Fix invalid sig generated for anonymous block param (&)

[prattik-wav]

Fixes #928

Problem

rbs_comments_to_sorbet_sigs generates invalid Ruby when a method uses Ruby 3.1+ anonymous block forwarding (def foo(&); end). The generated sig contains &block: inside params(), which is a syntax error:

# Input
#: (String) ?{ (String) -> void } -> String
def bar(request, &); end

# Generated (broken)
sig { params(request: String, &block: ::T.nilable(::T.proc.params(arg0: String).void)).returns(String) }
def bar(request, &); end

Named block params (&block) were already translated correctly.

Root Cause

Prism::BlockParameterNode#name is nil for anonymous &. The RBI::RBS::MethodTypeTranslator doesn't distinguish this case and falls back to the default name &block, which renders as &block: inside params() — a syntax error because & is not valid in a keyword argument position.

Fix

After translator.result produces the sig, check whether the original def_node has an anonymous block parameter (a BlockParameterNode with nil name). If so, strip the leading & from any block param name in the generated sig, producing the valid block: form.

# Generated (fixed)
sig { params(request: String, block: ::T.nilable(::T.proc.params(arg0: String).void)).returns(String) }
def bar(request, &); end

Testing

• Added test_translate_to_rbi_anonymous_block_param — covers the broken case from the issue
• Added test_translate_to_rbi_named_block_param_unchanged — regression guard confirming named &block still works
• Verified the fixed output compiles cleanly with RubyVM::InstructionSequence.compile

Fixes #928

[paracycle]

I don't think this generated code is correct either: https://sorbet.run/#%23%20typed%3A%20true%0Aclass%20Foo%0A%20%20extend%20T%3A%3ASig%0A%0A%20%20sig%20%7B%20params%28request%3A%20String%2C%20block%3A%20%3A%3AT.nilable%28%3A%3AT.proc.params%28arg0%3A%20String%29.void%29%29.void%20%7D%0A%20%20def%20bar%28request%2C%20%26%29%3B%20nil%3B%20end%0Aend

I think it should be:

Suggested change

	sig { params(request: String, block: ::T.nilable(::T.proc.params(arg0: String).void)).returns(String) }
	def bar(request, &); end
	sig { params(request: String, "&": ::T.nilable(::T.proc.params(arg0: String).void)).returns(String) }
	def bar(request, &); end

as you can see working here: https://sorbet.run/#%23%20typed%3A%20true%0Aclass%20Foo%0A%20%20extend%20T%3A%3ASig%0A%0A%20%20sig%20do%20%0A%20%20%20%20type_parameters%28%3AR%29%0A%20%20%20%20%20%20.params%28%22%26%22%3A%20%3A%3AT.proc.returns%28T.type_parameter%28%3AR%29%29%29%0A%20%20%20%20%20%20.returns%28T.type_parameter%28%3AR%29%29%0A%20%20end%0A%20%20def%20bar%28%26%29%3B%20T.unsafe%28nil%29%20end%0Aend%0A%0Afoo%20%3D%20Foo.new%0Aresult%20%3D%20foo.bar%20do%20%7Carg0%7C%0A%20%20%3Asymbol%0Aend%0A%0AT.reveal_type%28result%29

[paracycle]

I am not sure if this is the correct place to fix this problem. We need to solve it at the source.

[amomchilov]

Yeah the correct place to fix this is in the RBI gem that Spoom uses for this translation:

https://github.com/Shopify/rbi/blob/4a4f30b9ca743f7723ce451dc06a9b7cf62a42b0/lib/rbi/rbs/method_type_translator.rb#L46

[amomchilov]

This is overkill, we don't need to waste time compiling it. Perhaps assert Prism.parse_success?(res) at most, but we match the exact expected source we want, so there's no need to confirm this here

Suggested change

	# Must also be valid Ruby
	assert RubyVM::InstructionSequence.compile(res)

[amomchilov]

Yeah the correct place to fix this is in the RBI gem that Spoom uses for this translation:

https://github.com/Shopify/rbi/blob/4a4f30b9ca743f7723ce451dc06a9b7cf62a42b0/lib/rbi/rbs/method_type_translator.rb#L46