> Guido Berhoerster wrote:
> > I just became aware of the discussion on the
> contrib repo on
> > sw-porters-discuss through a cross-post, it does
> not seem to have
> > received attention there. The thread started here:
> >
> http://mail.opensolaris.org/pipermail/sw-porters-discu
> ss/2008-November/000027.html
> > The proposal can be read here:
> > http://www.genunix.org/wiki/index.php/ContribRepo
> > 
> > As this is about the future of OpenSolaris I'm
> posting my
> > thoughts on indiana-discuss.
> > With the current proposal the contrib repository
> becomes IMHO a
> > dumping ground for packages which have been built
> by someone
> > somehow. I consider a non-reproducible build, that
> is how, with
> > what patches, compiler options etc. the package was
> built,
> > problematic for a number of reasons:
> 
> I've heard repeatedly over time that barriers to
> entry are the excessive 
> amounts of processes put in place around
> opensolaris.org contributions.
> 

writing a .spec file and running diff -u is not an excessive amount of process. 
It is a reasonable safeguard. It's not a week's worth of paperwork, the spec 
file is not too much different from just a package manifest, and the diffs take 
30 seconds. 

> While all of the things you mention are certainly
> good for the long-term 
> management and health of the project, in the short
> term, the biggest 
> problem we face isn't non-reproducible builds -- it's
> a lack of packages.
> 

bad packages are worse than no packages. 

> For some of the same reasons that pkg(5) doesn't
> include a build system, 
> I don't think it's reasonable to require that
> contributed packages have 
> .spec files, since native ips package bundles ready
> for publishing 
> should be permittable too.  In addition, it also
> assumes that all build 
> environments can be automated; that also isn't
> necessarily true.
> 

if the build env. can't be automated then it's a bad package

> > * The first is security, if the package is built by
> some unknown
> >   submitter, a review of the binary package only
> cannot easily
> >   enforce quality and make sure no backdoors are
> injected etc.
> 
> ...which source code and a reproducible build does
> not guarantee as the 
> recent vulnerability in Debian that was there for
> years proved.
> 

which was found due to source, not due to someone getting bitten by it.

> > * Lastly, it is problematic with some FOSS licenses
> such as the
> >   GPL which require a distributor of binary
> packages to make
> >   available the _corresponding_ source, simply
> providing a URL to
> >   external sources (as proposed) is not sufficient.
> 
> For software that has licenses that require it, it
> obviously is, required.
> 
> > The restrictions excluding Sun employees is just
> ridiculous
> > considering that they make up the vast majority of
> the community.
> > Do SFE maintainers who are employed by Sun have to
> find non-Sun
> > sponsors or are they completely excluded? Either
> way this cuts
> > severely into resources.
> 
> Legal reasons as was cited previously.
> 
> > Cutting red tape is certainly appreciated but going
> from one
> > extreme to the other does not seem to be an
> improvement to me.
> > Please look at proven procedures of how similar
> projects operate,
> > e.g. Redhat/Fedora, Debian non-main, Ubuntu
> universe, FreeBSD
> > Ports, pkgsrc.
> 
> So far I haven't seen a reasonable compromise between
> either.  As far as 
> modeling the process on one of the distribution
> you've mentioned, I 
> think a key flaw with that idea is that they're
> nanaging 15,000+ 
> packages; we have 0.

and Package Build-Tron 5000 can turn 80% of random .tar.gz files in to packages 
in minutes, it also makes julienne fries.

distribute a .spec-ish file stub that does  "tar -zxvf $FILENAME ./configure; 
make; make install" as an automated check box and most random sourceforge junk 
will just work. if things need to be made different ( different cflags, etc ), 
specify them on the "upload package" page


> I think a better approach would be start lightweight
> and add processes 
> as the number of packages grows.

short-selling your future is a bad idea all around 


> It's silly to me that we're spending more time
> developing processes than 
> packages.

the two aren't mutually exclusive.
-- 
This message posted from opensolaris.org

Reply via email to