rubocop 1.71.1 → 1.92.1

data/lib/rubocop/runner.rb

@@ -62,14 +62,34 @@ module RuboCop
       @errors = []
       @warnings = []
       @aborting = false
+      @inspected_files = []
+      @diff_sources = {}
+      @report_queue = {}
+    end
+
+    # The `--diff` patches, one per file, from the source on disk (or stdin) to
+    # the source once every correction, redundant directive removal included,
+    # has been applied.
+    def diffs
+      @diffs ||= @diff_sources.filter_map do |file, (original_source, corrected_source)|
+        diff = UnifiedDiff.new(PathUtil.smart_path(file), original_source, corrected_source).to_s
+        diff unless diff.empty?
+      end
     end
 
     def run(paths)
-      target_files = find_target_files(paths)
+      @inspection_team_by_config = {}.compare_by_identity
+
+      # Compute the cache source checksum once to avoid potential
+      # inconsistencies between workers.
+      ResultCache.source_checksum
+
+      target_files = only_changed(find_target_files(paths))
+      build_project_index(target_files)
+
       if @options[:list_target_files]
         list_files(target_files)
       else
-        warm_cache(target_files) if @options[:parallel]
         inspect_files(target_files)
       end
     rescue Interrupt
@@ -86,19 +106,12 @@ module RuboCop
 
     private
 
-    # Warms up the RuboCop cache by forking a suitable number of RuboCop
-    # instances that each inspects its allotted group of files.
-    def warm_cache(target_files)
-      saved_options = @options.dup
-      if target_files.length <= 1
-        puts 'Skipping parallel inspection: only a single file needs inspection' if @options[:debug]
-        return
-      end
-      puts 'Running parallel inspection' if @options[:debug]
-      %i[autocorrect safe_autocorrect].each { |opt| @options[opt] = false }
-      Parallel.each(target_files) { |target_file| file_offenses(target_file) }
-    ensure
-      @options = saved_options
+    # The project index is deliberately built from the unfiltered file list:
+    # cross-file offenses must not depend on which files happen to have changed.
+    def only_changed(target_files)
+      return target_files unless @options[:changed]
+
+      ChangedFiles.new(@options[:changed]).filter(target_files).freeze
     end
 
     def find_target_files(paths)
@@ -112,12 +125,54 @@ module RuboCop
       target_files.each(&:freeze).freeze
     end
 
-    def inspect_files(files)
-      inspected_files = []
+    def build_project_index(target_files)
+      return unless project_index_enabled?
+
+      @project_index = ProjectIndexLoader.build_index(project_index_files(target_files))
+    end
+
+    # The index always covers the whole project, not just the files being
+    # inspected: cross-file offenses must not depend on which files a
+    # particular run happens to include, or single-file runs (editors, CI
+    # sharding) would see different offenses than full runs. The project is
+    # rooted at the directory of the configuration that enabled the index.
+    def project_index_files(target_files)
+      root = @config_store.for_pwd.base_dir_for_path_parameters
+      project_files = begin
+        find_target_files([root]) | target_files
+      rescue Errno::ENOENT
+        target_files
+      end
+
+      (project_files | bundled_gem_files).sort
+    end
+
+    def bundled_gem_files
+      return [] unless @config_store.for_pwd.for_all_cops['ProjectIndexIncludesGems']
+
+      ProjectIndexLoader.bundled_gem_source_files
+    end
+
+    def project_index_enabled?
+      return false unless @config_store.for_pwd.for_all_cops['UseProjectIndex']
+
+      unless ProjectIndexLoader.available?
+        ProjectIndexLoader.warn_unavailable
+        return false
+      end
+
+      true
+    end
 
+    def inspect_files(files) # rubocop:disable Metrics/AbcSize
       formatter_set.started(files)
+      formatters_started = true
+      file_iterator(files) do |file|
+        offenses, problems = collect_problems { process_file(file) }
+        succeeded = offenses.none? { |o| considered_failure?(o) && offense_displayed?(o) }
 
