First some background on what made me think of this.  On the apache 2.2.22 update that just happened on Mandriva 2010.2, there was a new symbol in httpd that one of the apache modules depended on.  urpmi updated the apache-modules before the apache-mpm (which contains httpd), and when the post scriplet tried to restart httpd, it failed because of the missing symbol.  Then when the apache-mpm was installed, the post scriplet didn't try to restart httpd, because it was no longer running.

This got me thinking that there are a couple of possible ways urpmi could handle post scriplets better that would have mitigated this problem.

One is that if urpmi is upgrading a bunch of packages that run the same post scriplet, to wait until they are all installed and just run it once.  A possible problem I could foresee with this is if not all of the packages get installed in the same transaction (when urpmi splits it up), it's possible all of the packages aren't downloaded yet such that urpmi can check their post scriplets.

The other idea is that with post scriplets that check to see if a service is running and restart it if it is, to remember interally in urpmi the result of the first time the test for the running service is run, and then use that result instead of re-running the same test for other post scriplets.  How would this have helped?  If it remembered when it tested on apache-modules that httpd service was running, even though restarting it at that point failed, when apache-mpm ran the same scriplet, it would know that httpd service had previously been running, so it would try to restart it even though it was no longer running.