A quick fix to address the false positive posted by Tiago Macarios in the mailing lists: http://lists.llvm.org/pipermail/cfe-dev/2016-December/051738.html
MallocChecker processes pointer escapes heuristically rather than regularly. Force it to treat the pointers passed to connectImpl() as escaping.
I dislike a few things about this patch - we're patching against implementation details and hardcoding names of private methods in an external library. I don't see a significantly better solution within the current approach though.
See also D27599 - a similar hack for another function.