C's offsetof() macro is an ANSI C library feature found in stddef.h. It evaluates to the offset (in bytes) of a given member within a struct or union type, an expression of type size_t. The offsetof() macro takes two parameters, the first being a structure name, and the second being the name of a member within the structure. It cannot be described as a C prototype.
Contents
Implementation
The "traditional" implementation of the macro relied on the compiler being not especially picky about pointers; it obtained the offset of a member by specifying a hypothetical structure that begins at address zero:
This works by casting a null pointer into a pointer to structure st, and then obtaining the address of member m within said structure. While this implementation works correctly in many compilers, it has undefined behavior according to the C standard, since it involves a dereference of a null pointer (although, one might argue that no dereferencing takes place, because the whole expression is calculated at compile time). It also tends to produce confusing compiler diagnostics if one of the arguments is misspelled. Some modern compilers (such as GCC) define the macro using a special form instead, e.g.
This builtin is especially useful with C++ classes or structs that declare a custom unary operator &.
Usage
It is useful when implementing generic data structures in C. For example, the Linux kernel uses offsetof() to implement container_of(), which allows something like a mixin type to find the structure that contains it:
This macro is used to retrieve an enclosing structure from a pointer to a nested element, such as this iteration of a linked list of my_struct objects:
The linux kernel implementation of container_of uses a GNU C extension called statement expressions. It's possible a statement expression was used to ensure type safety and therefore eliminate potential accidental bugs. There is, however, a way to implement the same behaviour without using statement expressions while still ensuring type safety:
At first glance, this implementation may seem more complex than necessary, and the unusual use of the conditional operator may seem out of place. A simpler implementation is possible:
This implementation would also serve the same purpose, however, there's a fundamental omission in terms of the original linux kernel implementation. The type of ptr is never checked against the type of the member, this is something that the linux kernel implementation would catch.
In the aforementioned type-checked implementation, the check is performed by the unusual use of the conditional operator. The constraints of the conditional operator specify that if the operands to the conditional operator are both pointers to a type, they must both be pointers to compatible types. In this case, despite the fact that the value of the third operand of the conditional expression will never be used, the compiler must perform a check to ensure that (ptr) and &((type *)0)->member are both compatible pointer types.
Limitations
Usage of offsetof is limited to POD types in C++98, and standard-layout classes in more recent versions (such as C++11), otherwise it has an undefined behavior. While most compilers will generate a correct result even in cases that don't respect the standard, there are edge cases when offsetof will either yield an incorrect value, generate a compile-time warning or error, or outright crash the program. This is especially the case for virtual inheritance. The following program will generate several warnings and print obviously suspicious results when compiled with gcc 4.7.3 on an amd64 architecture:
Output is:
offsetof(A,a) : 8 offsetof(B,b) : 8