-      each_inspected_file(files) { |file| inspected_files << file }
+        [offenses, succeeded, problems]
+      end
     ensure
       # OPTIMIZE: Calling `ResultCache.cleanup` takes time. This optimization
       # mainly targets editors that integrates RuboCop. When RuboCop is run
@@ -125,23 +180,101 @@ module RuboCop
       if files.size > 1 && cached_run?
         ResultCache.cleanup(@config_store, @options[:debug], @options[:cache_root])
       end
-
-      formatter_set.finished(inspected_files.freeze)
+      formatter_set.finished(@inspected_files.freeze) if formatters_started
       formatter_set.close_output_files
+      warn_about_elided_offenses
     end
 
-    def each_inspected_file(files)
-      files.reduce(true) do |all_passed, file|
-        offenses = process_file(file)
-        yield file
+    def file_iterator(files, &block)
+      all_passed = true
 
-        if offenses.any? { |o| considered_failure?(o) && offense_displayed?(o) }
-          break false if @options[:fail_fast]
+      on_start = ->(file, _index) { file_started(file) }
+      on_finish = lambda do |file, index, (offenses, passed, problems)|
+        all_passed &&= passed
+        finished_report(file, index, offenses, problems)
+      end
 
-          next false
-        end
+      if run_in_parallel?(files)
+        parallel_file_iterator(files, on_start, on_finish, &block)
+      else
+        serial_file_iterator(files, on_start, on_finish, &block)
+      end
+
+      process_remaining_report_queue
 
-        all_passed
+      all_passed
+    end
+
+    def finished_report(file, index, offenses, problems)
+      @report_queue[index] = [file, offenses, problems]
+      @next_index_to_report ||= 0
+      while @report_queue.key?(@next_index_to_report)
+        process_report_queue_entry(@next_index_to_report)
+        @next_index_to_report += 1
+      end
+    end
+
+    def process_report_queue_entry(index)
+      file, offenses, (errors, warnings) = @report_queue.delete(index)
+      @errors.concat(errors)
+      @warnings.concat(warnings)
+      file_problems(file, errors, warnings) unless errors.empty? && warnings.empty?
+      file_finished(file, offenses)
+    end
+
+    def process_remaining_report_queue
+      @report_queue.keys.sort.each do |index|
+        process_report_queue_entry(index)
+      end
+    end
+
+    def run_in_parallel?(files)
+      return false unless parallel_supported_by_options?
+
+      if files.size <= 1
+        puts 'Skipping parallel inspection: only a single file needs inspection' if @options[:debug]
+        return false
+      end
+
+      return false if project_index_disables_parallel?
+
+      puts 'Running parallel inspection' if @options[:debug]
+
+      true
+    end
+
+    # `--auto-gen-config` needs the offenses of every file in one place, and
+    # `--diff` collects the diffs in this process, where a worker's copy of
+    # them would be lost.
+    def parallel_supported_by_options?
+      @options[:parallel] && !@options[:auto_gen_config] && !@options[:diff]
+    end
+
+    def project_index_disables_parallel?
+      return false if @project_index.nil? || !Gem.win_platform?
+
+      if @options[:debug]
+        puts 'Skipping parallel inspection: the project index is enabled and parallel ' \
+             'inspection is not yet supported on Windows.'
+      end
+
+      true
+    end
+
+    def parallel_file_iterator(files, on_start, on_finish, &block)
+      Parallel.each(files, start: on_start, finish: on_finish, &block)
+    end
+
+    def serial_file_iterator(files, on_start, on_finish, &block)
+      files.each_with_index do |file, index|
+        on_start.call(file, index)
+        result = yield file
+        on_finish.call(file, index, result)
+
+        # Report and count the offending file before stopping so `--fail-fast`
+        # still shows its offenses and exits with a failing status.
+        _offenses, succeeded = result
+        break if @options[:fail_fast] && !succeeded
       end
     end
 
@@ -149,24 +282,37 @@ module RuboCop
       paths.each { |path| puts PathUtil.relative_path(path) }
     end
 
