Hacker News Reader

Could you explain in more detail what you mean about LALR error messages? I haven't used the parser generators you mention but I've implemented LALR parsing before, the error reporting situation doesn't seem drastically different from recursive descent to me. You can report the location where the parse failed, what was recognized before that point, what tokens would have been allowed to follow.

Is the problem that these specific tools (yacc/bison) don't report errors well? Or is the issue you're talking about that LR grammars will fail later on than LL grammars in general? Or the greater ambiguity (e.g. more possible lookaheads due to the greater parsing power)?

A bottom up parser such as LALR knows what it has built successfully, and knows that the next token makes no sense in the current state. So you have stuff in the parse stack that is headed somewhere but you don't know where, and you know you are at the earliest point you can detect a syntax error. Unless you rummage around in the parse stack for clues, there really isn't much you can do to make a good message, and the rummaging is often ad-hoc.

When parsing top-down as in recursive descent, you know what higher level syntactic structure you are in the middle of constructing, and anything successfully parsed so far is probably slotted into the AST in a meaningful way, so you simply have a lot more context to go on when constructing a message, and that context is an intuitive AST walk away.

I'll also add you can use two parsers: one that's super fast on code that's correctly written; if that fails, one that makes error messages easy to handle. sklogic said he used that strategy in his tools for program analysis. The extra code is negligible. I don't know how much extra time or maintenance burden but I figured marginal versus cost of handling errors in first place.

My FastParse (https://www.lihaoyi.com/fastparse/) parser combinator library does this; by default it runs without error logging, just giving you an error position, but if something fails you can ask it to re-do the parse keeping track of additional metadata to give you a nice error message with what tokens could have succeeded and a stack trace telling you why wants those tokens.

Tracing errors slows things down about 2x, which is why it isnt on by default, on the assumption that most parses are successful

> You can report the location where the parse failed, what was recognized before that point, what tokens would have been allowed to follow.

This is a garbage 'computer says no' kind of error message.

It doesn't tell you why those tokens would have been allowed to follow, why it is that those are the only tokens allowed to follow, why the token you used is not one of the allowed one, or what you could have more likely been intending to write to make it correct.

In a recursive descent parser you have the freedom to use more context and logic to generate your error messages.

What context or data would constitute a helpful error message in your opinion? Do you have an example of a recursive-descent parser that generates particularly good error messages?

A great example is if you forget to put 'end' at the end of a function at the end of a file in languages that have that.

Your idea of a good parser error message is being told that the end of file was unexpected and being given a list of a hundred possible tokens that could start another statement, including 'end' buried somewhere in the list, because that's what's valid there.

My idea of a good parser error message would be in a recursive descent parser looking for the 'end' token, not seeing it, seeing that it's end of file instead and saying 'missing 'end' or a new statement at end of file'.

See elm and rust for examples of compilers with good error messages.

Or clang.

> The problem with yacc/bison-style LALR parser generators is that you end up with an LALR parser. Which works great on syntactically correct code, but trying to get a reasonable error message out of one is about as much fun as repeatedly poking yourself in the eye with a sharp stick.

The standard ways of getting error messages from LR parsers are pretty bad, but there is a long line of work that's tried to make it better. If you'll forgive the immodesty on my part, I've been working on updating and fixing that work. There's a draft paper at https://arxiv.org/abs/1804.07133 and a beta-ish implementation at https://github.com/softdevteam/grmtools/, so you can see what you think!

[EDIT: fixed formatting]

For a really nice solution to the error message problem, see this recent strangeloop talk: https://www.youtube.com/watch?v=Jes3bD6P0To

Basically it uses the parse tree disambiguation from the GLR parser to look for the most likely mistake the user made - it's very clever.