Polish URI::Generic parameter signatures

[ksss]

Many parameters of URI::Generic are nilable.

For example, the relative path scheme is nil.

URI.parse('./').scheme
#=> nil

I loosened the type restrictions for some methods.

[ksss]

CI has failed.
Indeed, the following is true.

URI.parse("").send(:userinfo=, 'user:pass')
#=> ["user", "pass"]

But I think the following result should be written as a type

URI.parse("").userinfo = 'user:pass'
#=> "user:pass"

How should I test this case?

[soutaro]

@ksss Wow! 😮 I think the return type of userinfo= method is a tuple, and the value of x.userinfo = y is a behavior of the Ruby language.

[ksss]

@soutaro Thank you for reviewing.

As you said, the Ruby language specification seems to return the argument itself for methods ending with =, regardless of the implementation of the method.

I checked with steep, and this behavior is implemented.

(Maybe here?)
https://github.com/soutaro/steep/blob/89d42e06f5a2c77c55756c97a40e1a6c4b8fb3a4/lib/steep/type_construction.rb#L2739-L2744

However, there seems to be a little bit of confusion since it returns a different type than the documentation.

# Run with the type of `::URI::Generic#userinfo=` before this modification.

# uri: ::URI::Generic
uri = URI::Generic.new(nil, nil, nil, nil, nil, nil, nil, nil, nil)

# ::URI::Generic#userinfo= ~> (::String | nil)
# Sets userinfo, argument is string like 'name:pass'.
# ・(::String? userinfo) -> ::Array[::String | nil]?
userinfo = uri.userinfo = 'user:pass'

# userinfo: (::String | nil)
userinfo

I suggest that the returned type should be the argument type.

And how about the following modification to test this behavior?

diff --git a/test/stdlib/test_helper.rb b/test/stdlib/test_helper.rb
index cfeae24c..aac42bd7 100644
--- a/test/stdlib/test_helper.rb
+++ b/test/stdlib/test_helper.rb
@@ -245,6 +245,10 @@ module TypeAssertions

     begin
       result = spy.wrapped_object.__send__(method, *args, &block)
+      case method.to_s
+      when "[]=", /\w=\Z/
+        result = args.last
+      end
     rescue Exception => exn
       exception = exn
     end

[soutaro]

@ksss

However, there seems to be a little bit of confusion since it returns a different type than the documentation.

It makes sense... I think what we should fix is the implementation of uri library.
How about skipping the test of the method for now?

[soutaro]

@ksss I merged this PR to include the change to RBS 2.0.0, which will be bundled to Ruby 3.1.0! 🙇‍♂️ 🎉

[ksss]

@soutaro Thank you and GJ 🎄