On Wed, 2016-10-12 at 16:22 +0200, Mark Wielaard wrote:
I am no fan of redefining/open coding things which reasonable can be
expected to be provided through normal GNU system headers/libs. But
these seem fairly minor. However if there are larger changes needed we
should probably pull in the relevant gnulib modules.
Hmmm, I remember why I don't like such conditional defines/code.
The configure check was wrong... It should be:
diff --git a/configure.ac b/configure.ac
index feb2ed5..5360447 100644
--- a/configure.ac
+++ b/configure.ac
@@ -245,7 +245,9 @@ zip_LIBS="$LIBS"
LIBS="$save_LIBS"
AC_SUBST([zip_LIBS])
-AC_CHECK_DECLS([memrchr rawmemchr])
+AC_CHECK_DECLS([memrchr, rawmemchr],[],[],
+ [#define _GNU_SOURCE
+ #include <string.h>])
AC_CHECK_DECLS(powerof2, , , [#include <sys/param.h>])
AC_CHECK_LIB([stdc++], [__cxa_demangle], [dnl
At least now we know the fallback code really works, but please
explicitly say what kind of environment you tested on (and which you
didn't) next time.
Thanks,
Mark