Rob Crittenden wrote:
Nathan Kinder wrote:
> Richard Megginson wrote:
>> Andrew Bartlett wrote:
>>> On Wed, 2006-10-25 at 18:14 -0700, Pete Rowley wrote:
>>>
>>>> Andrew Bartlett wrote:
>>>>
>>>>> Our current policy is to generate these files for release
>>>>> tarballs, and
>>>>> for our 'unpacked' tree on
samba.org (current SVN checked
out).
>>>>>
>>>> OTOH they are required in order to do:
>>>>
>>>> cvs co
>>>> ./configure
>>>> make
>>>>
>>>
>>> Yeah, projects typically end up with an ./autogen.sh to make the right
>>> innovation of the configure generation tool.
>>>
>> I've found that using autoreconf usually does the right thing. When
>> I change configure.ac/in or Makefile.am or an .m4 file, I always run
>> autoreconf -vfi
>> -v, --verbose verbosely report processing
>> -f, --force consider all files obsolete
>> -i, --install copy missing auxiliary files
>> It takes a little longer, but I almost never have conflict or
>> timestamp problems. Plus, it's part of the standard autotools
>> package, and it is the way the autoconf/automake manuals recommend
>> rebuilding the autotool files.
>> For some projects, this won't work (e.g. for mozldap, you have to
>> just use autoconf-2.13, not autoreconf or autoreconf-2.13).
> As I just very recently found out, we also need a very specific
> version of libtool (1.5.22) to generate ltmain.sh if we want to be
> able to build a 64-bit Directory Server on Solaris. Running
> "autoreconf -fvi" will generate a new ltmain.sh that may be a version
> that we don't want to check in if we expect to be able to immediately
> run "configure; make install" after checking out the code.
>
> -NGK
The real pain is when not all of the files have changed and you check
in only those that did. This can cause an unwanted auto* rebuild.
I've taken to checking everything in at once whenever one thing
changes with:
cvs ci -f Makefile.am configure.in aclocal.m4 Makefile.in configure
This preserves the proper timestamp/dependency order (at least for me).
If autoreconf doesn't work for directory server, do we need to create an
autogen.sh script that "does the right thing"? I would rather not, but
if that is the only way to preserve the correct version of ltmain.sh,
then we need to do it. For example, we won't be able to use libtool on
RHEL4 (with the standard RH updates anyway) because the bundled version
of libtool is 1.5.6. However, Fedora Core 5 (and probably RHEL5) are ok
because they use 1.5.22. I fear this may not be the only problem with
autotool compatability - we also had a problem with hpux 11.23 ia64.
So, some options:
1) Only use autoreconf, and only run it on systems that use recent
versions of autotools (e.g. fc5) - do not allow the use of autotools on
RHEL4 or other less recent platforms.
2) Create an autogen.sh that first looks at the versions of the tools on
the system and bails out if it finds an incompatible version - note that
we would still have to run it on a modern system in order to actually
update the autoconf files
2a) Have autogen.sh "patch" ltmain.sh when run on less recent systems,
to allow us to use it on RHEL4 etc.
3) Make ltmain.sh (and possibly other files) somehow read-only with
respect to autoreconf
rob
------------------------------------------------------------------------
--
Fedora-directory-devel mailing list
Fedora-directory-devel(a)redhat.com
https://www.redhat.com/mailman/listinfo/fedora-directory-devel