Fully migrate to prism

It mostly continues to rely on tokens.
But for a few things like endless method defs
and multiline method continuations
it uses AST. These are either very difficult
or not possible to find just by checking tokens.

Because of multiline method calls, comments now don't need to be trimmed anymore.

Co-authored-by: Schneems <richard.schneeman+foo@gmail.com>

Author: Earlopain

CHANGELOG.md

@@ -1,6 +1,7 @@
 ## HEAD (unreleased)
 
-- Changed: Changed: Minimum supported Ruby version is now 3.3. (https://github.com/ruby/syntax_suggest/pull/246)
+- Changed: Minimum supported Ruby version is now 3.3. (https://github.com/ruby/syntax_suggest/pull/246)
+- Changed: Use native prism to analyse. (https://github.com/ruby/syntax_suggest/pull/251)
 
 ## 2.0.3
 

lib/syntax_suggest/api.rb

@@ -9,9 +9,6 @@
 
 # Prism is the new parser, replacing Ripper
 require "prism"
-# We need Ripper loaded for `Prism.lex_compat` even if we're using Prism
-# for lexing and parsing
-require "ripper"
 
 module SyntaxSuggest
   # Used to indicate a default value that cannot
@@ -188,7 +185,6 @@ def self.valid?(source)
 require_relative "clean_document"
 
 # Helpers
-require_relative "lex_all"
 require_relative "code_line"
 require_relative "code_block"
 require_relative "block_expand"
@@ -200,3 +196,5 @@ def self.valid?(source)
 require_relative "pathname_from_message"
 require_relative "display_invalid_blocks"
 require_relative "parse_blocks_from_indent_line"
+require_relative "visitor"
+require_relative "token"

lib/syntax_suggest/clean_document.rb

@@ -66,27 +66,9 @@ module SyntaxSuggest
   #
   # All of these problems are fixed by joining the whole heredoc into a single
   # line.
-  #
-  # ## Comments and whitespace
-  #
-  # Comments can throw off the way the lexer tells us that the line
-  # logically belongs with the next line. This is valid ruby but
-  # results in a different lex output than before:
-  #
-  #     1 User.
-  #     2   where(name: "schneems").
-  #     3   # Comment here
-  #     4   first
-  #
-  # To handle this we can replace comment lines with empty lines
-  # and then re-lex the source. This removal and re-lexing preserves
-  # line index and document size, but generates an easier to work with
-  # document.
-  #
   class CleanDocument
     def initialize(source:)
-      lines = clean_sweep(source: source)
-      @document = CodeLine.from_source(lines.join)
+      @document = CodeLine.from_source(source)
     end
 
     # Call all of the document "cleaners"
@@ -110,62 +92,6 @@ def to_s
       @document.join
     end
 
-    # Remove comments
-    #
-    # replace with empty newlines
-    #
-    #     source = <<~'EOM'
-    #       # Comment 1
-    #       puts "hello"
-    #       # Comment 2
-    #       puts "world"
-    #     EOM
-    #
-    #     lines = CleanDocument.new(source: source).lines
-    #     expect(lines[0].to_s).to eq("\n")
-    #     expect(lines[1].to_s).to eq("puts "hello")
-    #     expect(lines[2].to_s).to eq("\n")
-    #     expect(lines[3].to_s).to eq("puts "world")
-    #
-    # Important: This must be done before lexing.
-    #
-    # After this change is made, we lex the document because
-    # removing comments can change how the doc is parsed.
-    #
-    # For example:
-    #
-    #     values = LexAll.new(source: <<~EOM))
-    #       User.
-    #         # comment
-    #         where(name: 'schneems')
-    #     EOM
-    #     expect(
-    #       values.count {|v| v.type == :on_ignored_nl}
-    #     ).to eq(1)
-    #
-    # After the comment is removed:
-    #
-    #     values = LexAll.new(source: <<~EOM))
-    #       User.
-    #
-    #         where(name: 'schneems')
-    #     EOM
-    #     expect(
-    #      values.count {|v| v.type == :on_ignored_nl}
-    #    ).to eq(2)
-    #
-    def clean_sweep(source:)
-      # Match comments, but not HEREDOC strings with #{variable} interpolation
-      # https://rubular.com/r/HPwtW9OYxKUHXQ
-      source.lines.map do |line|
-        if line.match?(/^\s*#([^{].*|)$/)
-          $/
-        else
-          line
-        end
-      end
-    end
-
     # Smushes all heredoc lines into one line
     #
     #     source = <<~'EOM'
@@ -184,9 +110,9 @@ def join_heredoc!
       lines.each do |line|
         line.tokens.each do |token|
           case token.type
-          when :on_heredoc_beg
+          when :HEREDOC_START
             start_index_stack << line.index
-          when :on_heredoc_end
+          when :HEREDOC_END
             start_index = start_index_stack.pop
             end_index = line.index
             heredoc_beg_end_index << [start_index, end_index]
@@ -212,20 +138,10 @@ def join_heredoc!
     #     expect(lines[0].to_s).to eq(source)
     #     expect(lines[1].to_s).to eq("")
     #
-    # The one known case this doesn't handle is:
-    #
-    #     Ripper.lex <<~EOM
-    #       a &&
-    #        b ||
-    #        c
-    #     EOM
-    #
-    # For some reason this introduces `on_ignore_newline` but with BEG type
-    #
     def join_consecutive!
-      consecutive_groups = @document.select(&:ignore_newline_not_beg?).map do |code_line|
+      consecutive_groups = @document.select(&:consecutive?).map do |code_line|
         take_while_including(code_line.index..) do |line|
-          line.ignore_newline_not_beg?
+          line.consecutive?
         end
       end
 
