`Prism::Translation::Parser` inconsistency in parsing numeric literals with unary `+` (plus) and `-` (minus)

[Slike9]

There are numeric literals with + and - before, separated by a whitespace:

+ 42
- 42

+ and - are handled differently by Prism::Translation::Parser there:

 $ bundle exec ruby -Ilib -rprism -ve 'p Prism::Translation::ParserCurrent.parse("- 42")'
ruby 3.4.5 (2025-07-16 revision 20cda200d3) +PRISM [x86_64-linux]
s(:int, -42)

 $ bundle exec ruby -Ilib -rprism -ve 'p Prism::Translation::ParserCurrent.parse("+ 42")'
ruby 3.4.5 (2025-07-16 revision 20cda200d3) +PRISM [x86_64-linux]
s(:send,
  s(:int, 42), :+@)

- 42 is considered as a plain literal (seems not correct), while + 42 as an unary method call.

Prism itself considers the both as unary method calls.

I suspect handling - should be fixed.

[Earlopain]

Thanks for the report! This was originally done for rubocop because of #2501. The parser gem has trouble with code like this, for example also whitequark/parser#1016.

The test for that still exists (https://github.com/rubocop/rubocop/blob/aeff12f915d0d8c1f544492a5537049766b50b75/spec/rubocop/cop/style/numeric_literals_spec.rb#L121-L131). That said, the autocorrect that rubocop performs is not correct (because it acts on the faulty parser ast). While it is technically compatible, we should not mimick that.

More specifically:

class Integer
  def -@
    puts "Called unary -"
  end
end

a = -
123
a = -123 # Nothing printed

@koic do you have a problem with correcting the ast that prism produces here? You'd just have to update that one test expectation I linked above and skip the test on the parser gem when the next prism version is released.