For my project, I found sticking to LALR(1) quite beneficial. Not that there's anything special about LALR(1): you could use LL(k) instead if you prefer. Howe

For my project, I found sticking to LALR(1) quite beneficial. Not that there's anything special about LALR(1): you could use LL(k) instead if you prefer. However, by sticking to a small, deterministic syntax (and making the parser generator complain loudly in case of ambiguity), I can easily find edge cases I've never thought about before it comes around to bite me. And, in most cases, it's trivial to slightly modify the syntax so that the undesirable ambiguity does not arise. It's the equivalent of static typing for syntax: it takes effort to make bison run successfully, but when it does, I know that my grammar is free of ambiguity.