Prism::Translation::Ripper.lex emits :on_rbrace where MRI Ripper emits :on_embexpr_end

[ollym]

Reproduction

source = %q!a( id: nil }!

require "ripper"
require "prism"
require "prism/translation/ripper"

puts RUBY_DESCRIPTION
puts "Prism #{Prism::VERSION}"

p Ripper.lex(source)
p Prism::Translation::Ripper.lex(source)
p Prism.lex_compat(source).value

Expected

MRI 4.0.5 native Ripper.lex emits the unmatched closing brace as:

[[1, 11], :on_embexpr_end, "}", END]

Actual

Prism::Translation::Ripper.lex emits it as:

[[1, 11], :on_rbrace, "}", END]

Prism.lex_compat(source).value also emits :on_rbrace.

Environment

ruby 4.0.5 (2026-05-20 revision 64336ffd0e) +PRISM [arm64-darwin22] Prism 1.9.0

I also checked Prism 1.8.1, which showed the same :on_rbrace behavior.

Notes

The input is invalid Ruby, and both MRI and Prism raise with syntax errors enabled:

Ripper.lex(source, raise_errors: true)
# SyntaxError: syntax error, unexpected '}', expecting ')'

Prism::Translation::Ripper.lex(source, raise_errors: true)
# SyntaxError: unexpected '}'; expected a `)` to close the arguments

So this may be an invalid-input compatibility gap rather than a parser correctness bug.

I originally reported this against TruffleRuby because TruffleRuby 34.0.1’s ::Ripper.lex returns :on_rbrace here. @eregon pointed out that TruffleRuby uses Prism::Translation::Ripper for ::Ripper, so this appears to belong upstream in Prism.

This was found while removing obsolete TruffleRuby/Ripper guards from Haml after TruffleRuby added Ripper support: haml/haml#1210

[Earlopain]

Thanks for the report.

The translators are not tested very well against invalid syntax and I suspect the amount of differences are quite large. Ripper lexing a plain } as on_embexpr_end also doesn't look that correct to me. In prism that is just a plain bracket, which also is why the result differs. It's interesting that it currently works that way but personally I would put that under relying to much on internal parse state.

For your change in haml, it looks like you can simply stop modifying the source and only check for on_lbrace/on_rbrace for ripper and the translator (and also start the count at 0, not 1). I suspect the way it currently is was to work around some other ripper particularity that no longer applies.

[ollym]

Thanks for the feedback, implemented haml/haml@5021ed3

I'll close this ticket then.