[SSSD] [PATCHES] Improve portability of autoconf and automake

Jakub Hrozek jhrozek at redhat.com
Mon Sep 9 14:21:50 UTC 2013


On Mon, Sep 09, 2013 at 04:11:01PM +0200, Jakub Hrozek wrote:
> On Tue, Sep 03, 2013 at 08:14:09PM +0200, Lukas Slebodnik wrote:
> > On (03/09/13 08:32), Stephen Gallagher wrote:
> > >-----BEGIN PGP SIGNED MESSAGE-----
> > >Hash: SHA1
> > >
> > >On 09/03/2013 07:32 AM, Jakub Hrozek wrote:
> > >> On Mon, Sep 02, 2013 at 07:04:22PM +0200, Lukas Slebodnik wrote:
> > >>> ehlo
> > >>> 
> > >>> Patches are attached.
> > >>> 
> > >>> LS
> > >> 
> > >> (I started this review yesterday in the evening and then didn't had
> > >> the energy to finish it so some comments overlap with Sumit's)
> > >> 
> > >> The SSSD build still works fine with these patches on Fedora 19
> > >> (master) and for patches which could be applied on top of 1.9, I
> > >> also tested RHEL-5. I haven't tested RHEL-6 or any other
> > >> distribution.
> > >> 
> > >>> From 6e3b789f4b24198b2ec4b40fb09e8b97e578044a Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 01:12:01 +0200 Subject: [PATCH 2/9] AUTOTOOLS: add check
> > >>> for type intptr_t
> > >>> 
> > >>> We check whether HAVE_INTPTR_T is defined in definition of macro 
> > >>> discard_const_p, but autootols macro AC_CHECK_TYPE did not
> > >>> generate it. --- src/external/sizes.m4 | 3 +-- 1 file changed, 1
> > >>> insertion(+), 2 deletions(-)
> > >>> 
> > >>> diff --git a/src/external/sizes.m4 b/src/external/sizes.m4 index
> > >>> 53df61dedc3ae748c50ca9a3935632087834155d..1018d846016541043d81fb2a53609ad9c562071a
> > >>> 100644 --- a/src/external/sizes.m4 +++ b/src/external/sizes.m4 @@
> > >>> -37,8 +37,7 @@ AC_CHECK_SIZEOF(off_t) AC_CHECK_SIZEOF(size_t) 
> > >>> AC_CHECK_SIZEOF(ssize_t)
> > >>> 
> > >>> +AC_CHECK_TYPES([intptr_t]) AC_CHECK_TYPE(intptr_t, long long) 
> > >>> AC_CHECK_TYPE(uintptr_t, unsigned long long) 
> > >>> AC_CHECK_TYPE(ptrdiff_t, unsigned long long)
> > I removed these teo lines and replaced with:
> > +AC_CHECK_TYPES([intptr_t],
> > +               [],
> > +               [AC_DEFINE_UNQUOTED([intptr_t], [long long],
> > +                                   [Define to `long long'
> > +                                    if <stdint.h does not define.])])
> 
>                                           ^^^^^^^^^
>                                     you missed a closing bracket
> 
> Otherwise ACK.
> 
> If you agree, I will just add the bracket before pushing.
> 
> > 
> > >> 
> > >> I think this is OK but maybe we should have a ticket to get rid of
> > >> the obsolete AC_CHECK_TYPE invocation? See 
> > >> https://www.gnu.org/software/autoconf/manual/autoconf-2.67/html_node/Obsolete-Macros.html#Obsolete-Macros
> > >>
> > Yes, We should file a ticket to get rid of obsolete macros.
> > 
> > >> 
> > >for some details. The documentation explicitly cites using AC_CHECK_TYPE
> > >> for detecting pointer types as broken.
> > >> 
> > >> I would also like a comment added or the "AC_CHECK_TYPE(intptr_t" 
> > >> removed because currently the code looks a little odd with two 
> > >> subsequent lines checking for intptr_t.
