C++ assert implementation in assert.h

__assert is part of the implementation; in this case, a function in the library which will be called in case of assertion failure. And the (void) is simply to shut up compiler warnings about the unused results of the || operator.

Look at this line:

extern void __assert (const char *msg, const char *file, int line);

__assert is function that takes an assertion message, a file name and a line number as arguments. Basically, this is the method that prints out the error message and terminates the program when the assertion failed.

Then look at the macro definition above:

#define assert(EX) (void)((EX) || (__assert (#EX, __FILE__, __LINE__),0))

It defines the assert(EX) macro so, it first checks the EX expression and (because of the short-circuit operation of the C++ || operator) only if it fails, it calls the __assert function and passes the failed assertion exception as a string, and the exact location of the assert() method call in your source files. With this preprocessor trickery the assertion library achieves that when you type the following in your program

assert(a == 0);

and your assertion fails during the program run, you get the detailed

Assertion failed: a == 0 at program.c, line 23

error message that helps you to identify the exact location where the assertion was failing in your code.

The (void) part is just for make sure that the compiler won't put up some warnings about the unused result of the (EX) || 0 expression, see the other answers, the guys explained it well.

The remaining preprocessor define NDEBUG is used to turn of assertion generation at all compile-time, you your resulting executable will be smaller and faster.