3:28 p.m.
If libraries go in $LDFLAGS while AC_CHECK_LIBbing, they'll end up in
front of the object file name, which rarely works well. They belong
in $LIBS.
---
configure.in | 25 ++++++++++++-------------
1 files changed, 12 insertions(+), 13 deletions(-)
diff --git a/configure.in b/configure.in
index f735bba..5308364 100644
--- a/configure.in
+++ b/configure.in
@@ -542,14 +542,14 @@ AC_SUBST([LIBXML_LIBS])
dnl xmlURI structure has query_raw?
old_cflags="$CFLAGS"
-old_ldflags="$LDFLAGS"
+old_libs="$LIBS"
CFLAGS="$CFLAGS $LIBXML_CFLAGS"
-LDFLAGS="$LDFLAGS $LIBXML_LIBS"
+LIBS="$LIBS $LIBXML_LIBS"
AC_CHECK_MEMBER([struct _xmlURI.query_raw],
[AC_DEFINE([HAVE_XMLURI_QUERY_RAW], [], [Have query_raw field in libxml2 xmlURI
structure])],,
[#include <libxml/uri.h>])
CFLAGS="$old_cflags"
-LDFLAGS="$old_ldflags"
+LIBS="$old_libs"
dnl GnuTLS library
GNUTLS_CFLAGS=
@@ -579,15 +579,15 @@ dnl Old versions of GnuTLS uses types like 'gnutls_session'
instead
dnl of 'gnutls_session_t'. Try to detect this type if defined so
dnl that we can offer backwards compatibility.
old_cflags="$CFLAGS"
-old_ldflags="$LDFLAGS"
+old_libs="$LIBS"
CFLAGS="$CFLAGS $GNUTLS_CFLAGS"
-LDFLAGS="$LDFLAGS $GNUTLS_LIBS"
+LIBS="$LIBS $GNUTLS_LIBS"
AC_CHECK_TYPE([gnutls_session],
AC_DEFINE([GNUTLS_1_0_COMPAT],[],
[enable GnuTLS 1.0 compatibility macros]),,
[#include <gnutls/gnutls.h>])
CFLAGS="$old_cflags"
-LDFLAGS="$old_ldflags"
+LIBS="$old_libs"
dnl Cyrus SASL
@@ -685,12 +685,12 @@ if test "x$with_polkit" = "xyes" -o
"x$with_polkit" = "xcheck"; then
[use PolicyKit for UNIX socket access checks])
old_CFLAGS=$CFLAGS
- old_LDFLAGS=$LDFLAGS
+ old_LIBS=$LIBS
CFLAGS="$CFLAGS $POLKIT_CFLAGS"
- LDFLAGS="$LDFLAGS $POLKIT_LIBS"
+ LIBS="$LIBS $POLKIT_LIBS"
AC_CHECK_FUNCS([polkit_context_is_caller_authorized])
CFLAGS="$old_CFLAGS"
- LDFLAGS="$old_LDFLAGS"
+ LIBS="$old_LIBS"
AC_PATH_PROG([POLKIT_AUTH], [polkit-auth])
if test "x$POLKIT_AUTH" != "x"; then
@@ -1682,20 +1682,19 @@ if test "x$with_hal" = "xyes" -o
"x$with_hal" = "xcheck"; then
[use HAL for host device enumeration])
old_CFLAGS=$CFLAGS
- old_LDFLAGS=$LDFLAGS
+ old_LIBS=$LIBS
CFLAGS="$CFLAGS $HAL_CFLAGS"
- LDFLAGS="$LDFLAGS $HAL_LIBS"
+ LIBS="$LIBS $HAL_LIBS"
AC_CHECK_FUNCS([libhal_get_all_devices],,[with_hal=no])
AC_CHECK_FUNCS([dbus_watch_get_unix_fd])
CFLAGS="$old_CFLAGS"
- LDFLAGS="$old_LDFLAGS"
+ LIBS="$old_LIBS"
fi
fi
AM_CONDITIONAL([HAVE_HAL], [test "x$with_hal" = "xyes"])
AC_SUBST([HAL_CFLAGS])
AC_SUBST([HAL_LIBS])
-
dnl udev/libpciaccess library check for alternate host device enumeration
UDEV_CFLAGS=
UDEV_LIBS=
--
1.6.5.3.100.g75959
5:24 a.m.
On Mon, Nov 23, 2009 at 09:28:24PM +0000, Nix wrote:
If libraries go in $LDFLAGS while AC_CHECK_LIBbing, they'll end
up in
front of the object file name, which rarely works well. They belong
in $LIBS.
What platforms / scenarios show this problem ? Empirically it has
been working fine on our commonly tested platforms, so it'll be
important to know where to look/test for this problem.