> > >> 
> > >>> From e5d01ae5690d52b83dbdb621bf4e4077fc65abf4 Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 05:16:58 +0200 Subject: [PATCH 3/9] AUTOTOOLS: Add
> > >>> -LLIBDIR to PYTHON_LIBS
> > >>> 
> > >>> Detect directory with python libraries and add this directory to
> > >>> the list of directories to be searched for linker. --- 
> > >>> src/external/python.m4 | 3 ++- 1 file changed, 2 insertions(+), 1
> > >>> deletion(-)
> > >>> 
> > >>> diff --git a/src/external/python.m4 b/src/external/python.m4 
> > >>> index
> > >>> 00487a746a44a945e7bcd92a8f1f2ddd2b4eba80..5f48bc92618b2fbc23610f67f86e38f659977c5a
> > >>> 100644 --- a/src/external/python.m4 +++ b/src/external/python.m4 
> > >>> @@ -19,7 +19,8 @@ dnl versions of python PYTHON_LIBS="`$PYTHON -c
> > >>> \"from distutils import sysconfig; \ print \\\"
> > >>> \\\".join(sysconfig.get_config_var('LIBS').split() + \ 
> > >>> sysconfig.get_config_var('SYSLIBS').split()) + \ -            '
> > >>> -lpython' + sysconfig.get_config_var('VERSION')\"`" +
> > >>> ' -lpython' + sysconfig.get_config_var('VERSION') + \ +
> > >>> ' -L' + sysconfig.get_config_var('LIBDIR')\"`" 
> > >>> AC_MSG_RESULT([yes]) else AC_MSG_ERROR([no. Please install python
> > >>> devel package]) -- 1.8.3.1
> > >>> 
> > >> 
> > >> ACK. Out of curiosity, what directory is Python's LIBDIR on
> > >> FreeBSD? /usr/local/lib ?
> > >>
> > Yes, but portable way is to call "sysconfig.get_config_var('LIBDIR')"
> > or to use pkg-config (I don't like second option)
> 
> Yes, I like LIBDIR, better, too.
> 
> ACK
> 
> > 
> > >>> From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31 Aug
> > >>> 2013 05:36:49 +0200 Subject: [PATCH 4/9] AUTOTOOLS: Add missing
> > >>> AC_MSG_RESULT
> > >> Ack, but Sumit is right about adding [].
> > >> 
> > Changed.
> 
> ACK
> 
> > 
> > >>> From c338c7d721d6bef4fda6ad84321dd3096a14612a Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 07:08:32 +0200 Subject: [PATCH 5/9] AUTOMAKE: Use
> > >>> portable way to link with dlopen
> > >> 
> > >> I couldn't apply this patch on top of 1.9. It works on Fedora 19, 
> > >> though.
> > >>
> > I din't have a problem with this, beacuse I cherry-picked all patches directly from
> > master. I don't want to complicate this.
> > So you have two options either to use "cherry-pick" form master or "git am --3way".
> > 
> > I created patch from master and 1-9 branch and patches was very similar.
> > I don't know why it can not be applied cleanly.
