Hi Paul, > > I can reproduce it on an Ubuntu 16.04.x machine: > > > m4.c:166:1: error: function might be candidate for attribute 'pure' if it > > is known to return normally [-Werror=suggest-attribute=pure] > > fault_handler (int signo) > > ^ > > ... > > > It is not clear to me whether the function 'fault_handler' should be > > marked with attribute 'pure' or not. > > 'pure' does sound dicey, as the intent is for 'pure' to be useful to a > compiler > when doing optimization and it's satisfying that intent, but on the other > hand > the function does have side effects visible to the caller.
I was hesitating as well, until I read the first sentence in the description of 'pure': "Calls to functions that have no observable effects on the state of the program other than to return a value..." Evidently _exit(1) is a side effect (the biggest side effect a function can have). So, 'fault_handler' should not be marked 'pure'. > Anyway, GCC 10.1 doesn't issue the diagnostic for me, I hope because GCC has > been improved since the GCC 5 that comes with Ubuntu 16.04. Right, it's corrected in GCC 8 or newer. > So I hope we can > leave this alone, as we can assume that whoever prepares m4 distributions has > a > more-modern GCC. OK. I've switched the weekly tarball generator [2] to use Ubuntu 20.04; it contains a GCC >= 8. Bruno [1] https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Common-Function-Attributes.html [2] https://gitlab.com/gnu-m4/ci-distcheck
