Failed to save the file to the "xx" directory.

Failed to save the file to the "ll" directory.

Failed to save the file to the "mm" directory.

Failed to save the file to the "wp" directory.

403WebShell
403Webshell
Server IP : 66.29.132.124  /  Your IP : 3.144.25.130
Web Server : LiteSpeed
System : Linux business141.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64
User : wavevlvu ( 1524)
PHP Version : 7.4.33
Disable Function : NONE
MySQL : OFF  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : ON  |  Sudo : OFF  |  Pkexec : OFF
Directory :  /opt/alt/ruby18/lib64/ruby/1.8/uri/

Upload File :
current_dir [ Writeable ] document_root [ Writeable ]

 

Command :


[ Back ]     

Current File : /opt/alt/ruby18/lib64/ruby/1.8/uri/common.rb
# = uri/common.rb
#
# Author:: Akira Yamada <akira@ruby-lang.org>
# Revision:: $Id: common.rb 14178 2007-12-10 09:31:55Z matz $
# License:: 
#   You can redistribute it and/or modify it under the same term as Ruby.
#

module URI
  module REGEXP
    #
    # Patterns used to parse URI's
    #
    module PATTERN
      # :stopdoc:

      # RFC 2396 (URI Generic Syntax)
      # RFC 2732 (IPv6 Literal Addresses in URL's)
      # RFC 2373 (IPv6 Addressing Architecture)

      # alpha         = lowalpha | upalpha
      ALPHA = "a-zA-Z"
      # alphanum      = alpha | digit
      ALNUM = "#{ALPHA}\\d"

      # hex           = digit | "A" | "B" | "C" | "D" | "E" | "F" |
      #                         "a" | "b" | "c" | "d" | "e" | "f"
      HEX     = "a-fA-F\\d"
      # escaped       = "%" hex hex
      ESCAPED = "%[#{HEX}]{2}"
      # mark          = "-" | "_" | "." | "!" | "~" | "*" | "'" |
      #                 "(" | ")"
      # unreserved    = alphanum | mark
      UNRESERVED = "-_.!~*'()#{ALNUM}"
      # reserved      = ";" | "/" | "?" | ":" | "@" | "&" | "=" | "+" |
      #                 "$" | ","
      # reserved      = ";" | "/" | "?" | ":" | "@" | "&" | "=" | "+" | 
      #                 "$" | "," | "[" | "]" (RFC 2732)
      RESERVED = ";/?:@&=+$,\\[\\]"

      # uric          = reserved | unreserved | escaped
      URIC = "(?:[#{UNRESERVED}#{RESERVED}]|#{ESCAPED})"
      # uric_no_slash = unreserved | escaped | ";" | "?" | ":" | "@" |
      #                 "&" | "=" | "+" | "$" | ","
      URIC_NO_SLASH = "(?:[#{UNRESERVED};?:@&=+$,]|#{ESCAPED})"
      # query         = *uric
      QUERY = "#{URIC}*"
      # fragment      = *uric
      FRAGMENT = "#{URIC}*"

      # domainlabel   = alphanum | alphanum *( alphanum | "-" ) alphanum
      DOMLABEL = "(?:[#{ALNUM}](?:[-#{ALNUM}]*[#{ALNUM}])?)"
      # toplabel      = alpha | alpha *( alphanum | "-" ) alphanum
      TOPLABEL = "(?:[#{ALPHA}](?:[-#{ALNUM}]*[#{ALNUM}])?)"
      # hostname      = *( domainlabel "." ) toplabel [ "." ]
      HOSTNAME = "(?:#{DOMLABEL}\\.)*#{TOPLABEL}\\.?"

      # RFC 2373, APPENDIX B:
      # IPv6address = hexpart [ ":" IPv4address ]
      # IPv4address   = 1*3DIGIT "." 1*3DIGIT "." 1*3DIGIT "." 1*3DIGIT
      # hexpart = hexseq | hexseq "::" [ hexseq ] | "::" [ hexseq ]
      # hexseq  = hex4 *( ":" hex4)
      # hex4    = 1*4HEXDIG
      #
      # XXX: This definition has a flaw. "::" + IPv4address must be
      # allowed too.  Here is a replacement.
      #
      # IPv4address = 1*3DIGIT "." 1*3DIGIT "." 1*3DIGIT "." 1*3DIGIT
      IPV4ADDR = "\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}"
      # hex4     = 1*4HEXDIG
      HEX4 = "[#{HEX}]{1,4}"
      # lastpart = hex4 | IPv4address
      LASTPART = "(?:#{HEX4}|#{IPV4ADDR})"
      # hexseq1  = *( hex4 ":" ) hex4
      HEXSEQ1 = "(?:#{HEX4}:)*#{HEX4}"
      # hexseq2  = *( hex4 ":" ) lastpart
      HEXSEQ2 = "(?:#{HEX4}:)*#{LASTPART}"
      # IPv6address = hexseq2 | [ hexseq1 ] "::" [ hexseq2 ]
      IPV6ADDR = "(?:#{HEXSEQ2}|(?:#{HEXSEQ1})?::(?:#{HEXSEQ2})?)"