+    # Keeps the errors and warnings a file causes apart from the rest of the
+    # run's. A parallel worker inspects files in a copy of the runner that is
+    # thrown away, so they have to travel back with the file's offenses.
+    def collect_problems
+      errors = @errors
+      warnings = @warnings
+      @errors = []
+      @warnings = []
+
+      [yield, [@errors, @warnings]]
+    ensure
+      @errors = errors
+      @warnings = warnings
+    end
+
     def process_file(file)
-      file_started(file)
-      offenses = file_offenses(file)
+      file_offenses(file)
     rescue InfiniteCorrectionLoop => e
       raise e if @options[:raise_cop_error]
 
       errors << e
       warn Rainbow(e.message).red
-      offenses = e.offenses.compact.sort.freeze
-    ensure
-      file_finished(file, offenses || [])
+      e.offenses.compact.sort.freeze
     end
 
     def file_offenses(file)
       file_offense_cache(file) do
         source, offenses = do_inspection_loop(file)
         offenses = add_redundant_disables(file, offenses.compact.sort, source)
-        offenses.sort.reject(&:disabled?).freeze
+        offenses = offenses.reject(&:disabled?) unless @options[:display_suppressed]
+        offenses.sort.freeze
       end
     end
 
@@ -190,7 +336,7 @@ module RuboCop
 
       if real_run_needed
         offenses = yield
-        save_in_cache(cache, offenses)
+        save_in_cache(cache, offenses) unless Cop::Registry.global.warnings?(file)
       end
 
       offenses
@@ -203,9 +349,11 @@ module RuboCop
         if redundant_updated
           # Do one extra inspection loop if any redundant disables were
           # removed. This is done in order to find rubocop:enable directives that
-          # have now become useless.
-          _source, new_offenses = do_inspection_loop(file)
-          offenses |= new_offenses
+          # have now become useless. It finds the offenses that are left again,
+          # located in the corrected source, so only the corrected ones are kept
+          # from before, as `inspect_and_correct` does.
+          _source, new_offenses = do_inspection_loop(file, team.updated_source)
+          offenses = offenses.select(&:corrected?) | new_offenses
         end
       end
       offenses
@@ -218,12 +366,20 @@ module RuboCop
       team = Cop::Team.mobilize([Cop::Lint::RedundantCopDisableDirective], config, @options)
       return if team.cops.empty?
 
+      team.defer_corrections = @options[:diff]
+
       team.cops.first.offenses_to_check = offenses
       yield team
     end
 
     def check_for_redundant_disables?(source)
-      return false if source.disabled_line_ranges.empty? || except_redundant_cop_disable_directive?
+      return false if except_redundant_cop_disable_directive?
+      # Detached `disable-next` directives produce no disabled ranges but must
+      # still be reported as redundant.
+      if source.disabled_line_ranges.empty? &&
+         source.comment_config.detached_next_directives.empty?
+        return false
+      end
 
       !@options[:only]
     end
@@ -245,7 +401,17 @@ module RuboCop
       formatter_set.file_started(file, cli_options: @options, config_store: @config_store)
     end
 
+    # Formatters get the messages without the colors added for the terminal.
+    def file_problems(file, errors, warnings)
+      formatter_set.file_problems(file, plain_messages(errors), plain_messages(warnings))
+    end
+
+    def plain_messages(problems)
+      problems.map { |problem| Rainbow::StringUtils.uncolor(problem.to_s) }
+    end
+
     def file_finished(file, offenses)
+      @inspected_files << file
       offenses = offenses_to_report(offenses)
       formatter_set.file_finished(file, offenses)
     end
@@ -254,10 +420,6 @@ module RuboCop
       @cached_run ||=
         (@options[:cache] == 'true' ||
          (@options[:cache] != 'false' && @config_store.for_pwd.for_all_cops['UseCache'])) &&
-        # When running --auto-gen-config, there's some processing done in the
-        # cops related to calculating the Max parameters for Metrics cops. We
-        # need to do that processing and cannot use caching.
-        !@options[:auto_gen_config] &&
         # We can't cache results from code which is piped in to stdin
         !@options[:stdin]
     end
@@ -266,41 +428,107 @@ module RuboCop
       return unless cache
       # Caching results when a cop has crashed would prevent the crash in the
       # next run, since the cop would not be called then. We want crashes to
-      # show up the same in each run.
+      # show up the same in each run. Only this file's errors and warnings are
+      # visible here, see `collect_problems`.
       return if errors.any? || warnings.any?
 
