Sure, but that's an aliasing issue, not a type punning issue. I agree that the code you cite there is a violation of the aliasing rules, and will not work "as

Sure, but that's an aliasing issue, not a type punning issue. I agree that the code you cite there is a violation of the aliasing rules, and will not work "as expected" on modern compilers unless one does the equivalent of gcc's -fno-strict-aliasing. And I agree that the corrigendum doesn't apply in this case. Once you hand different-typed pointers to the same memory to people, whether via union or just casting pointers, the aliasing rules will up and bite you.