class RuboCop::Cop::Style::Copyright


an offense is reported.
that RuboCop scans, a comment that matches this regex must be found or
This regex string is treated as an unanchored regex. For each file
β€”-
Notice: β€˜^Copyright ((c) )?2d{3} Acme Inc’
Style/Copyright:
β€”-
[source,yaml]

config/default.yml. The default can be changed as follows:
The default regexp for an acceptable copyright notice can be found in
Check that a copyright notice was given in each source file.

def autocorrect(corrector)

def autocorrect(corrector)
  token = insert_notice_before(processed_source)
  range = token.nil? ? range_between(0, 0) : token.pos
  corrector.insert_before(range, "#{autocorrect_notice}\n")
end

def autocorrect_notice

def autocorrect_notice
  cop_config['AutocorrectNotice']
end

def encoding_token?(processed_source, token_index)

def encoding_token?(processed_source, token_index)
  return false if token_index >= processed_source.tokens.size
  token = processed_source.tokens[token_index]
  token.comment? && /\A#.*coding\s?[:=]\s?(?:UTF|utf)-8/.match?(token.text)
end

def insert_notice_before(processed_source)

def insert_notice_before(processed_source)
  token_index = 0
  token_index += 1 if shebang_token?(processed_source, token_index)
  token_index += 1 if encoding_token?(processed_source, token_index)
  processed_source.tokens[token_index]
end

def notice

def notice
  cop_config['Notice']
end

def notice_found?(processed_source)

def notice_found?(processed_source)
  notice_regexp = Regexp.new(notice.lines.map(&:strip).join)
  multiline_notice = +''
  processed_source.tokens.each do |token|
    break unless token.comment?
    multiline_notice << token.text.sub(/\A# */, '')
    break if notice_regexp.match?(token.text)
  end
  multiline_notice.match?(notice_regexp)
end

def on_new_investigation

def on_new_investigation
  return if notice.empty? || notice_found?(processed_source)
  verify_autocorrect_notice!
  message = format(MSG, notice: notice)
  if processed_source.blank?
    add_global_offense(message)
  else
    offense_range = source_range(processed_source.buffer, 1, 0)
    add_offense(offense_range, message: message) do |corrector|
      autocorrect(corrector)
    end
  end
end

def shebang_token?(processed_source, token_index)

def shebang_token?(processed_source, token_index)
  return false if token_index >= processed_source.tokens.size
  token = processed_source.tokens[token_index]
  token.comment? && /\A#!.*\z/.match?(token.text)
end

def verify_autocorrect_notice!

def verify_autocorrect_notice!
  if autocorrect_notice.nil? || autocorrect_notice.empty?
    raise Warning, "#{cop_name}: #{AUTOCORRECT_EMPTY_WARNING}"
  end
  regex = Regexp.new(notice)
  return if autocorrect_notice.gsub(/^# */, '').match?(regex)
  message = "AutocorrectNotice '#{autocorrect_notice}' must match Notice /#{notice}/"
  raise Warning, "#{cop_name}: #{message}"
end