Update parser so that void type is rejected depending on the context

[soutaro]

No description provided.

[Morriar]

We're getting more places where we specify what's allowed: allow_selfq, allow_eof, allow_void. Since we already pass the parser everywhere, I wonder if we should add these options as context rather than passing them individually to each method:

parser.context.allow_void

[soutaro]

@Morriar I'd like to keep allow_void (and allow_selfq) as dedicated parameter. They are not part of the parser’s global state, but rather values that change frequently during parsing. So I don’t think it’s reasonable to write code that preserves and restores the context, like:

parser_context old_context = parser.context;
parser.context.allow_void = true;

parse_type(parser);

parser.context = old_context;

I’m planning to introduce two more flags for parsing self and class/instance types (#2336). These flags are static during parsing, so I agree that making them attributes of the parser makes sense.

[soutaro]

These flags are static during parsing

Oops — they are constant while parsing a type, but they can change when parsing an entire signature.

[ksss]

The parse_type_arg is also used for parse_method_type and parse_signature.

Regarding the name allow_void, it is appropriate for type parsing, but for signature parsing, I think there is a possibility that it could be interpreted as allowing or disallowing all uses of void, regardless of the context in which it is used.

For example, looks disallowed to me.

RBS::Parser.parse_method_type("() -> void", allow_void: false)

What do you think?

[soutaro]

It makes sense. I added two more arg types, parse_method_type_arg and parse_signature_arg, and use them for parse_method_type and parse_signature.

The two structs are identical currently, but a few attributes will be added to parse_method_type_arg to support options for self/instance/class type parsing.