Re: [RFC] sigaltstack + longjmp to protect from C stack overflow

[Paul Eggert]

Thanks for looking into this.  A few comments:

'handle_sigsegv' tries to guess direction of stack growth, but shouldn't 
stack growth direction be determined at startup?  That should be more 
reliable than guessing from what may be a stray pointer reference.

init_signals need not use xmalloc to allocate the alternate stack, since 
it's of fixed size and Emacs always needs it.  We should be able to 
allocate the stack statically.

sys/resource.h is not available everywhere; include it only if 
HAVE_SYS_RESOURCE_H is defined.  Similarly, use struct rlimit and 
getrlimit only if HAVE_GETRLIMIT is defined, and use sigaltstack only if 
HAVE_SIGALTSTACK is defined (you'll need to modify configure.ac for 
that).  And the code should do the right thing (i.e., stick with 
thread_fatal_action instead of sigsegv_action) if SA_SIGINFO and/or 
SA_ONSTACK are not defined.  Come to think of it, check for all the 
above issues and do the sigsegv_action stuff only if all the macros 
check out.

There's no need to abort if sigaltstack fails.  Just let Emacs go on as 
it does now, and use thread_fatal_action.