Problems with core review

Ralf Corsepius rc040203 at freenet.de
Wed Feb 7 05:56:50 UTC 2007


On Tue, 2007-02-06 at 21:34 -0800, Christopher Stone wrote:
> On 2/6/07, Christopher Stone <chris.stone at gmail.com> wrote:
> > On 2/6/07, Ralf Corsepius <rc040203 at freenet.de> wrote:
> > > On Tue, 2007-02-06 at 20:34 +0000, Joe Orton wrote:
> > > > On Tue, Feb 06, 2007 at 11:49:11AM -0800, Christopher Stone wrote:
> > > > > Here are the issues in question:
> > > > >
> > > > > 1) Replace use of $RPM_SOURCE_DIR with %{SOURCEx}
> > > > >
> > > > > I asked about this in #fedora-extras since I did not understand
> > > > > rpmlints Error message. f13 responded by saying you should just use
> > > > > %{SOURCEx}.
> > > > >
> > > > > I agree with f13 on this issue because it is easier to identify in the
> > > > > spec file where the source files are used.
> > > >
> > > > Con: it makes renumering Sources a pain, it's harder to use since you
> > > > have to remember numbers not filenames.  This number/filename mapping
> > > > trick doesn't scale well as anybody who has maintained spec files with
> > > > more than a handful of patches knows.
> > > >
> > > > Insufficient justification for change.
> > > >
> > > > > 2) Add empty %build section even though its not required
> > > > >
> > > > > All php-pear packages include an empty %build section and php-pear
> > > > > should not be an exception.  This was disccussed at length when
> > > > > creating the php-pear spec file template.  Ville has real world
> > > > > examples how this can cause problems.
> > > >
> > > > What are they, how do they apply to this package?
> > > rpm doesn't generate debug-infos if %build is not present.
> > >
> > > > > Technical reason for changing:  rpm is unpredictable with no %build,
> > > > > consistency among all pear packages
> > > >
> > > > It's worked predictably for the history of this package.
> > > Only if all those package had been noarch'ed.
> > >
> > > If not, you surely have broken debug-infos.
> >
> > I brought up the fact that all php-pear packages are noarch, yet this
> > requirement was imposed on all php-pear packages anyway.
> >
> > The php-pear default spec template adds an empty %build section even
> > though I argued against such an addition.
> >
> > Therefore, I do not see why php-pear should be an exception to this
> > rule.  Why is it imposed on all other php-pear packages except for
> > php-pear itself?
> >
> 
> It should also be noted, that if rpm can't figure out how to do things
> correctly with arch specific packages that do not have a %build, then
> can we really trust rpm to function properly on packages that are
> noarch?

There isn't much I can add to this ;)

I've brought up the %build/debug-info bug several times before (IIRC, I
even bugzilla'ed it), as well as the general unreliability of debug-info
generation, but so far, I have seen no indication of improvement -
Instead, people preferred to shoot at me.

> I think this is a valid argument.  If the logic inside rpm is so bad
> that it can't handle things properly without certain tags, then we
> better be really careful about how we format our spec files.

Well, I am not sure about this - Instead of playing with symptoms
(missing %build) we should aim at getting the real causes fixed 
(read: redhat-rpm-config).

Ralf





More information about the Fedora-maintainers mailing list