noarch rpm feature query/issue

[Mike Bonnet]

Peter Robinson wrote:
> I have a query about the new noarch feature for building rpms. I have
> a package called gupnp-vala which the contents there of can be noarch.
> It built fine in koji as noarch so I presume it passed the rpmdiff or
> what ever is run against the rpms. But it contains a pkgconfig file
> which is located in /usr/lib or /usr/lib64 so pacakges built against
> it (none actually in Fedora yet, but one awaiting review) fail to
> build on either 32 or 64 bit platforms depending on which of the build
> packages gets the lime light.
> 
> So my query is, I presume due to the pkgconfig this package shouldn't
> be noarch, so there must be a bug there somewhere which caused it to
> get past koji. But as its noarch AFAICT other than the location of the
> pkgconfig file it seems somewhat of a waste to lose the advantage of
> the noarch stuff for the sake of a pkgconfig file as I presume most
> devel packages could be noarch.

The rpmdiff you're talking about is only run against noarch 
*subpackages*.  This is a noarch base package, and like every other 
package, we assume the maintainer is handling arch issues appropriately. 
  In this case, a noarch package should not be dropping files under 
%{_lib}, since this will obviously be different depending on which arch 
it was built on.  Maybe there should be a check in redhat-rpm-config to 
avoid this?  I'm not sure how noarch packages are supposed to handle 
pkgconfig files.

> Cheers,
> Peter
> 
> BTW I see noarch packages are located in each of the separate arch
> trees as opposed to being in a noarch dir which looks like the
> advantage of the noarch in space saving isn't happening. What other
> advantages does it offer?

noarch packages are often hardlinked into the different dirs, so though 
they appear multiple times they don't consume additional space.