Recommending software upgrades with Mojave.

•We characterize 96 bug reports spanning 5 versions of OpenSSH. The results show that 41% of the bugs were due to upgrades. In addition, we find that the majority of the bugs are caused by both the user's environment and input.•We propose Mojave, the first recommendation system for software upgrades. Mojave leverages data from the existing and new users, machine learning, and static and dynamic source analyses to recommend for or against an upgrade to new users.•We evaluate Mojave for 5 upgrade problems with 3 widely used applications. Across all our experiments, Mojave accurately predicts the upgrade failure for 93–100% of the users.