[Fedora-packaging] Drafts for next Tuesday

[Ville Skyttä]

On Thursday 20 March 2008, Tom "spot" Callaway wrote:

> Perl Guidelines (spot) :
> http://fedoraproject.org/wiki/PackagingDrafts/Perl

1) cpanspec should be pushed further down in the draft.

2) Versioned MODULE_COMPAT_ Requires:
"This is to ensure that all perl modules are built against the appropriate 
version of perl."

This rationale is wrong - it doesn't ensure that, but that packages have a 
dependency to a perl that uses stuff from dirs versioned with that version 
number.

3) ExtUtils::Build doesn't exist AFAIK, did you mean Module::Build?

> Eclipse Plugin Guidelines (spot) :
> http://fedoraproject.org/wiki/PackagingDrafts/EclipsePlugins
> OpenOffice.org extensions guidelines (Caolan McNamara) :
> http://fedoraproject.org/wiki/PackagingDrafts/OpenOffice.orgExtensions

These look mostly sane, will need to read them some more though.

> Secure BuildRoot (Lubomir Kundrak) :
> http://fedoraproject.org/wiki/PackagingDrafts/SecureBuildRoot

-1 to any buildroot suggestion that doesn't propose implementing it internally 
in rpm aiming for eventual deprecation and elimination of the BuildRoot tag 
(and related "rm -rf"'s) in specfiles.

Anyway, specific to the submitted draft, both alternatives cause buildroot 
trashing even with innocent "rpm -q --specfile foo.spec" or "rpmbuild -bp 
foo.spec" or "rpmbuild -bs foo.spec".

Additionally, the second makes dangerous assumptions which can wreak havoc 
in %clean when one overrides the build root (eg. %buildroot in ~/.rpmmacros).

> Register VirtualProvides (Patrice Dumas) :
> http://fedoraproject.org/wiki/PackagingDrafts/ProvidesList

+1 to the general idea, however I'm somewhat uncertain about 
server(port_name), it needs more explanation.  Consider for example the 
tomcat5 package: it's configured to use port 8080 by default.  I don't think 
server(webcache) would describe it well at all.  Also, changing servers to 
run in non-default ports is pretty common and kind of breaks the "contract" 
of server(port_name), but perhaps that's just a documentation issue.