Ross Patterson

This is ANTLR bug #[2677](https://github.com/antlr/antlr4/issues/2677), originally opened on 2019-10-26. I ran into it during my work to improve parse-tree- and errors-file testing, and verified that it was still a bug...

> Target agnostic features should be tested in ANTLR reposiroty. I'm not sure what @KvanTTT meant by this comment, but I interpret it to mean "_the `antlr4` repo should test...

It seems to me that we have multiple needs for the test suite. All of them, of course, are solved by the _test-all-grammars-every-time_ system. What @kaby76 seems to be building...

> My general opinion is that lark.lark should reflect the syntax as defined by the BNF inside of load_grammar.py and not the semantic checks implemented on top of that (and...

I missed two lines that I should have deleted in the test files, which I used for before/after testing (`... 'grammars/lark.lark-ORIG' ...`). I'll remove them.

> I missed two lines that I should have deleted in the test files, which I used for before/after testing (`... 'grammars/lark.lark-ORIG' ...`). I'll remove them. Fixed in 9493f81e9eab6dbe42096938bbcfc1f0cf5a2135.

> As for what to do with lark.lark, in my mind it's still an open question, and depends heavily on what are its main uses. Is it to parse and...

> > I think having an inaccurate reference grammar for human purposes > > @RossPatterson Can you explain which option this refers to? I don't think I advocated at any...

> @RossPatterson Would you be okay adapting your PR to solution 2? > > That means getting lark.lark as close as possible to load_grammar, and adding a validating visitor for...

> To note: lark.lark should probably still use ebnf and not restrict itself to bnf like the hardcoded grammar does, I'm personally biased in favor of BNF, because it's so...