Revert "[ruby/prism] Add Ripper :on_sp events for Prism.lex_compat and Prism::Translation::Ripper"

This reverts commit 35a7b51.

This broke syntax_suggest.

https://github.com/ruby/ruby/actions/runs/21167011751/job/60874111912

Author: hsbt

lib/prism.rb

@@ -61,7 +61,8 @@ def initialize(version)
   #   Prism::lex_compat(source, **options) -> LexCompat::Result
   #
   # Returns a parse result whose value is an array of tokens that closely
-  # resembles the return value of Ripper::lex.
+  # resembles the return value of Ripper::lex. The main difference is that the
+  # `:on_sp` token is not emitted.
   #
   # For supported options, see Prism::parse.
   def self.lex_compat(source, **options)
@@ -71,8 +72,9 @@ def self.lex_compat(source, **options)
   # :call-seq:
   #   Prism::lex_ripper(source) -> Array
   #
-  # This wraps the result of Ripper.lex. It produces almost exactly the
-  # same tokens. Raises SyntaxError if the syntax in source is invalid.
+  # This lexes with the Ripper lex. It drops any space events but otherwise
+  # returns the same tokens. Raises SyntaxError if the syntax in source is
+  # invalid.
   def self.lex_ripper(source)
     LexRipper.new(source).result # steep:ignore
   end

lib/prism/lex_compat.rb

@@ -226,7 +226,7 @@ def state
     end
 
     # Tokens where state should be ignored
-    # used for :on_sp, :on_comment, :on_heredoc_end, :on_embexpr_end
+    # used for :on_comment, :on_heredoc_end, :on_embexpr_end
     class IgnoreStateToken < Token
       def ==(other) # :nodoc:
         self[0...-1] == other[0...-1]
@@ -611,10 +611,10 @@ def self.build(opening)
     BOM_FLUSHED = RUBY_VERSION >= "3.3.0"
     private_constant :BOM_FLUSHED
 
-    attr_reader :options
+    attr_reader :source, :options
 
-    def initialize(code, **options)
-      @code = code
+    def initialize(source, **options)
+      @source = source
       @options = options
     end
 
@@ -624,14 +624,12 @@ def result
       state = :default
       heredoc_stack = [[]] #: Array[Array[Heredoc::PlainHeredoc | Heredoc::DashHeredoc | Heredoc::DedentingHeredoc]]
 
-      result = Prism.lex(@code, **options)
-      source = result.source
+      result = Prism.lex(source, **options)
       result_value = result.value
       previous_state = nil #: State?
       last_heredoc_end = nil #: Integer?
-      eof_token = nil
 
-      bom = source.slice(0, 3) == "\xEF\xBB\xBF"
+      bom = source.byteslice(0..2) == "\xEF\xBB\xBF"
 
       result_value.each_with_index do |(token, lex_state), index|
         lineno = token.location.start_line
@@ -743,7 +741,6 @@ def result
 
             Token.new([[lineno, column], event, value, lex_state])
           when :on_eof
-            eof_token = token
             previous_token = result_value[index - 1][0]
 
             # If we're at the end of the file and the previous token was a
@@ -766,7 +763,7 @@ def result
                   end_offset += 3
                 end
 
-                tokens << Token.new([[lineno, 0], :on_nl, source.slice(start_offset, end_offset - start_offset), lex_state])
+                tokens << Token.new([[lineno, 0], :on_nl, source.byteslice(start_offset...end_offset), lex_state])
               end
             end
 
@@ -860,89 +857,7 @@ def result
       # We sort by location to compare against Ripper's output
       tokens.sort_by!(&:location)
 
