Sorry. It looks like the skip broken code works exactly the opposite way I thought: It removes the packages that leads to the problem. I've commited two test cases that give an idea why this might not be the best strategy (hmm, may be I should add some doc strings...). Florian