If the other transformed versions of the Riemann solvers are working, then there is probably some static error in the common interface kernel.
I generally don’t find debuggers to be very useful, also what pyfr is doing is turning the mako into a hardware specific language, such as cuda, and then running that. So a python debugger probably would struggle with that.
Take another look at your mako macro for the interface flux and see if you can spot the bug. If not, you can run with -b openmp and print the source to screen. The source for the interface flux can then be copied into a .c file which you can try compiling with gcc. That should then give you some error messages. However, it takes a bit of unpicking to workout how the errors from gcc translate to errors in the mako.
For example you can add a print(src) here: PyFR/provider.py at 16afb73b9fd70817c146c43e4966f848e1a084af · PyFR/PyFR · GitHub