beanz added a comment. This patch ends up hiding the problem, not fixing it. If it unblocks something it might be ok, but it doesn't actually fix the problem.
The underlying problem is that liblldb and lldb-mi have their own copies of LLVM with their own copies of the GlobalParser static (llvm/lib/Support/CommandLine.cpp), and the corresponding static for the debug cl::opt (llvm/lib/Support/Debug.cpp). When the library exports the LLVM symbols, the dynamic loader loads lldb-mi and it makes the cl::opt constructor point at the same instance of the GlobalParser (following link-once ODR rules). When the library doesn't export the LLVM symbols, the cl::opt variables inside liblldb resolve to the GlobalParser inside liblldb (because it is local), and the cl::opt variables inside lldb-mi resolve inside lldb-mi because the loader cannot see the GlobalParser from liblldb because the symbol isn't exported. This makes two instances of the GlobalParser exist at the same time. This does work around the problem Todd was seeing. Unfortunately, it doesn't actually fix the problem, and if lldb-mi is using cl::opt across the library boundary it will cause subtle and difficult to diagnose bugs. https://reviews.llvm.org/D26093 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
