Re: RFS: b5

[Panu Kalliokoski <atehwa@sange.fi>]

On Wed, Mar 01, 2006 at 12:09:49AM +0800, Ying-Chun Liu wrote:
> >>Native packages should ideally only be packages that have no real use
> >>outside of Debian.
[...]
> > If you could show me some document that explains whether and why
> > native packages are not preferred for software that could live
> > outside Debian, I'd appreciate that very much.
> 
> http://lintian.debian.org/reports/Tnative-package-with-dash-version.html

Okay, that kind-of answers the "whether" question -- although I'm not so
sure about the wording: "if ... software was written specifically to be
turned into a Debian package" does not sound, to me, like it means only
software that has no _real use_ outside Debian.  I mean, my software
_was_ written to be turned into Debian packages, and still, it is usable
outside Debian.  Moreover, the actual warning doesn't even relate to
this question, and the remark seems to be more of a loosely-written
explanation than a guideline.

But it still lacks the "why"... maybe I will learn about the "why"
during this discussion.

On Tue, Feb 28, 2006 at 05:12:32PM +0100, Thijs Kinkhorst wrote:
> I'm not sure if it's actually clearly written down, so it might be more
> of an opinion, but it's customary at least to package debian-specific
> things as native and others as non-native packages.

Maybe native-packaging software is more common _outside_ debian?  I've
seen a couple of other people, too, not DD's, including the debian/
subdir right in their release source.

> I'm not sure where you're getting the political and social aspects from,
> as I am only talking about technical aspects.

Yes, I'm sorry I misinterpreted your intent.

> If someone is packaging your software for other distributions and you're
> using a non-native package, it's very clear which part is the shared
> part by all distributions, and which are the Debian-specific changes
> that are being made. 

Hm.  All of these packages are written so that "make install" will
install them (except for python-selecting which uses distutils), or have
installation documentation.  Hard to get more standard than that...  I
understand someone could get nervous if they notice that there is
debian-specific stuff around, but as I've been including the debian/
directory in the release source for years, I guess it's not that
horrible after all.

I really try to keep my packages portable, and not by writing
distro-/OS-specific quirks, but by avoiding use of constructs that are
nonportable.  As such, the packages really don't have any
Debian-specific about them other than metadata.  It is entirely possible
to write a package that is simultaneously RPM-native, Debian-native and
NetBSD-pkgsrc-native, which suggests that "native" is somewhat of a
misnomer.

> This would e.g. also make an NMU more transparent.

Interesting!  What does this mean?  That the NMU only touches the
.diff.gz but not the .orig.tar.gz?  But in both cases (native or
non-native), there is only one file that changes; the only difference is
that in the non-native case, there is also one additional file that does
not change.  But maybe you mean something completely different.

> It's your call, but since making them non-native is not really that much
> more work, I'd recommend doing it that way.

Well, I'm not really complaining, but I really want first to understand
why this work is any good.

-- 
personal contact:	panu.kalliokoski@helsinki.fi, +35841 5323835
technical contact:	atehwa@iki.fi, http://www.iki.fi/atehwa/
PGP fingerprint:	0EA5 9D33 6590 FFD4 921C  5A5F BE85 08F1 3169 70EC