3 ms·

I've had good luck with a mix of recursive descent -- for high fidelity error messages -- and Pratt at the expression level. I suspect that if expressions got s

by kabdib 6y ago

I've had good luck with a mix of recursive descent -- for high fidelity error messages -- and Pratt at the expression level. I suspect that if expressions got seriously complicated in my applications (e.g., lots of different types with complicated promotion and mixing rules) that I'd write the expression parser in R-D as well. I'm happy giving up some performance for a better chance at finding problems. Who cares if your compiler clocks at a bazillion lines a second if all it can say is "line 101: syntax error" or mumble something indistinct about a reduce-reduce conflict. I'm not writing commercial compilers, just little language processors. And my users need all the help they can get.

cuspycode 6y ago
I too have had great experiences from doing recursive descent. I don't understand why so many people don't like it. The techniques for handling left- and right associativity of infix operators have been well known since many decades, so I am surprised that the original article thinks it is still an issue. But I have met some people who felt uncomfortable by recursive descent parsers that were implemented via recursive procedure calls, just because they didn't quite understand recursive procedure calls in the first place. So maybe that's part of the reason.
- userbinator 6y ago
  My experience has been that those who don't like RD have a mostly theoretical/academic background and are likewise similarly unimpressed with "simple" or "unsophisticated" code in general.
  - afiori 6y ago
    For the author his case against RD is not for the implementation. At the end of the articles he calls for LR reference grammars (I imagine both for fuzzy testing and documentation) to ensure that your perfectly usable RD parser with nice error messages does not miss some edge cases.
- tomp 6y ago
  > I don't understand why so many people don't like it. 2 reasons: - it doesn’t warn you about ambiguous grammar, so you don’t know when you screw up - it requires a lot of boilerplate for operator precedence (one function for each level), in a LR parser generator that’s trivial so I can spend time on more important things

by kabdib 6y ago

cuspycode 6y ago

I too have had great experiences from doing recursive descent. I don't understand why so many people don't like it. The techniques for handling left- and right associativity of infix operators have been well known since many decades, so I am surprised that the original article thinks it is still an issue. But I have met some people who felt uncomfortable by recursive descent parsers that were implemented via recursive procedure calls, just because they didn't quite understand recursive procedure calls in the first place. So maybe that's part of the reason.

userbinator 6y ago
My experience has been that those who don't like RD have a mostly theoretical/academic background and are likewise similarly unimpressed with "simple" or "unsophisticated" code in general.
- afiori 6y ago
  For the author his case against RD is not for the implementation. At the end of the articles he calls for LR reference grammars (I imagine both for fuzzy testing and documentation) to ensure that your perfectly usable RD parser with nice error messages does not miss some edge cases.
tomp 6y ago
> I don't understand why so many people don't like it. 2 reasons: - it doesn’t warn you about ambiguous grammar, so you don’t know when you screw up - it requires a lot of boilerplate for operator precedence (one function for each level), in a LR parser generator that’s trivial so I can spend time on more important things