---
configure.in | 25 ++++++++++++-------------
1 files changed, 12 insertions(+), 13 deletions(-)
diff --git a/configure.in b/configure.in
index f735bba..5308364 100644
--- a/configure.in
+++ b/configure.in
@@ -542,14 +542,14 @@ AC_SUBST([LIBXML_LIBS])
dnl xmlURI structure has query_raw?
old_cflags="$CFLAGS"
-old_ldflags="$LDFLAGS"
+old_libs="$LIBS"
CFLAGS="$CFLAGS $LIBXML_CFLAGS"
-LDFLAGS="$LDFLAGS $LIBXML_LIBS"
+LIBS="$LIBS $LIBXML_LIBS"
AC_CHECK_MEMBER([struct _xmlURI.query_raw],
[AC_DEFINE([HAVE_XMLURI_QUERY_RAW], [], [Have query_raw field in libxml2 xmlURI
structure])],,
[#include <libxml/uri.h>])
CFLAGS="$old_cflags"
-LDFLAGS="$old_ldflags"
+LIBS="$old_libs"
dnl GnuTLS library
GNUTLS_CFLAGS=
@@ -579,15 +579,15 @@ dnl Old versions of GnuTLS uses types like 'gnutls_session'
instead
dnl of 'gnutls_session_t'. Try to detect this type if defined so
dnl that we can offer backwards compatibility.
old_cflags="$CFLAGS"
-old_ldflags="$LDFLAGS"
+old_libs="$LIBS"
CFLAGS="$CFLAGS $GNUTLS_CFLAGS"
-LDFLAGS="$LDFLAGS $GNUTLS_LIBS"
+LIBS="$LIBS $GNUTLS_LIBS"
AC_CHECK_TYPE([gnutls_session],
AC_DEFINE([GNUTLS_1_0_COMPAT],[],
[enable GnuTLS 1.0 compatibility macros]),,
[#include <gnutls/gnutls.h>])
CFLAGS="$old_cflags"
-LDFLAGS="$old_ldflags"
+LIBS="$old_libs"
dnl Cyrus SASL
@@ -685,12 +685,12 @@ if test "x$with_polkit" = "xyes" -o
"x$with_polkit" = "xcheck"; then
[use PolicyKit for UNIX socket access checks])
old_CFLAGS=$CFLAGS
- old_LDFLAGS=$LDFLAGS
+ old_LIBS=$LIBS
CFLAGS="$CFLAGS $POLKIT_CFLAGS"
- LDFLAGS="$LDFLAGS $POLKIT_LIBS"
+ LIBS="$LIBS $POLKIT_LIBS"
AC_CHECK_FUNCS([polkit_context_is_caller_authorized])
CFLAGS="$old_CFLAGS"
- LDFLAGS="$old_LDFLAGS"
+ LIBS="$old_LIBS"
AC_PATH_PROG([POLKIT_AUTH], [polkit-auth])
if test "x$POLKIT_AUTH" != "x"; then
@@ -1682,20 +1682,19 @@ if test "x$with_hal" = "xyes" -o
"x$with_hal" = "xcheck"; then
[use HAL for host device enumeration])
old_CFLAGS=$CFLAGS
- old_LDFLAGS=$LDFLAGS
+ old_LIBS=$LIBS
CFLAGS="$CFLAGS $HAL_CFLAGS"
- LDFLAGS="$LDFLAGS $HAL_LIBS"
+ LIBS="$LIBS $HAL_LIBS"
AC_CHECK_FUNCS([libhal_get_all_devices],,[with_hal=no])
AC_CHECK_FUNCS([dbus_watch_get_unix_fd])
CFLAGS="$old_CFLAGS"
- LDFLAGS="$old_LDFLAGS"
+ LIBS="$old_LIBS"
fi
fi
AM_CONDITIONAL([HAVE_HAL], [test "x$with_hal" = "xyes"])
AC_SUBST([HAL_CFLAGS])
AC_SUBST([HAL_LIBS])
-
dnl udev/libpciaccess library check for alternate host device enumeration
UDEV_CFLAGS=
UDEV_LIBS=
ACK
Daniel
--
|: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :|
|: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :|
|: http://autobuild.org -o- http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|
4:26 p.m.
On 25 Nov 2009, Daniel P. Berrange spake thusly:
On Mon, Nov 23, 2009 at 09:28:24PM +0000, Nix wrote:
> If libraries go in $LDFLAGS while AC_CHECK_LIBbing, they'll end up in
> front of the object file name, which rarely works well. They belong
> in $LIBS.
What platforms / scenarios show this problem ? Empirically it has
been working fine on our commonly tested platforms, so it'll be
important to know where to look/test for this problem.
Autoconf has always sorted $LDFLAGS before object files. When
-Wl,--as-needed is turned on by default (as a number of distros do),
such libraries are silently dropped because nothing needs symbols from
them at that point in the link, just as if they were static
libraries. Without -Wl,--as-needed, every single library named in the
link is added to DT_NEEDED, so this works.
Nonetheless, it is plainly obvious to any Unix hand that
cc -L/usr/lib -lfoo -o test thing-that-uses-foo.o
is less likely to work than
cc -L/usr/lib -o test thing-that-uses-foo.o -lfoo
and if you put libraries in LDFLAGS, the former's often what you get.
It'll probably fail for you, too, if HAL (say) is statically linked,
because the -Wl,--as-needed behaviour for shared libraries is exactly
how static libraries have always worked.
The GNU Autoconf manual has long stated
-- Variable: LDFLAGS
[...]
Don't use this variable to pass library names (`-l') to the
linker; use `LIBS' instead.
(I'm surprised that no Gentooers, for instance, have reported this
already. A *lot* of distros turn on -Wl,--as-needed by default these
days, often by modifying the built-in GCC specs and providing a fallback
specs file that specifies -Wl,--no-as-needed for those very rare things,
such as glibc's testsuite, that actually *require* linkage with shared
libraries from which no symbols are required.)
5477
days inactive
5479
days old
2 comments
2 participants
participants (2)
-
Daniel P. Berrange -
Nix