On 18 June 2014 12:23, Mark Derricutt <[email protected]> wrote: > Interesting, I hadn't noticed this was only a warning. > > The main problem I have is the moment you start relying on these fake > reactors you end up breaking C.I. builds with dependant changes. >
Why? Your CI should know that project B depends upon A. If you checkin B first, then yes you will get problems. If you checkin the correct order then the CI will stack the builds so you wont have a problem. Either in this thread or elsewhere was the guidance to avoid have a snapshot repo for developers. But this is needed for CI, as each build should be using a clean local repo - so it needs to pull snapshots from somewhere else as they wont be in your now empty repo, nor build as part of the reactor. > > This is my main beef with multiple git repositories for projects - if you > have a pairing of dependant changes ( an API change and an Implementation > change ) - without specific tooling about sharing local-repos for discreet > dependent change sets you end up with false-negative build failures. > This is a process issue. Either your CI is too aggressive with its code change checking window, or you need to slow down commits, and make sure they are in the correct order, so that you avoid the false-positive failures.