       cache.save(offenses)
     end
 
-    def do_inspection_loop(file)
-      processed_source = get_processed_source(file)
+    def do_inspection_loop(file, source = nil)
+      processed_source = initial_processed_source(file, source)
+      original_source = processed_source.raw_source
       # This variable is 2d array used to track corrected offenses after each
       # inspection iteration. This is used to output meaningful infinite loop
       # error message.
       offenses_by_iteration = []
+      corrected_source = nil
 
-      # When running with --autocorrect, we need to inspect the file (which
-      # includes writing a corrected version of it) until no more corrections
-      # are made. This is because automatic corrections can introduce new
-      # offenses. In the normal case the loop is only executed once.
+      # When running with --autocorrect, we need to inspect the file until no
+      # more corrections are made. This is because automatic corrections can
+      # introduce new offenses. In the normal case the loop is only executed
+      # once. The corrections are kept in memory while iterating and written
+      # back to the file when the loop is done.
       iterate_until_no_changes(processed_source, offenses_by_iteration) do
-        # The offenses that couldn't be corrected will be found again so we
-        # only keep the corrected ones in order to avoid duplicate reporting.
-        !offenses_by_iteration.empty? && offenses_by_iteration.last.select!(&:corrected?)
-        new_offenses, updated_source_file = inspect_file(processed_source)
-        offenses_by_iteration.push(new_offenses)
+        team, updated_source_file = inspect_and_correct(processed_source, offenses_by_iteration)
 
         # We have to reprocess the source to pickup the changes. Since the
         # change could (theoretically) introduce parsing errors, we break the
         # loop if we find any.
         break unless updated_source_file
 
-        processed_source = get_processed_source(file)
+        # Autocorrect has happened, don't use the prism result since it is stale.
+        # With --stdin the corrected source is kept in @options[:stdin] instead.
+        corrected_source = team.updated_source
+        processed_source = get_processed_source(file, nil, source: corrected_source)
       end
 
       # Return summary of corrected offenses after all iterations
-      offenses = offenses_by_iteration.flatten.uniq
-      [processed_source, offenses]
+      [processed_source, offenses_by_iteration.flatten.uniq]
+    ensure
+      finalize_corrections(file, original_source, corrected_source)
+    end
+
+    # `source` is the corrected source to continue from when the corrections
+    # are kept in memory rather than written to the file (`--diff`).
+    def initial_processed_source(file, source)
+      return get_processed_source(file, nil, source: source) if source
+
+      # We can reuse the prism result since the source did not change yet.
+      get_processed_source(file, @prism_result)
+    end
+
+    def inspect_and_correct(processed_source, offenses_by_iteration)
+      # The offenses that couldn't be corrected will be found again so we
+      # only keep the corrected ones in order to avoid duplicate reporting.
+      !offenses_by_iteration.empty? && offenses_by_iteration.last.select!(&:corrected?)
+      team, new_offenses, updated_source_file = inspect_iteration(processed_source)
+      offenses_by_iteration.push(new_offenses)
+
+      [team, updated_source_file]
+    end
+
+    # Write the file once, even when the loop was left through an exception
+    # (e.g. an infinite correction loop), like the per-iteration writes used
+    # to be. Under `--diff` nothing is written at all.
+    def finalize_corrections(file, original_source, corrected_source)
+      if @options[:diff]
+        record_diff(file, original_source, corrected_source)
+      elsif corrected_source
+        Util.replace_file_contents(file, corrected_source)
+      end
+    end
+
+    # `--diff` reports what autocorrection would do instead of doing it. With
+    # `--stdin` the corrected source is handed back through the options rather
+    # than deferred, so that is where the new source comes from. A file can go
+    # through more than one inspection loop, so the diff runs from the source
+    # the first loop started with to the one the last loop ended with.
+    def record_diff(file, original_source, corrected_source)
+      return unless original_source
+
+      new_source = @options[:stdin] || corrected_source
+      # The original source is the file as it is on disk, so the corrected one
+      # has to go through the same line ending conversion `File.write` would
+      # apply, or on Windows every single line reads as changed.
+      new_source = new_source ? Util.emulate_write_read_cycle(new_source) : original_source
+
+      sources = (@diff_sources[file] ||= [original_source])
+      sources[1] = new_source
+    end
+
+    def inspect_iteration(processed_source)
+      team = mobilize_team(processed_source)
+      team.defer_corrections = @options[:diff] || in_memory_corrections_possible?
+      offenses, updated_source_file = inspect_file(processed_source, team)
+      [team, offenses, updated_source_file]
+    end
+
+    # Custom ruby extractors may derive their fragments from the file on
+    # disk rather than from the passed processed source, so corrections can
+    # only be kept in memory when the default extractor is used.
+    def in_memory_corrections_possible?
+      self.class.ruby_extractors.one?
     end
 
