Eric Blake wrote: > Ingo Weinhold <ingo_weinhold <at> gmx.de> writes: > > > > If SA_ONSTACK exists, use it even if SA_SIGINFO doesn't. This does at least > > allow for a somewhat controlled program termination on stack overflow. E.g. > > in Haiku this prevents the debug server dialog box from popping up. > > In my mind, it would be cleaner if we provided a stub > > #ifndef SA_SIGINFO > # define SA_SIGINFO 0 > #endif > > in signal.in.h rather than using #if inside the body of a method, similar to > the SA_NODEFER stub added for Interix. Does that approach work for you?
Then both SA_SIGINFO and SA_ONSTACK were defined and the SIGACTION_WORKS macro in c-stack.c would be defined to 1. Which would obviously go horribly wrong in the signal handler. If both SA_SIGINFO and SA_ONSTACK were stubbed, SIGACTION_WORKS could be defined differently: #define SIGACTION_WORKS (SA_SIGINFO && SA_ONSTACK) All "#if defined SA_SIGINFO" and "#if defined SA_ONSTACK" checks would need to be changed, too. Not sure if that's such a good idea. Stubbing only SA_ONSTACK as Paolo suggests should be relatively harmless, I suppose. CU, Ingo
