https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108439
--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> --- Aliasing rules are NOT a local thing and they are not about what the type of the pointer is, rather than they are about the access after the casting. They are also global ... GCC used to have strict aliasing warnings which might warn during optimizations but it had many false positives so it was tuned down a lot. The rules are simple in most cases (there are complex parts to it when dealing with unions and structures but more complex in the implementation of the compiler). Some reading on this subject: https://developers.redhat.com/blog/2020/06/02/the-joys-and-perils-of-c-and-c-aliasing-part-1 https://gist.github.com/shafik/848ae25ee209f698763cffee272a58f8 There are many more papers about C/C++ aliasing rules if you want to go read up on why GCC is doing the "correct" thing here.
