`Prism::Translation::Parser` incompatibility AST for `foo.~@ bar`

[koic]

AST regression for foo.~@ bar in Prism 1.7.0 (translation Parser33):

Expected Behavior

Prism 1.6.0 returns "~@":

$ bundle exec ruby -rprism -rprism/translation/parser33 -ve 'p Prism::Translation::Parser33.parse("foo.~@ bar")'
ruby 4.0.0dev (2025-11-08T11:52:39Z master 4365c4fb6b) +PRISM [arm64-darwin24]
s(:send,
  s(:send, nil, :foo), :"~@",
  s(:send, nil, :bar))

and Parser gem returns "~@" too:

$ bundle exec ruby -rparser/ruby33 -ve 'p Parser::Ruby33.parse("foo.~@ bar")'
ruby 4.0.0dev (2025-11-08T11:52:39Z master 4365c4fb6b) +PRISM [arm64-darwin24]
s(:send,
  s(:send, nil, :foo), :"~@",
  s(:send, nil, :bar))

Actual Behavior

Prism 1.7.0 returns "~":

$ bundle exec ruby -rprism -rprism/translation/parser33 -ve 'p Prism::Translation::Parser33.parse("foo.~@ bar")'
ruby 4.0.0dev (2025-11-08T11:52:39Z master 4365c4fb6b) +PRISM [arm64-darwin24]
s(:send,
  s(:send, nil, :foo), :~,
  s(:send, nil, :bar))

This breaks compatibility with Parser's AST for unary ~@ calls. And "foo.!@ bar" has the same issue as "foo.~@ bar".

[Earlopain]

I'm fairly sure this is a bug in the parser gem whitequark/parser#1084. Also see #3694. Ruby drops the @ from the method call. This is similar to :~@:

$ ruby-parse --33 -e ":~@"
(sym :~)

$ ruby -ve "puts :~@"
ruby 3.4.7 (2025-10-08 revision 7a5688e2a2) +PRISM [x86_64-linux]
~

Here the parser gem correctly drops the @. This seems consistent with that?

[koic]

The Parser gem AST for the example in the title is shown below. The result is :"~@".

$ ruby-parse --33 -e "foo.~@ bar"                                                         
(send
  (send nil :foo) :"~@"
  (send nil :bar))

[Earlopain]

Yeah, I understand. This is what I'm talking about:

foo = Object.new
def foo.singleton_method_added(name)
  puts "Added `#{name}`"
end

def foo.~@
  puts "called"
end

# Both call the same method
foo.~@
~foo

$ ruby -v --parser=parse.y test.rb # Need to use parse.y, only fixed for prism on ruby-dev
ruby 3.4.8 (2025-12-17 revision 995b59f666) [x86_64-linux]
Added `singleton_method_added`
Added `~`
called
called

Basically the same as whitequark/parser#486 / whitequark/parser#488. That's why I think this is a parser bug.

[koic]

Ah, I see. Unlike +@ and -@, for ! and ~ it may be correct to omit the trailing @. I'll think about how to handle the incompatibility. In any case, it seems like this should be resolved outside of Prism. Thank you.

[Earlopain]

Yeah, it's weird syntax. I didn't think that it would actually affect rubocop. If it would help, I wouldn't mind trying to fix this in the parser gem, let me know.