-      # Add :on_sp tokens
-      tokens = add_on_sp_tokens(tokens, source, result.data_loc, bom, eof_token)
-
-      Result.new(tokens, result.comments, result.magic_comments, result.data_loc, result.errors, result.warnings, source)
-    end
-
-    def add_on_sp_tokens(tokens, source, data_loc, bom, eof_token)
-      new_tokens = []
-
-      prev_token_state = Translation::Ripper::Lexer::State.cached(Translation::Ripper::EXPR_BEG)
-      prev_token_end = bom ? 3 : 0
-
-      tokens.each do |token|
-        line, column = token.location
-        start_offset = source.line_to_byte_offset(line) + column
-        # Ripper reports columns on line 1 without counting the BOM, so we adjust to get the real offset
-        start_offset += 3 if line == 1 && bom
-
-        if start_offset > prev_token_end
-          sp_value = source.slice(prev_token_end, start_offset - prev_token_end)
-          sp_line = source.line(prev_token_end)
-          sp_column = source.column(prev_token_end)
-          # Ripper reports columns on line 1 without counting the BOM
-          sp_column -= 3 if sp_line == 1 && bom
-          continuation_index = sp_value.byteindex("\\")
-
-          # ripper emits up to three :on_sp tokens when line continuations are used
-          if continuation_index
-            next_whitespace_index = continuation_index + 1
-            next_whitespace_index += 1 if sp_value.byteslice(next_whitespace_index) == "\r"
-            next_whitespace_index += 1
-            first_whitespace = sp_value[0...continuation_index]
-            continuation = sp_value[continuation_index...next_whitespace_index]
-            second_whitespace = sp_value[next_whitespace_index..]
-
-            new_tokens << IgnoreStateToken.new([
-              [sp_line, sp_column],
-              :on_sp,
-              first_whitespace,
-              prev_token_state
-            ]) unless first_whitespace.empty?
-
-            new_tokens << IgnoreStateToken.new([
-              [sp_line, sp_column + continuation_index],
-              :on_sp,
-              continuation,
-              prev_token_state
-            ])
-
-            new_tokens << IgnoreStateToken.new([
-              [sp_line + 1, 0],
-              :on_sp,
-              second_whitespace,
-              prev_token_state
-            ]) unless second_whitespace.empty?
-          else
-            new_tokens << IgnoreStateToken.new([
-              [sp_line, sp_column],
-              :on_sp,
-              sp_value,
-              prev_token_state
-            ])
-          end
-        end
-
-        new_tokens << token
-        prev_token_state = token.state
-        prev_token_end = start_offset + token.value.bytesize
-      end
-
-      unless data_loc # no trailing :on_sp with __END__ as it is always preceded by :on_nl
-        end_offset = eof_token.location.end_offset
-        if prev_token_end < end_offset
-          new_tokens << IgnoreStateToken.new([
-            [source.line(prev_token_end), source.column(prev_token_end)],
-            :on_sp,
-            source.slice(prev_token_end, end_offset - prev_token_end),
-            prev_token_state
-          ])
-        end
-      end
-
-      new_tokens
+      Result.new(tokens, result.comments, result.magic_comments, result.data_loc, result.errors, result.warnings, Source.for(source))
     end
   end
 

lib/prism/lex_ripper.rb

@@ -19,6 +19,8 @@ def result
 
       lex(source).each do |token|
         case token[1]
+        when :on_sp
+          # skip
         when :on_tstring_content
           if previous[1] == :on_tstring_content && (token[2].start_with?("\#$") || token[2].start_with?("\#@"))
             previous[2] << token[2]

test/prism/fixtures/bom_leading_space.txt

@@ -39,8 +39,6 @@ class RipperTest < TestCase
 
     # Skip these tests that we haven't implemented yet.
     omitted_sexp_raw = [
-      "bom_leading_space.txt",
-      "bom_spaces.txt",
       "dos_endings.txt",
       "heredocs_with_fake_newlines.txt",
       "heredocs_with_ignored_newlines.txt",
@@ -94,7 +92,7 @@ def test_lexer
       assert_equal(expected, lexer.parse[0].to_a)
       assert_equal(lexer.parse[0].to_a, lexer.scan[0].to_a)
 
-      assert_equal(%i[on_int on_sp on_op], Translation::Ripper::Lexer.new("1 +").lex.map(&:event))
+      assert_equal(%i[on_int on_op], Translation::Ripper::Lexer.new("1 +").lex.map(&:event))
       assert_raise(SyntaxError) { Translation::Ripper::Lexer.new("1 +").lex(raise_errors: true) }
     end
 
@@ -123,17 +121,15 @@ def assert_ripper_sexp_raw(source)
     def assert_ripper_lex(source)
       prism = Translation::Ripper.lex(source)
       ripper = Ripper.lex(source)
-
-      # Prism emits tokens by their order in the code, not in parse order
-      ripper.sort_by! { |elem| elem[0] }
+      ripper.reject! { |elem| elem[1] == :on_sp } # Prism doesn't emit on_sp
+      ripper.sort_by! { |elem| elem[0] } # Prism emits tokens by their order in the code, not in parse order
 
       [prism.size, ripper.size].max.times do |i|
         expected = ripper[i]
         actual = prism[i]
-
         # Since tokens related to heredocs are not emitted in the same order,
         # the state also doesn't line up.
-        if expected && actual && expected[1] == :on_heredoc_end && actual[1] == :on_heredoc_end
+        if expected[1] == :on_heredoc_end && actual[1] == :on_heredoc_end
           expected[3] = actual[3] = nil
         end