(svn r10779) -Fix: GCC 4.2 makes (by default for -O2, -O3 and -Os) an assumption that signed integers do not overflow, whereas it is very likely that some of the legacy code does so. It furthermore breaks with the assumed standard (the behaviour is not actually defined in the standard) about the behaviour when signed integers wrap.

config.lib

@@ -838,6 +838,14 @@ make_cflags_and_ldflags() {
 		CFLAGS="$CFLAGS -fno-strict-aliasing"
 	fi
 
+	# GCC 4.2+ automatically assumes that signed overflows do
+	# not occur in signed arithmetics, whereas we are not
+	# sure that they will not happen. It furthermore complains
+	# about it's own optimized code in some places.
+	if [ $cc_version -ge 42 ]; then
+		CFLAGS="$CFLAGS -fno-strict-overflow"
+	fi
+
 	if [ "$os" != "CYGWIN" ] && [ "$os" != "FREEBSD" ] && [ "$os" != "OPENBSD" ] && [ "$os" != "MINGW" ] && [ "$os" != "MORPHOS" ] && [ "$os" != "OSX" ] && [ "$os" != "WINCE" ] && [ "$os" != "PSP" ] && [ "$os" != "OS2" ]; then
 		LIBS="$LIBS -lpthread"
 		LIBS="$LIBS -lrt"