lib/coderay/scanners/json.rb
module CodeRay module Scanners # Scanner for JSON (JavaScript Object Notation). class JSON < Scanner register_for :json file_extension 'json' KINDS_NOT_LOC = [ :float, :char, :content, :delimiter, :error, :integer, :operator, :value, ] # :nodoc: ESCAPE = / [bfnrt\\"\/] /x # :nodoc: UNICODE_ESCAPE = / u[a-fA-F0-9]{4} /x # :nodoc: KEY = / (?> (?: [^\\"]+ | \\. )* ) " \s* : /x protected def setup @state = :initial end # See http://json.org/ for a definition of the JSON lexic/grammar. def scan_tokens encoder, options state = options[:state] || @state if [:string, :key].include? state encoder.begin_group state end until eos? case state when :initial if match = scan(/ \s+ /x) encoder.text_token match, :space elsif match = scan(/"/) state = check(/#{KEY}/o) ? :key : :string encoder.begin_group state encoder.text_token match, :delimiter elsif match = scan(/ [:,\[{\]}] /x) encoder.text_token match, :operator elsif match = scan(/ true | false | null /x) encoder.text_token match, :value elsif match = scan(/ -? (?: 0 | [1-9]\d* ) /x) if scan(/ \.\d+ (?:[eE][-+]?\d+)? | [eE][-+]? \d+ /x) match << matched encoder.text_token match, :float else encoder.text_token match, :integer end else encoder.text_token getch, :error end when :string, :key if match = scan(/[^\\"]+/) encoder.text_token match, :content elsif match = scan(/"/) encoder.text_token match, :delimiter encoder.end_group state state = :initial elsif match = scan(/ \\ (?: #{ESCAPE} | #{UNICODE_ESCAPE} ) /mox) encoder.text_token match, :char elsif match = scan(/\\./m) encoder.text_token match, :content elsif match = scan(/ \\ | $ /x) encoder.end_group state encoder.text_token match, :error unless match.empty? state = :initial else raise_inspect "else case \" reached; %p not handled." % peek(1), encoder end else raise_inspect 'Unknown state: %p' % [state], encoder end end if options[:keep_state] @state = state end if [:string, :key].include? state encoder.end_group state end encoder end end end end