The $ prefix is ambiguous

[gamesaucer]

Steps to reproduce: Parse the following grammar:

$Start = $Part
$Part = .

Expected result: The validity of this Peggy grammar is decided by whether rule names can start with $.
Actual result: It is not a valid Peggy grammar because rule Part is not defined.

Because Peggy allows the dollar sign $ to start rule names AND has a $ prefix operator, one of the following must be true:

1. The first $ is always parsed as an operator. In this case, a rule cannot reliably be matched simply by writing its name. There is also no way to match $Part without returning the text rather than the match. This is the current implementation.
2. The first $ is always parsed as part of the identifier. In this case, to return the matched text of a nonterminal, a space is required before the nonterminal to disambiguate between $ Part (return text of rule Part) and $Part (return match of rule $Part), or the expression needs to be wrapped in parentheses. This makes it inconsistent with the operator's behaviour before terminals.
3. The $ is parsed as part of the identifier if that identifier exists. In this case, adding a rule called Part would break existing references to a rule called $Part or vice versa. It would also it becomes ambiguous what a $ does before a nonterminal and require an extra parsing pass to determine how it is interpreted.

All those options seem undesirable to me. Instead, I'd advocate for either disallowing $ at the start of rule names, or changing the $ operator to something that cannot appear at the start of an identifier. Unfortunately, any alteration to this behaviour would be a breaking change. So if it's decided this won't be fixed, I understand why that's the case.

[hildjj]

This is a valid issue, and I'm surprised it hasn't come up before. I don't have a better solution that disallowing $ as a start character in identifiers.

@Mingun any insights?

[Mingun]

Yes, I think we should forbid that symbol in identifiers (only at start or everywhere if something$something could be treated ambiguously). It's amazing how it hasn't been noticed before

[hildjj]

I think $ is fine in the middle of identifier.

top = fb

fb = $foo$bar

foo$bar = foo $bar

$foo$bar = 'baz'

foo = 'foo'

bar = 'b' 'a'+ 'r'

this matches "foobaaar", but not "baz".