> > 
> > >>> 
> > >>> --- Makefile.am  | 4 ++-- configure.ac | 1 + 2 files changed, 3
> > >>> insertions(+), 2 deletions(-)
> > >>> 
> > >>> diff --git a/Makefile.am b/Makefile.am index
> > >>> 9e68ad9b3a16c9106311eb833a97b01714656a61..1d08dd08b19a56017fbff908811e6a58ab0f7a74
> > >>> 100644 --- a/Makefile.am +++ b/Makefile.am @@ -719,7 +719,7 @@
> > >>> sssd_be_SOURCES = \ src/providers/dp_refresh.c \ 
> > >>> $(SSSD_FAILOVER_OBJ) sssd_be_LDADD = \ -    -ldl \ +
> > >>> $(LIBADD_DL) \ $(SSSD_LIBS) \ $(CARES_LIBS) \ 
> > >>> $(SSSD_INTERNAL_LTLIBS) @@ -1096,7 +1096,7 @@
> > >>> simple_access_tests_CFLAGS = \ $(CHECK_CFLAGS) \ -DUNIT_TESTING 
> > >>> simple_access_tests_LDADD = \ -    -ldl \ +    $(LIBADD_DL) \ 
> > >>> $(SSSD_LIBS) \ $(CARES_LIBS) \ $(CHECK_LIBS) \ diff --git
> > >>> a/configure.ac b/configure.ac index
> > >>> 19bd1928f8a2149d3bf8e8f7a8d46ab7343a0bd6..d16054efae73e8844ac236d0220f274684c5ce8d
> > >>> 100644 --- a/configure.ac +++ b/configure.ac @@ -20,6 +20,7 @@
> > >>> m4_ifdef([AM_PROG_AR], [AM_PROG_AR]) AC_DISABLE_STATIC 
> > >>> AC_PROG_INSTALL AC_PROG_LIBTOOL +LT_LIB_DLLOAD
> > >> 
> > >> I thought that LT_* macros were low-level libtool macros? Can you
> > >> check if AC_LIBTOOL_DLOPEN would do the same thing? Also I would
> > >> expect that any macro that would set $LD should be called before
> > >> AC_PROG_LIBTOOL not after.
> > >>
> > Macro LT_LIB_DLLOAD can be safely use, it is not private.
> > 
> > 5.1 Autoconf macros exported by libtool
> > http://www.gnu.org/software/libtool/manual/html_node/Autoconf-macros.html
> 
> Right, I think I was confused by the LT_ prefix.
> 
> ACK
> 
> > 
> > >>> From a98852a669c5df89a0c611cc9324be32e6d05c0e Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 08:50:47 +0200 Subject: [PATCH 6/9] AUTOMAKE: Use
> > >>> portable way to link with gettext
> > >> ACK
> 
> Still ACK
> 
> > >> 
> > >>> From beb84b3339a79de8f0f4a57b2e61642542b0e4ef Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 11:15:03 +0200 Subject: [PATCH 7/9] AUTOTOOLS: Add
> > >>> directories for searching ldap headers and libs
> > >>> 
> > >>> --- src/external/ldap.m4 | 4 ++-- 1 file changed, 2
> > >>> insertions(+), 2 deletions(-)
> > >>> 
> > >>> diff --git a/src/external/ldap.m4 b/src/external/ldap.m4 index
> > >>> 8899df9d7c73dae25b97347022651d768986333e..3a99ddfcc7e7ebee94272d382e2876ec8649770a
> > >>> 100644 --- a/src/external/ldap.m4 +++ b/src/external/ldap.m4 @@
> > >>> -9,14 +9,14 @@ dnl
> > >>> ---------------------------------------------------------------------------
> > >>>
> > >>> 
> > >dnl - Check for Mozilla LDAP or OpenLDAP SDK
> > >>> dnl
> > >>> ---------------------------------------------------------------------------
> > >>>
> > >>>  -for p in /usr/include/openldap24; do +for p in
> > >>> /usr/include/openldap24 /usr/local/include; do if test -f
> > >>> "${p}/ldap.h"; then OPENLDAP_CFLAGS="${OPENLDAP_CFLAGS} -I${p}" 
> > >>> break; fi done
> > >>> 
> > >>> -for p in /usr/lib64/openldap24 /usr/lib/openldap24; do +for p in
> > >>> /usr/lib64/openldap24 /usr/lib/openldap24 /usr/local/lib ; do if
> > >>> test -f "${p}/libldap.so"; then OPENLDAP_LIBS="${OPENLDAP_LIBS}
> > >>> -L${p}" break;
> > >> 
> > >> Don't we also need to add /usr/local/lib64 ? (Sorry, I don't know
> > >> how multilib works on FreeBSD)..