      # IPv6prefix  = ( hexseq1 | [ hexseq1 ] "::" [ hexseq1 ] ) "/" 1*2DIGIT
      # unused

      # ipv6reference = "[" IPv6address "]" (RFC 2732)
      IPV6REF = "\\[#{IPV6ADDR}\\]"

      # host          = hostname | IPv4address
      # host          = hostname | IPv4address | IPv6reference (RFC 2732)
      HOST = "(?:#{HOSTNAME}|#{IPV4ADDR}|#{IPV6REF})"
      # port          = *digit
      PORT = '\d*'
      # hostport      = host [ ":" port ]
      HOSTPORT = "#{HOST}(?::#{PORT})?"

      # userinfo      = *( unreserved | escaped |
      #                    ";" | ":" | "&" | "=" | "+" | "$" | "," )
      USERINFO = "(?:[#{UNRESERVED};:&=+$,]|#{ESCAPED})*"

      # pchar         = unreserved | escaped |
      #                 ":" | "@" | "&" | "=" | "+" | "$" | ","
      PCHAR = "(?:[#{UNRESERVED}:@&=+$,]|#{ESCAPED})"
      # param         = *pchar
      PARAM = "#{PCHAR}*"
      # segment       = *pchar *( ";" param )
      SEGMENT = "#{PCHAR}*(?:;#{PARAM})*"
      # path_segments = segment *( "/" segment )
      PATH_SEGMENTS = "#{SEGMENT}(?:/#{SEGMENT})*"

      # server        = [ [ userinfo "@" ] hostport ]
      SERVER = "(?:#{USERINFO}@)?#{HOSTPORT}"
      # reg_name      = 1*( unreserved | escaped | "$" | "," |
      #                     ";" | ":" | "@" | "&" | "=" | "+" )
      REG_NAME = "(?:[#{UNRESERVED}$,;:@&=+]|#{ESCAPED})+"
      # authority     = server | reg_name
      AUTHORITY = "(?:#{SERVER}|#{REG_NAME})"

      # rel_segment   = 1*( unreserved | escaped |
      #                     ";" | "@" | "&" | "=" | "+" | "$" | "," )
      REL_SEGMENT = "(?:[#{UNRESERVED};@&=+$,]|#{ESCAPED})+"

      # scheme        = alpha *( alpha | digit | "+" | "-" | "." )
      SCHEME = "[#{ALPHA}][-+.#{ALPHA}\\d]*"

      # abs_path      = "/"  path_segments
      ABS_PATH = "/#{PATH_SEGMENTS}"
      # rel_path      = rel_segment [ abs_path ]
      REL_PATH = "#{REL_SEGMENT}(?:#{ABS_PATH})?"
      # net_path      = "//" authority [ abs_path ]
      NET_PATH   = "//#{AUTHORITY}(?:#{ABS_PATH})?"

      # hier_part     = ( net_path | abs_path ) [ "?" query ]
      HIER_PART   = "(?:#{NET_PATH}|#{ABS_PATH})(?:\\?(?:#{QUERY}))?"
      # opaque_part   = uric_no_slash *uric
      OPAQUE_PART = "#{URIC_NO_SLASH}#{URIC}*"

      # absoluteURI   = scheme ":" ( hier_part | opaque_part )
      ABS_URI   = "#{SCHEME}:(?:#{HIER_PART}|#{OPAQUE_PART})"
      # relativeURI   = ( net_path | abs_path | rel_path ) [ "?" query ]
      REL_URI = "(?:#{NET_PATH}|#{ABS_PATH}|#{REL_PATH})(?:\\?#{QUERY})?"

      # URI-reference = [ absoluteURI | relativeURI ] [ "#" fragment ]
      URI_REF = "(?:#{ABS_URI}|#{REL_URI})?(?:##{FRAGMENT})?"

      # XXX:
      X_ABS_URI = "
        (#{PATTERN::SCHEME}):                     (?# 1: scheme)
        (?:
           (#{PATTERN::OPAQUE_PART})              (?# 2: opaque)
        |
           (?:(?:
             //(?:
                 (?:(?:(#{PATTERN::USERINFO})@)?  (?# 3: userinfo)
                   (?:(#{PATTERN::HOST})(?::(\\d*))?))?(?# 4: host, 5: port)
               |
                 (#{PATTERN::REG_NAME})           (?# 6: registry)
               )
             |
             (?!//))                              (?# XXX: '//' is the mark for hostport)
             (#{PATTERN::ABS_PATH})?              (?# 7: path)
           )(?:\\?(#{PATTERN::QUERY}))?           (?# 8: query)
        )
        (?:\\#(#{PATTERN::FRAGMENT}))?            (?# 9: fragment)
      "
      X_REL_URI = "
        (?:
          (?:
            //
            (?:
              (?:(#{PATTERN::USERINFO})@)?       (?# 1: userinfo)
                (#{PATTERN::HOST})?(?::(\\d*))?  (?# 2: host, 3: port)
            |
              (#{PATTERN::REG_NAME})             (?# 4: registry)
            )
          )
        |
          (#{PATTERN::REL_SEGMENT})              (?# 5: rel_segment)
        )?
        (#{PATTERN::ABS_PATH})?                  (?# 6: abs_path)
        (?:\\?(#{PATTERN::QUERY}))?              (?# 7: query)
        (?:\\#(#{PATTERN::FRAGMENT}))?           (?# 8: fragment)
      "
      # :startdoc:
    end # PATTERN

