Python 3.12 Peg Parser Grammar has a small problem

[hasii2011]

Pycharm and Python 3.12 in general accept lines in this format

class MyMetaBaseWxCommand(BaseWxCommandMeta, ABCMeta):  # type: ignore
    pass

However, the Python parser generated by the above burps on the '# type: ignore' comment. Says it is a syntax error. Not sure why

[kaby76]

The grammar is scraped from the pegen grammar 3.12.1. If it doesn't parse, it's caused by one of the following:

• The pegen grammar is incorrect.
• The lexical structure is not being implemented correctly from the doc.
• The lexical structure not being described correctly in the documentation.

Pegen does not take a description for the lexical structure. It's backwards captured from an implementation then placed into the Python documentation--something that is really bad practice. We have to read the source code for the Python3 compiler and find out which.

[hasii2011]

Ok, thanks for the quick response. Was hoping it was a quick fix. I am not familiar with g4 files, mainly just a consumer.

[RobEin]

The lexical analysis documentation does not mention the generation of the TYPE_COMMENT token.
Maybe another documentation describes it somewhere, but I haven't found it so far.

The solution for the comment in the example (# type: ignore) could be that the lexer would recognize it as a plain COMMENT (or perhaps as a hidden TYPE_COMMENT).
In other words, the tokenizer would be statement-sensitive.
I'm afraid that this cannot be implemented in the lexer and that's probably why they don't write about it in the lexical analysis documentation.
I still have to think about that.

Until then, I temporarily set the TYPE_COMMENT tokens to hidden in my own repository.
This way, there is no parsing for type comments, but no errors are generated either.

[kaby76]

In other words, the tokenizer would be statement-sensitive.

We might be able to define an Antlr4 "lexer mode" to work around a parser-state-dependent lex, but "lexer modes" are basically hacks for the real deal. It would be best to have a parser-state aware lexer option for Antlr5. @ericvergnaud

[hasii2011]

So I generated new lexer/parser files this morning and verified using the modified file provided by @RobEin ; They worked great and got me around this issue with the mypy commented files. This resolves the issue for me now as I don't look at that construct in my visitor code

Thanks very much for this.