Add Ripper :on_sp events for Prism.lex_compat and Prism::Translation::Ripper

* Handle line continuations.
* Handle space at the end of file in LexCompat.

Co-authored-by: Earlopain <[email protected]>

Author: eregon

lib/prism.rb

@@ -61,8 +61,7 @@ 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. The main difference is that the
-  # `:on_sp` token is not emitted.
+  # resembles the return value of Ripper::lex.
   #
   # For supported options, see Prism::parse.
   def self.lex_compat(source, **options)
@@ -72,9 +71,8 @@ def self.lex_compat(source, **options)
   # :call-seq:
   #   Prism::lex_ripper(source) -> Array
   #
-  # 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.
+  # This wraps the result of Ripper.lex. It produces almost exactly 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_comment, :on_heredoc_end, :on_embexpr_end
+    # used for :on_sp, :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 :source, :options
+    attr_reader :options
 
-    def initialize(source, **options)
-      @source = source
+    def initialize(code, **options)
+      @code = code
       @options = options
     end
 
@@ -624,12 +624,13 @@ def result
       state = :default
       heredoc_stack = [[]] #: Array[Array[Heredoc::PlainHeredoc | Heredoc::DashHeredoc | Heredoc::DedentingHeredoc]]
 
-      result = Prism.lex(source, **options)
+      result = Prism.lex(@code, **options)
+      source = result.source
       result_value = result.value
       previous_state = nil #: State?
       last_heredoc_end = nil #: Integer?
 
-      bom = source.byteslice(0..2) == "\xEF\xBB\xBF"
+      bom = source.slice(0..2) == "\xEF\xBB\xBF"
 
       result_value.each_with_index do |(token, lex_state), index|
         lineno = token.location.start_line
@@ -763,7 +764,7 @@ def result
                   end_offset += 3
                 end
 
-                tokens << Token.new([[lineno, 0], :on_nl, source.byteslice(start_offset...end_offset), lex_state])
+                tokens << Token.new([[lineno, 0], :on_nl, source.slice(start_offset...end_offset), lex_state])
               end
             end
 
@@ -857,7 +858,88 @@ def result
       # We sort by location to compare against Ripper's output
       tokens.sort_by!(&:location)
 
-      Result.new(tokens, result.comments, result.magic_comments, result.data_loc, result.errors, result.warnings, Source.for(source))
+      # Add :on_sp tokens
+      tokens = add_on_sp_tokens(tokens, source, result.data_loc, bom)
+
+      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)
+      new_tokens = []
+
+      prev_token_state = Translation::Ripper::Lexer::State.new(Translation::Ripper::EXPR_BEG)
+      prev_token_end = 0
+
+      tokens.each do |token|
+        first_token = new_tokens.empty?
+        line, column = token.location
+        start_offset = source.line_to_byte_offset(line) + column
+
+        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)
+          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
+        prev_token_end += 3 if first_token && bom
+      end
+
+      unless data_loc # no trailing :on_sp with __END__ as it is always preceded by :on_nl
+        end_offset = source.source.bytesize
+        end_offset -= 3 if bom
+        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
     end
   end
 

lib/prism/lex_ripper.rb

@@ -19,8 +19,6 @@ 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]

lib/prism/parse_result.rb

@@ -71,9 +71,9 @@ def lines
     end
 
     # Perform a byteslice on the source code using the given byte offset and
-    # byte length.
-    def slice(byte_offset, length)
-      source.byteslice(byte_offset, length) or raise
+    # byte length, or using a Range.
+    def slice(...)
+      source.byteslice(...) or raise
     end
 
     # Converts the line number to a byte offset corresponding to the start of that line

rbi/prism/parse_result.rbi

@@ -25,8 +25,7 @@ class Prism::Source
   sig { returns(T::Array[String]) }
   def lines; end
 
-  sig { params(byte_offset: Integer, length: Integer).returns(String) }
-  def slice(byte_offset, length); end
+  def slice(...); end
 
   sig { params(byte_offset: Integer).returns(Integer) }
   def line(byte_offset); end

test/prism/ruby/ripper_test.rb

@@ -74,6 +74,10 @@ class RipperTest < TestCase
       "strings.txt",
       "whitequark/dedenting_heredoc.txt",
       "whitequark/procarg0.txt",
+      "dos_endings.txt",
+      "seattlerb/str_lit_concat_bad_encodings.txt",
+      "seattlerb/utf8_bom.txt",
+      "unparser/corpus/semantic/dstr.txt",
     ]
 
     Fixture.each_for_current_ruby(except: incorrect | omitted_sexp_raw) do |fixture|
@@ -92,7 +96,7 @@ 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_op], Translation::Ripper::Lexer.new("1 +").lex.map(&:event))
+      assert_equal(%i[on_int on_sp on_op], Translation::Ripper::Lexer.new("1 +").lex.map(&:event))
       assert_raise(SyntaxError) { Translation::Ripper::Lexer.new("1 +").lex(raise_errors: true) }
     end
 
@@ -121,15 +125,17 @@ def assert_ripper_sexp_raw(source)
     def assert_ripper_lex(source)
       prism = Translation::Ripper.lex(source)
       ripper = Ripper.lex(source)
-      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 emits tokens by their order in the code, not in parse order
+      ripper.sort_by! { |elem| elem[0] }
 
       [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[1] == :on_heredoc_end && actual[1] == :on_heredoc_end
+        if expected && actual && expected[1] == :on_heredoc_end && actual[1] == :on_heredoc_end
           expected[3] = actual[3] = nil
         end