.

• Character literals such as 'a' are of type int in C and of type char in C++, which means that sizeof 'a' will generally give different results in the two languages: in C++, it will be 1, while in C it will be sizeof(int). As another consequence of this type difference, in C, 'a' will always be a signed expression, regardless of whether or not char is a signed or unsigned type, whereas for C++ this is compiler implementation specific.
• C++ assigns internal linkage to namespace-scoped const variables unless they are explicitly declared extern, unlike C in which extern is the default for all file-scoped entities. In practice this does not lead to silent semantic changes between identical C and C++ code but instead will lead to a compile-time or linkage error.
• In C, use of inline functions requires manually adding a prototype declaration of the function using the extern keyword in exactly one translation unit to ensure a non-inlined version is linked in, whereas C++ handles this automatically. In more detail, C distinguishes two kinds of definitions of inline functions: ordinary external definitions (where extern is explicitly used) and inline definitions. C++, on the other hand, provides only inline definitions for inline functions. In C, an inline definition is similar to an internal (i.e. static) one, in that it can coexist in the same program with one external definition and any number of internal and inline definitions of the same function in other translation units, all of which can differ. This is a separate consideration from the linkage of the function, but not an independent one. C compilers are afforded the discretion to choose between using inline and external definitions of the same function when both are visible. C++, however, requires that if a function with external linkage is declared inline in any translation unit then it must be so declared (and therefore also defined) in every translation unit where it is used, and that all the definitions of that function be identical, following the ODR. Static inline functions behave identically in C and C++.
• Both C99 and C++ have a Boolean type bool with constants true and false, but they are defined differently. In C++, bool is a built-in type and a reserved keyword. In C99, a new keyword, _Bool, is introduced as the new Boolean type. The header stdbool.h provides macros bool, true and false that are defined as _Bool, 1 and 0, respectively. Therefore, true and false have type int in C. This is likely to change in C23 however, whose draft includes changing bool, true, and false to become keywords, and giving true and false the type bool.
• In C it is implementation-defined whether a bit field of type int is signed or unsigned while in C++ it is always signed to match the underlying type.

Linking C and C++ code

While C and C++ maintain a large degree of source compatibility, the object files their respective compilers produce can have important differences that manifest themselves when intermixing C and C++ code. Notably:

• C compilers do not name mangle symbols in the way that C++ compilers do.
• Depending on the compiler and architecture, it also may be the case that calling conventions differ between the two languages.

For these reasons, for C++ code to call a C function foo(), the C++ code must prototype foo() with extern "C". Likewise, for C code to call a C++ function bar(), the C++ code for bar() must be declared with extern "C".

A common practice for header files to maintain both C and C++ compatibility is to make its declaration be extern "C" for the scope of the header:

/* Header file foo.h */
#ifdef __cplusplus /* If this is a C++ compiler, use C linkage */
extern "C" {
#endif

/* These functions get C linkage */
void foo();

struct bar { /* ... */ };

#ifdef __cplusplus /* If this is a C++ compiler, end C linkage */
}
#endif

Differences between C and C++ linkage and calling conventions can also have subtle implications for code that uses function pointers. Some compilers will produce non-working code if a function pointer declared extern "C" points to a C++ function that is not declared extern "C".

Take the following code:

void my_function();
extern "C" void foo(void (*fn_ptr)(void));

void bar()
{
   foo(my_function);
}

This may raise a warning, as with Sun Microsystems' C++ compiler, because my_function() is not declared with C linkage and calling conventions, but is being passed to the C function foo().