> > >> 
> > I have a 64-bit virtual machine and directory /usr/local/lib64 does not exist.
> 
> OK, thank you for checking. ACK.
> 
> > 
> > 
> > >>> From 2e939d50ce4837a76bb4850e6fb6b1c1435c27c1 Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 12:17:38 +0200 Subject: [PATCH 8/9] AUTOTOOLS: Use
> > >>> pkg-config to detect libraries.
> > >>> 
> > >>> We used pkg-config only as a fallback if header files was not
> > >>> found, but detection of library failed in case of available
> > >>> header file and linking problem (missing -Ldir).
> > >>> 
> > >>> This patch prefers pkg-config.
> > >> 
> > >> The other reason is that in the past, some of the libraries didn't
> > >> ship a .pc file. Since you would like to have these patches in 1.9
> > >> (I assume), can you check that they have pkg-config files in the
> > >> older versions, too?
> > >> 
> > FreeBSD have .pc files.
> > popt-devel does not ship a .pc file on fedora 19 :-)
> > We need fallback.
> 
> ACK
> 
> > 
> > >>> From faa431369b66684024272392da507d374695aaea Mon Sep 17 00:00:00
> > >>> 2001 From: Lukas Slebodnik <lslebodn at redhat.com> Date: Sat, 31
> > >>> Aug 2013 15:32:39 +0200 Subject: [PATCH 9/9] AUTOTOOLS: Refactor
> > >>> unicode library detection
> > >> 
> > >> Seems to work fine, but see Sumit's comment in his response..
> > Changed 
> > 
> 
> ACK
> 
> > >
> > >
> > >I ran these patches in a scratch-build against RHEL 6 (which is the
> > >oldest supported OS we have for master). It builds successfully, from
> > >which I will assume that all of the macros we are relying on either
> > >exist or are falling-back gracefully.
> > I tested srpm from master and sssd-1-9 with various mock configs. So it must work.
> > >
> > >
> > >
> > >Patch 0002: See Sumit and Jakub's comments about the macro.
> > >
> > >Patch 0003: Ack
> > >
> > >Patch 0004: See earlier comments about escaping.
> > As I mentioned before. It is changed.
> > 
> > >
> > >Patch 0005: Ack
> > >
> > >Patch 0006: Ack
> > >
> > >Patch 0007: are /usr/local/include and /usr/local/lib not standard
> > >lookup paths on BSD? The reason for this entry here is entirely for
> > >support of RHEL 5 (which we have dropped) because we were carrying a
> > >parallel-installable version of the openldap client libraries in a
> > >non-standard location.
> > Standard directories are /usr/include and /usr/lib, but all packages
> > are installed to /usr/local/.
> > I would like to push these patches also to 1-9-branch.
> > 
> > >
> > >Patch 0008: I *think* the original reasoning for this behavior was
> > >that we intended a Solaris port which didn't support pkg-config
> > >properly at the time. I'm not sure whether Solaris/Illumos today has
> > >since implemented it properly. That said, we're not currently focusing
> > >on that OS, so I'm willing to leave that as "patches welcome".
> > >
> > >Patch 0009: Can we drop libunistring in 1.11/1.12? We only originally
> > >supported it because we were wary that pulling in Glib would have
> > >issues with dueling implementations of D-BUS and a mainloop, but we've
> > >since proven that this isn't a problem. Of course, if Glib is not
> > >available on our target platforms, this becomes less obvious.
> > I think we can leave it as it. libunistring can be untested.
> > The similar situation is with nss and libcrypto.
> > 
> 
> I also suppose that some distributions that prefer lower footprint would
> like to keep libunistring support in. It was tested in RHEL and the code
> didn't change afterwards.

btw I also just tested mock builds with F-19 and RHEL-6 and they passed
just fine.



More information about the sssd-devel mailing list