[kazinator]

The main advantage of recursive descent is that since parser rules correspond to functions in the host language, you can put a breakpoint on them in the debugger and understand how you got there. You can't do that with a table-driven LALR(1) parser like Bison.

I think you cover that with "debuggability" remark.

Here is something else. Yacc/Bison parsing uses its own stack for the symbols. Whereas recursive descent uses the regular run-time stack.

Pop quiz: which stack is visible to your garbage collector, and which isn't?

In TXR Lisp, which uses a Bison/Byacc generated parser, GC is suspended while yyparse() executes. Otherwise it would prematurely collect anything that is only stored in the Yacc stack, and so there would have to be a GC hook to traverse that stack (which would depend on undocumented features of the generated code, and likely have to have some separate coding for Bison versus Byacc.)