Post History

Whether null pointer checking should be the caller's responsibility or the callee is debatable and probably a matter of opinion, local convention, sometimes logical choices but in all cases should be documented explicitly.

Consider for example `memcpy()`: it is the caller's responsibility, the behavior is undefined if either pointer is invalid unless the size is `0` and if the destination and source ranges overlap. There are good reasons for this: `memcpy` is meant to be as efficient as possible so not extra tests are needed. This allows compilers to generate optimal inline code for `memcpy` calls, sometimes as small as a single instruction.

Conversely, consider `free()`: this is function specifically accepts a null pointer argument, so no test is required at the call site to free any allocated data, whether it was allocated successfully or not as long as the pointers were initialized to `NULL`.

The original rationale of the C language was to favor performance at the cost of increased programmer's responsibility. Note that it is always the programmer's responsibility to pass valid pointers along with enough information for the callee to access only relevant data via the pointer. Small functions can be more efficient without null checking, larger ones can have extra checks at minimal cost (quite negligible on CPUs with branch prediction). Handling null pointers gracefully is more a question of quality of implementation in this case (eg: `printf("%s", (char *)NULL);`).

Another sound approach is too add assertions in the function prologue to perform null checking only in DEBUG builds. Such functions should be documented as taking non null pointers and purposely adorned in the source code if annotations are available to specify this restriction (via `__attribute__((__nonnull__))`, `static` size specifications, or similar annotations)