Debugging rpm package installation order

Jeffrey Johnson n3npq at
Fri Aug 9 22:49:28 CEST 2013

On Aug 9, 2013, at 4:39 PM, Jacek Konieczny wrote:

> On Fri, 09 Aug 2013 11:58:57 -0400
> Jeffrey Johnson <n3npq at> wrote:
>> Maybe. I can say that the existing rpm implementation has been used
>> by many for years without any change whatsoever.
>> FIx your LOOP before guessing at what is broken.
> But THERE IS NO LOOP. RPM shows loop due to some imaginary dependency.

Before shouting (or comparing to rpm-4.5 behavior), you need some additional information.

Meanwhile you have reported a LOOP message (and you have already seen that the
_dependency_whiteout macro changes/simplifies rpm behavior).

> Package A requires package B which requires /some/dir from package FHS.
> But my RPM shows: package A requires package B which requires /some/dir
> from package C which requires package D which requires /some/other/dir
> from package E which requires package A. And package C does not contain
> /some/dir, nor package E contains /some/other/dir. RPM sees dependencies
> which are not there, so it reports loop. Details in my previous mail

rpm-5.x has additional dependency rules:

	1) every file "requires" its parent directory
	2) every symlink "requires" its target end-point

I put the "requires" in quote solely because yoi will _NOT_ see this
information with
	rpm -qp --requires foo*.rpm
queries. Otherwise the additional "requires" behave exactly like
other Requires:, both as dependency assertions and as pre-requsites
for topologically sorting.

73 de Jeff

More information about the pld-devel-en mailing list