@@ -275,14 +191,15 @@ def join_groups(groups)
         @document[line.index] = CodeLine.new(
           tokens: lines.map(&:tokens).flatten,
           line: lines.join,
-          index: line.index
+          index: line.index,
+          consecutive: false
         )
 
         # Hide the rest of the lines
         lines[1..].each do |line|
           # The above lines already have newlines in them, if add more
           # then there will be double newline, use an empty line instead
-          @document[line.index] = CodeLine.new(line: "", index: line.index, tokens: [])
+          @document[line.index] = CodeLine.new(line: "", index: line.index, tokens: [], consecutive: false)
         end
       end
       self

lib/syntax_suggest/code_line.rb

@@ -27,21 +27,56 @@ class CodeLine
     # Returns an array of CodeLine objects
     # from the source string
     def self.from_source(source)
-      tokens_for_line = LexAll.new(source: source).each_with_object(Hash.new { |h, k| h[k] = [] }) { |token, hash| hash[token.line] << token }
+      source = +source
+      parse_result = Prism.parse_lex(source)
+      ast, tokens = parse_result.value
+
+      clean_comments!(source, parse_result.comments)
+
+      visitor = Visitor.new
+      visitor.visit(ast)
+      tokens.sort_by! { |token, _state| token.location.start_line }
+
+      prev_token = nil
+      tokens.map! do |token, _state|
+        prev_token = Token.new(token, prev_token, visitor)
+      end
+
+      tokens_for_line = tokens.each_with_object(Hash.new { |h, k| h[k] = [] }) { |token, hash| hash[token.line] << token }
       source.lines.map.with_index do |line, index|
         CodeLine.new(
           line: line,
           index: index,
-          tokens: tokens_for_line[index + 1]
+          tokens: tokens_for_line[index + 1],
+          consecutive: visitor.consecutive_lines.include?(index + 1)
         )
       end
     end
 
+    # Remove comments that apear on their own in source. They will never be the cause
+    # of syntax errors and are just visual noise. Example:
+    #
+    #   source = +<<~RUBY
+    #     # Comment-only line
+    #     foo # Inline comment
+    #   RUBY
+    #   CodeLine.clean_comments!(source, Prism.parse(source).comments)
+    #   source # => "\nfoo # Inline comment\n"
+    def self.clean_comments!(source, comments)
+      # Iterate backwards since we are modifying the source in place and must preserve
+      # the offsets. Prism comments are sorted by their location in the source.
+      comments.reverse_each do |comment|
+        next if comment.trailing?
+        source.bytesplice(comment.location.start_offset, comment.location.length, "")
+      end
+    end
+
     attr_reader :line, :index, :tokens, :line_number, :indent
-    def initialize(line:, index:, tokens:)
+    def initialize(line:, index:, tokens:, consecutive:)
       @tokens = tokens
       @line = line
       @index = index
+      @consecutive = consecutive
       @original = line
       @line_number = @index + 1
       strip_line = line.dup
@@ -150,91 +185,36 @@ def <=>(other)
       index <=> other.index
     end
 
-    # [Not stable API]
-    #
-    # Lines that have a `on_ignored_nl` type token and NOT
-    # a `BEG` type seem to be a good proxy for the ability
-    # to join multiple lines into one.
-    #
-    # This predicate method is used to determine when those
-    # two criteria have been met.
-    #
-    # The one known case this doesn't handle is:
-    #
-    #     Ripper.lex <<~EOM
-    #       a &&
-    #        b ||
-    #        c
-    #     EOM
-    #
-    # For some reason this introduces `on_ignore_newline` but with BEG type
-    def ignore_newline_not_beg?
-      @ignore_newline_not_beg
+    # Can this line be logically joined together
+    # with the following line? Determined by walking
+    # the AST
+    def consecutive?
+      @consecutive
     end
 
