[syslog-ng] A couple of notes regarding packagability

[Gergely Nagy]

Balazs Scheidler <bazsi at balabit.hu> writes:

For the record, I made a separate tree for Debian packaging, it's at
git://git.madhouse-project.org/debian/syslog-ng.git - the non-debian
specific patches are on the 'patched' branch.

I'm not going to submit the new debianisation for inclusion, as it does
not work with pre-squeeze releases.

>> * autogen.sh is not executable in a fresh checkout, and this repeatedly
>>   bites me over and over again.
>
> hmm, it is executable for me, and is marked with the 'x' permission in
> the git repository, at least according to gitweb:
>
> http://git.balabit.hu/?p=bazsi/syslog-ng-3.3.git;a=tree;h=8a7f6854d26a7ea8fa8c1ff57554046481eb0786;hb=8a7f6854d26a7ea8fa8c1ff57554046481eb0786

Hrm, indeed. Must be something stupid on my buildds then. Nevertheless,
for my own sanity, and until I can figure out what goes wrong:

http://git.madhouse-project.org/debian/syslog-ng/commit/?h=patched&id=a6b9824b87e7edc6232fe4712c3456a679766253

>> * autogen.sh will always try to call git submodule init when
>>   autogenerating the submodules. I'd like a way to skip this step, and
>>   just use whatever is in the subdirectories, assuming there IS
>>   something.
>> 
>>   The reason I want this, is because I'm building from a git checkout,
>>   not a make dist'd tarball, and my build chroots do not have network
>>   access (nor do I want to give them any).
>
> I was thinking about this already too, so knocking on open doors. I've
> commited a Proof-of-concept on my autogen branch.

Looks ok to me! Though, I did something different locally:

http://git.madhouse-project.org/debian/syslog-ng/commit/?h=patched&id=d31718e16d757920e3b8e5c0b9745eb7f834847a

I kinda like how I can explicitly disable the git call. Perhaps we could
combine the two? Though, that might be an overkill, in which case, I
like your approach better.

>> * It would also be nice if autogen.sh used the --copy flag for
>>   libtoolize and automake.
>
> ok, not done anything on this.

http://git.madhouse-project.org/debian/syslog-ng/commit/?h=patched&id=c2d24b620629a659229803fb41311b926bc95f27

>> * When the system-wide libmongo-client support patch is applied, and
>>   syslog-ng is built with --with-libmongo-client=system, make distclean
>>   will fail, as it tries to clean in modules/afmongodb/libmongo-client/,
>>   while that hasn't been configured in.
>> 
>>   I do not know an easy solution around this, except maybe overriding
>>   DIST_SUBDIRS with SUBDIRS for the distclean-recursive target in
>>   modules/afmongodb/ - but that sounds like a crude hack, and I haven't
>>   tested if it actually works.
>> 
>
> I'd have to think about this. The reason is that AC_CONFIG_SUBDIRS() is
> not invoked in this case, but calling that would run the configure
> script for libmongo-client, even if it wasn't used.

Yeah, I wouldn't want to run configure in there if using the system
library. I'd rather tweak distclean to not go there, either, in that
case.

I'll play with the DIST_SUBDIRS override, and report back.

>> PS: syslog-ng 3.3.0beta2 packages for Debian squeeze & sid should be
>> available by the end of the coming week, for i386, amd64 and
>> kfreebsd-amd64 at least. I just have to iron out a few issues and add
>> some finishing touches to my build farm.
>
> very cool.

Packages reporting in, as of ~5 minutes ago! Though, only for amd64 at
this time (but built for both stable & unstable).

-- 
|8]