I reworked the musttail patchkit to make it work without optimization and also give better error messages and fixed some bugs and increased the test coverage.
The no optimization support is not perfect: it cannot handle returning small structures which would need to be allocated to registers first. There are also still cases when musttail cannot give an accurate error message (mainly because of complex control flow making tree-tailcall giving up). But overall it works significantly better. Jason's feedback on the C++ parser is also addressed.