glibc-headers RPM bug? (was: Re: Kernel dependency issues)

[Panu Matilainen]

On Sat, 13 Dec 2008, Michael Schwendt wrote:

> On Fri, 12 Dec 2008 22:43:04 +0200 (EET), Panu wrote:
>
>> Yup, just checked (from F-10 glibc.spec):
>>
>> %package devel
>> Summary: Object files for development using standard C libraries.
>> Group: Development/Libraries
>> Requires(pre): /sbin/install-info
>> Requires(pre): %{name}-headers = %{version}-%{release}
>
>   -R, --requires
>          List packages on which this package depends.
>
>  $ rpm -qR glibc-devel|grep ^glib
>  glibc = 2.9-2
>  glibc-headers = 2.9-2
>
> As -R/--requires lists any sort of Requires() [even post/postun/...],
> the results of this query are misleading, at least. A requirement
> that is not a requirement, because rpm -e ignores it.

No disagreement there, the way rpm reports, handles and verifies 
dependencies is misleading and buggy too in places.

>> "Requires(pre): pkg1" means "pkg1" is required to be present during
>> execution of %pre scriptlet of the package and nothing else. It does not
>> imply "Requires: pkg1", that would have to be separately added if pkg1 is
>> needed during installation *and* runtime.
>
> That ought to be fixed in glibc-devel, then.
>
> glibc-devel doesn't do anything in %pre that would require glibc-headers.
> Perhaps it used to be PreReq (I won't look at old pkg releases now).

That was by guess too.

> Changing the Requires(pre) into a normal Requires would be plausible.
> The packages share a single %_includedir/gnu -- erasure ordering is a
> mess anyway with regard to unowned directories and complex dep-chains.
> It would not be wrong to include the directory in both packages.

Yup, and besides regular Requires should guarantee correct order (which 
has been the case for several years) unless there are loops, which is then 
another issue (Fedora package set is FULL of dependency loops).

 	- Panu -