-    # Determines if the given line has a trailing slash
+    # Determines if the given line has a trailing slash.
+    # Simply check if the line contains a backslash after
+    # the content of the last token.
     #
     #     lines = CodeLine.from_source(<<~EOM)
     #       it "foo" \
     #     EOM
     #     expect(lines.first.trailing_slash?).to eq(true)
     #
     def trailing_slash?
-      last = @tokens.last
-
-      # Older versions of prism diverged slightly from Ripper in compatibility mode
-      case last&.type
-      when :on_sp
-        last.value == TRAILING_SLASH
-      when :on_tstring_end
-        true
-      else
-        false
-      end
+      return unless (last = @tokens.last)
+      @line.byteindex(TRAILING_SLASH, last.location.end_column) != nil
     end
 
-    # Endless method detection
-    #
-    # From https://github.com/ruby/irb/commit/826ae909c9c93a2ddca6f9cfcd9c94dbf53d44ab
-    # Detecting a "oneliner" seems to need a state machine.
-    # This can be done by looking mostly at the "state" (last value):
-    #
-    #   ENDFN -> BEG (token = '=' ) -> END
-    #
     private def set_kw_end
-      oneliner_count = 0
-      in_oneliner_def = nil
-
       kw_count = 0
       end_count = 0
 
-      @ignore_newline_not_beg = false
       @tokens.each do |token|
         kw_count += 1 if token.is_kw?
         end_count += 1 if token.is_end?
-
-        if token.type == :on_ignored_nl
-          @ignore_newline_not_beg = !token.expr_beg?
-        end
-
-        if in_oneliner_def.nil?
-          in_oneliner_def = :ENDFN if token.state.allbits?(Ripper::EXPR_ENDFN)
-        elsif token.state.allbits?(Ripper::EXPR_ENDFN)
-          # Continue
-        elsif token.state.allbits?(Ripper::EXPR_BEG)
-          in_oneliner_def = :BODY if token.value == "="
-        elsif token.state.allbits?(Ripper::EXPR_END)
-          # We found an endless method, count it
-          oneliner_count += 1 if in_oneliner_def == :BODY
-
-          in_oneliner_def = nil
-        else
-          in_oneliner_def = nil
-        end
       end
 
-      kw_count -= oneliner_count
-
       @is_kw = (kw_count - end_count) > 0
       @is_end = (end_count - kw_count) > 0
     end

lib/syntax_suggest/left_right_token_count.rb

@@ -49,21 +49,22 @@ def count_end
     #
     # Example:
     #
+    #   token = CodeLine.from_source("{").first.tokens.first
     #   left_right = LeftRightTokenCount.new
-    #   left_right.count_token(Token.new(1, :on_lbrace, "{", Ripper::EXPR_BEG))
+    #   left_right.count_token(Token.new(token)
     #   left_right.count_for_char("{")
     #   # => 1
     #   left_right.count_for_char("}")
     #   # => 0
     def count_token(token)
       case token.type
-      when :on_tstring_content
+      when :STRING_CONTENT
         # ^^^
         # Means it's a string or a symbol `"{"` rather than being
         # part of a data structure (like a hash) `{ a: b }`
         # ignore it.
-      when :on_words_beg, :on_symbols_beg, :on_qwords_beg,
-           :on_qsymbols_beg, :on_regexp_beg, :on_tstring_beg
+      when :PERCENT_UPPER_W, :PERCENT_UPPER_I, :PERCENT_LOWER_W,
+           :PERCENT_LOWER_I, :REGEXP_BEGIN, :STRING_BEGIN
         # ^^^
         # Handle shorthand syntaxes like `%Q{ i am a string }`
         #
@@ -72,25 +73,18 @@ def count_token(token)
         # can be used
         char = token.value[-1]
         @count_for_char[char] += 1 if @count_for_char.key?(char)
-      when :on_embexpr_beg
+      when :EMBEXPR_BEGIN
         # ^^^
         # Embedded string expressions like `"#{foo} <-embed"`
         # are parsed with chars:
         #
-        # `#{` as :on_embexpr_beg
-        #  `}` as :on_embexpr_end
-        #
-        # We cannot ignore both :on_emb_expr_beg and :on_embexpr_end
-        # because sometimes the lexer thinks something is an embed
-        # string end, when it is not like `lol = }` (no clue why).
+        # `#{` as :EMBEXPR_BEGIN
+        #  `}` as :EMBEXPR_END
         #
         # When we see `#{` count it as a `{` or we will
         # have a mis-match count.
         #
-        case token.value
-        when "\#{"
-          @count_for_char["{"] += 1
-        end
+        @count_for_char["{"] += 1
       else
         @end_count += 1 if token.is_end?
         @kw_count += 1 if token.is_kw?