On 1 June 2016 at 21:21, Mehdi Amini <mehdi.am...@apple.com> wrote: > I'm not sure how to be robust against that other than putting all the > projects in the same repo and asking developers to build them all before push.
I'm strongly against a single repo with all in, or asking to build LLDB when the commit is in Compiler-RT. As usual, appropriate tests are all we need. I share your view that we already have a window of possible breakage due to co-dependant commits and therefore can afford the same behaviour in Git. IIGIR, sub-modules are meant to coordinate the repositories, not to solve the problematic short window of doom. Even having a single repository with all in won't help this problem. Unless there is a way to link two commits to two different submodules and make the bots know that they can't be tested separately, we should just accept that some bots will break if they're extremely unlucky. Having said that, pre-commit bots will break with most co-dependant patches, and we must have a way to either ignore those failures and still merge, or to mark them as co-dependants and test again. cheers, --renato _______________________________________________ lldb-dev mailing list lldb-dev@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-dev