     def iterate_until_no_changes(source, offenses_by_iteration)
@@ -344,16 +572,9 @@ module RuboCop
 
     def inspect_file(processed_source, team = mobilize_team(processed_source))
       extracted_ruby_sources = extract_ruby_sources(processed_source)
-      offenses = extracted_ruby_sources.flat_map do |extracted_ruby_source|
-        report = team.investigate(
-          extracted_ruby_source[:processed_source],
-          offset: extracted_ruby_source[:offset],
-          original: processed_source
-        )
-        @errors.concat(team.errors)
-        @warnings.concat(team.warnings)
-        report.offenses
-      end
+      offenses = team.investigate_fragments(extracted_ruby_sources, original: processed_source)
+      @errors.concat(team.errors)
+      @warnings.concat(team.warnings)
       [offenses, team.updated_source_file?]
     end
 
@@ -373,30 +594,41 @@ module RuboCop
 
     def mobilize_team(processed_source)
       config = @config_store.for_file(processed_source.path)
-      Cop::Team.mobilize(mobilized_cop_classes(config), config, @options)
+      @inspection_team_by_config ||= {}.compare_by_identity
+      @inspection_team_by_config[config] ||= assemble_team(config)
     end
 
-    def mobilized_cop_classes(config) # rubocop:disable Metrics/AbcSize
+    def mobilized_cop_classes(config)
       @mobilized_cop_classes ||= {}.compare_by_identity
       @mobilized_cop_classes[config] ||= begin
-        cop_classes = Cop::Registry.all
-
         # `@options[:only]` and `@options[:except]` are not qualified until
-        # needed so that the Registry can be fully loaded, including any
-        # cops added by `require`s.
+        # needed so that the registry contains any cops added by `require`s,
+        # which register eagerly when their files are loaded.
         qualify_option_cop_names
 
         OptionsValidator.new(@options).validate_cop_options
 
-        if @options[:only]
-          cop_classes.select! { |c| c.match?(@options[:only]) }
-        else
-          filter_cop_classes(cop_classes, config)
-        end
+        # Filtering by badge keeps lazy-loaded cops unloaded; only the cops
+        # that survive the filter and are enabled will be loaded.
+        Cop::Registry.global.filter_by_badge(@options) { |badge| mobilize_cop_badge?(badge, config) }
+      end
+    end
+
+    # Lets a runner that is reused from run to run pick up a different `--only`
+    # or `--except`.
+    def forget_mobilized_cops
+      @mobilized_cop_classes = nil
+    end
 
-        cop_classes.reject! { |c| c.match?(@options[:except]) }
+    def mobilize_cop_badge?(badge, config)
+      return false if badge.department == :Test
+      return false if badge.match_name?(@options[:except])
 
-        Cop::Registry.new(cop_classes, @options)
+      if @options[:only]
+        badge.match_name?(@options[:only])
+      else
+        # use only cops that link to a style guide if requested.
+        !style_guide_cops_only?(config) || config.for_cop(badge.to_s)['StyleGuide']
       end
     end
 
@@ -404,19 +636,12 @@ module RuboCop
       %i[only except].each do |option|
         next unless @options[option]
 
