http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47409
--- Comment #3 from John Regehr <regehr at cs dot utah.edu> 2011-01-24 16:43:58 UTC --- (In reply to comment #2) > Not sure if this is a bug at all, structure assignment should be implementable > using memcpy or memmove and thus the side effects that will happen on it are > not very well defined. Hi Jakub- "volatile" isn't a very strong guarantee, but realistically people assume that if a volatile is on the RHS of an assignment, a load from that location occurs (and a store, if on the LHS). My guess is that violating this contract will confuse embedded systems developers and break previously working code. Also, if the rule for volatile becomes significantly different from what I said above, we won't be able to do volatile testing anymore since violations will be sort of meaningless.
