Bug 82024

FWIW, the script I posted works even in the case of
a initial install, as that package is fully, not partially,
installed. Yes, payload unpacking may exit, but that is
detectable with --verify, and blocking signals during
payload unpacking is certainly an option.

OTOH, blocking signals leads to lack of responsiveness,
and causes those with short attention spans to start blasting
with "kill -9", creating a worse problem than before.

Jeff,  I am not getting the results you say I should.

For example I create a partial package from the following spec:
   Summary:        Partial Package
   Name:           partial
   Version:        1.0
   Release:        1
   Group:          Broken/Tools
   Copyright:      No Importante
   Vendor:         Nadia
   BuildArchitectures:     noarch
   Provides:               %{name} = %{version}

   %description
   A partial package.

   %post
   exit 1

   %files

I build it with:

   rpmbuild -ba partial.spec

I install it (not upgrade) with:

   rpm -ivh partial-1.0-1.noarch.rpm

It installs but fails in the post sections (on purpose of course).
I run:

   rpm -V partial

This gives no output and returns 0.  Secondly if I do:

   rpm -qa --qf "%{NAME}\n" | grep partial

I get one instance shown of partial (which would make the 
sort uniq hack invalid).

If failing %post is your definition of "partial install",
then the answer is
   Fix your packaging
THere are bigger problems than busted %post that need to be solved.

Look, I was doing this as a test case for a very real problem.  Just because
the kernel panic's does not mean that there is a bug in the kernel.  Sometimes
this is true, but often it may mean that something in the kernel's environment
(i.e. the hardware) has gone way south and the only right thing to do is panic.
In the same way, as one is doing error checking inside your post script, you 
may come across cases that the only right thing to do is exit with an error.
Furthmore, if you come accross a system that you are upgrading, and you can
see that the present instance of a package on the system is partially installed
you had better not simply upgrade anyway, but instead stop with a meaningful
error message that will give someone a clue.

Furthermore, the multiple instance method of checking for partial packages
does not allow for one to legitamitly have multiple instances of a particular
package (of the same name) installed.  There are more examples of this 
than just the kernel RPM's.

Lastly, though there may be bigger problems.  It is a problem.  No need to get
your feathers ruffled.  If RedHat cannot fix it anytime soon, well we will
have to cope, but it is a problem.  It would be real nice to see it in AS 3.0,
though.

In fact, the package with "exit 1" in %post is completely
installed with --install. Yes, any side effects from running
%post that occur after the exit are not performed.

Otherwise, the check I gave is adequate for detecting duplicately
named packages, which occur in practice only for the kernel.

This behavior is not going to change in the near future. not
because my feathres are ruffled, but because my marching orders
are different.

RHEL 2.1 is in maintenance mode, this will not be addressed in it.