-        @options[option].map! do |cop_name|
+        @options[option] = @options[option].map do |cop_name|
           Cop::Registry.qualified_cop_name(cop_name, "--#{option} option")
         end
       end
     end
 
-    def filter_cop_classes(cop_classes, config)
-      # use only cops that link to a style guide if requested
-      return unless style_guide_cops_only?(config)
-
-      cop_classes.select! { |cop| config.for_cop(cop)['StyleGuide'] }
-    end
-
     def style_guide_cops_only?(config)
       @options[:only_guide_cops] || config.for_all_cops['StyleGuideCopsOnly']
     end
@@ -452,7 +677,28 @@ module RuboCop
     end
 
     def offenses_to_report(offenses)
-      offenses.select { |o| offense_displayed?(o) }
+      offenses = offenses.select { |o| offense_displayed?(o) }
+      offenses = offense_limit.filter(offenses) if offense_limit
+      offenses
+    end
+
+    # Files are reported in order, so which offenses survive the limit does not
+    # depend on how the run was parallelized.
+    def offense_limit
+      return @offense_limit if defined?(@offense_limit)
+
+      max = @options[:max_offenses_per_cop]
+      @offense_limit = (OffenseLimit.new(max.to_i) if max)
+    end
+
+    def warn_about_elided_offenses
+      elided = offense_limit&.elided_per_cop
+      return unless elided&.any?
+
+      warn Rainbow(
+        "#{elided.values.sum} more offenses were not reported because of " \
+        "--max-offenses-per-cop=#{offense_limit.max} (#{elided.keys.join(', ')})."
+      ).yellow
     end
 
     def supports_safe_autocorrect?(offense)
@@ -472,24 +718,30 @@ module RuboCop
     end
 
     def minimum_severity_to_fail
-      @minimum_severity_to_fail ||= begin
-        # Unless given explicitly as `fail_level`, `:info` severity offenses do not fail
-        name = @options[:fail_level] || :refactor
-
-        # autocorrect is a fake level - use the default
-        RuboCop::Cop::Severity.new(name == :autocorrect ? :refactor : name)
-      end
+      @minimum_severity_to_fail ||=
+        RuboCop::Cop::Severity.minimum_to_fail(@options, @config_store.for_pwd)
     end
 
-    # rubocop:disable Metrics/MethodLength
-    def get_processed_source(file)
+    # rubocop:disable-next Metrics/MethodLength
+    def get_processed_source(file, prism_result, source: nil)
       config = @config_store.for_file(file)
       ruby_version = config.target_ruby_version
       parser_engine = config.parser_engine
 
-      processed_source = if @options[:stdin]
+      processed_source = if source
+                           ProcessedSource.new(
+                             Util.emulate_write_read_cycle(source),
+                             ruby_version,
+                             file,
+                             parser_engine: parser_engine
+                           )
+                         elsif @options[:stdin]
                            ProcessedSource.new(
-                             @options[:stdin], ruby_version, file, parser_engine: parser_engine
+                             @options[:stdin],
+                             ruby_version,
+                             file,
+                             parser_engine: parser_engine,
+                             prism_result: prism_result
                            )
                          else
                            begin
@@ -504,16 +756,26 @@ module RuboCop
       processed_source.registry = mobilized_cop_classes(config)
       processed_source
     end
-    # rubocop:enable Metrics/MethodLength
 
     # A Cop::Team instance is stateful and may change when inspecting.
     # The "standby" team for a given config is an initialized but
     # otherwise dormant team that can be used for config- and option-
     # level caching in ResultCache.
     def standby_team(config)
-      @team_by_config ||= {}.compare_by_identity
-      @team_by_config[config] ||=
-        Cop::Team.mobilize(mobilized_cop_classes(config), config, @options)
+      @standby_team_by_config ||= {}.compare_by_identity
+      @standby_team_by_config[config] ||= assemble_team(config)
+    end
+
+    def assemble_team(config)
+      team = Cop::Team.mobilize(mobilized_cop_classes(config), config, @options)
+
+      if @project_index
+        team.cops.each do |cop|
+          cop.project_index = @project_index
+        end
+      end
+
+      team
     end
   end
 end