On 12/04/2013 10:38 AM, Rhys Ulerich wrote: >> Reviving this thread; I've fixed the code duplication, and now have only >> one spot to work on to fix Rhys' original complaint. > > Thanks Eric. Any tentative fix that I should ask my user to try out > on that BG system? I'm unsure what to try in light of the reduced > duplication.
Yes, can you please try with latest gnulib.git plus this patch:
(I still wish there were a more reliable way to detect whether a
compiler supports -C alongside -E; I'm a bit worried about just probing
for whether it works, without knowing what behavior to expect from
compilers that don't support it, or worse support it but with a
different meaning than what gcc gives it)
diff --git a/m4/absolute-header.m4 b/m4/absolute-header.m4
index 89ff5be..ccd3c6c 100644
--- a/m4/absolute-header.m4
+++ b/m4/absolute-header.m4
@@ -1,4 +1,4 @@
-# absolute-header.m4 serial 16
+# absolute-header.m4 serial 17
dnl Copyright (C) 2006-2013 Free Software Foundation, Inc.
dnl This file is free software; the Free Software Foundation
dnl gives unlimited permission to copy and/or distribute it,
@@ -61,9 +61,11 @@ AC_DEFUN([gl_ABSOLUTE_HEADER_ONE],
dnl detect the absolute name of <dirent.h>, <signal.h>, <poll.h>
dnl and others. The workaround is to force preservation of comments
dnl through option -C. This ensures all necessary #line directives
- dnl are present. GCC supports option -C as well.
- case "$host_os" in
- aix*) gl_absname_cpp="$ac_cpp -C" ;;
+ dnl are present. GCC supports option -C as well. Various Blue Gene
+ dnl compilers for Intrepid also support -C, needed for detecting
+ dnl when compiling for that system.
+ case $host_os:$CC in
+ aix*:* | *:bg* | *:blrts_*) gl_absname_cpp="$ac_cpp -C" ;;
*) gl_absname_cpp="$ac_cpp" ;;
esac
changequote(,)
--
1.8.3.1
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
