Conversation

Phew.. It was a lot of work to fix all the bugs, reduce compile times to reasonable amounts, and actually implement the parser, but I finally have a full MinCaml parser implemented using my own LR parser generator: github.com/osa1/mincaml/b Not the final product!

Brendan Zabarauskas

@brendanzab

May 16, 2021

Oooh nice! How does your parser generator compare to LALRPOP? Other than being able to be used via a procedural macro?

sinan

@_osa1

May 16, 2021

LALRPOP: Mature, good error messages, generates LALR(1). Mine: Prototype, terrible error messages, generates canonical LR(1). I'm using mine as a research device and for learning. I will be implementing GLR, good errors + error recovery etc.

Brendan Zabarauskas

@brendanzab

May 16, 2021

Nice! Very cool! And thanks for the info. Curious if you've seen this post, and if you have any thoughts: matklad.github.io/2018/06/06/mod

sinan

@_osa1

May 16, 2021

Very interesting post! I think I agree that lossless parse trees are a good idea and semantic actions feel like a "layering violation". tree-sitter gets this right. I should probably implement a mode in my parser generator to disallow semantic actions and return lossless trees.

Brendan Zabarauskas

@brendanzab

Replying to

@_osa1

Yeah, I guess I still wish I could have my grammar separate from the implementation, and include that as part of my specification… and then maybe 'layer on' some additional stuff somehow to make it more user-friendly… but yeah I dunno what that would look like.

8:29 AM · May 16, 2021Twitter Web App