[pca] Packages and false patch information

[Martin Paul]

Dan,

> My query is, can PCA be changed to recognise this situation? I see both
> patchdiag.xref and showrev -p have package names associated with each of the
> patches.

Thomas and Laurent already summed it up pretty well. For the archives, 
here's what I wrote in a private message about a similar issue some time 
ago; it is still valid:

I know, and I have put quite some time into this issue in the past. At 
first I came up with a script to identify such "partly installed 
patches" and it seemed to work fine. I built the functionality into pca, 
but soon discovered that it caused problems. It would find some false 
positives (e.g reinstall kernel patches which don't need a reinstall), 
and it fails to identify some special cases, especially with 
pre-installed patches.

At the end I found out (and it was confirmed by a Sun engineer) that 
there just isn't enough information in the system to determine which 
patches need to be reinstalled.

Sun's official line (it's mentioned in some InfoDoc) is that *all* 
patches have to be reinstalled as soon as a package is added to a system 
which already had patches installed. I wouldn't do that, as it might 
cause more issues than it solves.

I really recommend deciding on an install cluster and stick to that. If 
there are more packages than you need, ignore that. Sun only tests 
patches on official clusters, too, and you might see failing patch 
installs when you later add or remove packages. If you *must* add a 
package, grep for the package name+version through patchdiag.xref and 
identify which patches have to be re-installed manually.

I'm not happy about all that either, but we're talking about a package 
system which is far from perfect, and it's hard to put something perfect 
onto such a basis :) Hope you are not too disappointed ..

Martin.