On 2025-03-12 13:29, Collin Funk wrote: > This is reported in Gnulib here [1]. > > It is not a bug, in the Gnulib documentation there is a warning about > this [2]:
Cough, cough. I cannot find anywhere in the GCC or Clang documentation that displacement of null pointers is allowed (i.e. that it constitutes a documented extension). Therefore, an expression like ((char *) 0) + 1 means "invoke undefined behavior here", allowing an optimizer to assume that the code is unreachable, with whatever consequences that further entails. The offsetof macro has widely been implemented using null pointer arithmetic. Compilers which come with <stddef.h> whose offsetof is implemented in that traditional way are de facto defining null pointer arithmetic as a documented extension. This is because the include file they provide is human readable, and is processed in the ordinary way like any other file: whatever the file depends on must be defined for the sake of that file. Note, though, it looks like Clang does not use the traditional trick for defining offsetof; it uses __builtin_offsetof: https://clang.llvm.org/doxygen/____stddef__offsetof_8h_source.html Therefore in the case of Clang, we cannot look to its offsetof definition to hypothesize that it has a null pointer arithmetic extension.
