Use of .sa_sigaction and .sa_handler in same project(man page clarification)

Question

While looking up the man pages for sigaction, I stumbled upon a confusing note in the Linux man pages:

On some architectures a union is involved: do not assign to both sa_handler and sa_sigaction(sigaction man page).

There are two possible ways to read it:

1. Does this mean, as I understand it, that you are not allowed to set .sa_sigaction and .sa_handler in the same struct sigaction.
2. Or is one only allowed to use either .sa_sigaction OR .sa_handler in the whole project?

struct sigaction sigactionArray[SIGNAL_SIGACTION_ARRAY_SIZE] = {
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_IGN, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_handler = SIG_DFL, .sa_flags = SA_NODEFER},
   {.sa_sigaction = HandleBacktraceSignals, .sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK}
};

According to interpretation 1. the above piece of code would be fine, but when definition 2. is right, it would be invalid since I mix.

Which one is true? Note: On my architecture, it is defined as a union

Answer 1

It's just talking about the same structure. It explains why: they might be part of a union, so they use the same memory in the structure. E.g. it might actually be like:

struct sigaction {
    union {
        void     (*sa_handler)(int);
        void     (*sa_sigaction)(int, siginfo_t *, void *);
    },
    sigset_t   sa_mask;
    int        sa_flags;
    void     (*sa_restorer)(void);
};

Therefore, writing to one of them overwrites the other.