    # :stopdoc:

    # for URI::split
    ABS_URI = Regexp.new('^' + PATTERN::X_ABS_URI + '$', #'
                         Regexp::EXTENDED, 'N').freeze
    REL_URI = Regexp.new('^' + PATTERN::X_REL_URI + '$', #'
                         Regexp::EXTENDED, 'N').freeze

    # for URI::extract
    URI_REF     = Regexp.new(PATTERN::URI_REF, false, 'N').freeze
    ABS_URI_REF = Regexp.new(PATTERN::X_ABS_URI, Regexp::EXTENDED, 'N').freeze
    REL_URI_REF = Regexp.new(PATTERN::X_REL_URI, Regexp::EXTENDED, 'N').freeze

    # for URI::escape/unescape
    ESCAPED = Regexp.new(PATTERN::ESCAPED, false, 'N').freeze
    UNSAFE  = Regexp.new("[^#{PATTERN::UNRESERVED}#{PATTERN::RESERVED}]",
                         false, 'N').freeze

    # for Generic#initialize
    SCHEME   = Regexp.new("^#{PATTERN::SCHEME}$", false, 'N').freeze #"
    USERINFO = Regexp.new("^#{PATTERN::USERINFO}$", false, 'N').freeze #"
    HOST     = Regexp.new("^#{PATTERN::HOST}$", false, 'N').freeze #"
    PORT     = Regexp.new("^#{PATTERN::PORT}$", false, 'N').freeze #"
    OPAQUE   = Regexp.new("^#{PATTERN::OPAQUE_PART}$", false, 'N').freeze #"
    REGISTRY = Regexp.new("^#{PATTERN::REG_NAME}$", false, 'N').freeze #"
    ABS_PATH = Regexp.new("^#{PATTERN::ABS_PATH}$", false, 'N').freeze #"
    REL_PATH = Regexp.new("^#{PATTERN::REL_PATH}$", false, 'N').freeze #"
    QUERY    = Regexp.new("^#{PATTERN::QUERY}$", false, 'N').freeze #"
    FRAGMENT = Regexp.new("^#{PATTERN::FRAGMENT}$", false, 'N').freeze #"
    # :startdoc:
  end # REGEXP

  module Util # :nodoc:
    def make_components_hash(klass, array_hash)
      tmp = {}
      if array_hash.kind_of?(Array) &&
          array_hash.size == klass.component.size - 1
        klass.component[1..-1].each_index do |i|
          begin
            tmp[klass.component[i + 1]] = array_hash[i].clone
          rescue TypeError
            tmp[klass.component[i + 1]] = array_hash[i]
          end
        end

      elsif array_hash.kind_of?(Hash)
        array_hash.each do |key, value|
          begin
            tmp[key] = value.clone
          rescue TypeError
            tmp[key] = value
          end
        end
      else
        raise ArgumentError, 
          "expected Array of or Hash of components of #{klass.to_s} (#{klass.component[1..-1].join(', ')})"
      end
      tmp[:scheme] = klass.to_s.sub(/\A.*::/, '').downcase

      return tmp
    end
    module_function :make_components_hash
  end

  module Escape
    include REGEXP

    #
    # == Synopsis
    #
    #   URI.escape(str [, unsafe])
    #
    # == Args
    #
    # +str+::
    #   String to replaces in.
    # +unsafe+::
    #   Regexp that matches all symbols that must be replaced with codes.
    #   By default uses <tt>REGEXP::UNSAFE</tt>.
    #   When this argument is a String, it represents a character set.
    #
    # == Description
    #
    # Escapes the string, replacing all unsafe characters with codes.
    #
    # == Usage
    #
    #   require 'uri'
    #
    #   enc_uri = URI.escape("http://example.com/?a=\11\15")
    #   p enc_uri
    #   # => "http://example.com/?a=%09%0D"
    #
    #   p URI.unescape(enc_uri)
    #   # => "http://example.com/?a=\t\r"
    #
    #   p URI.escape("@?@!", "!?")
    #   # => "@%3F@%21"
    #
    def escape(str, unsafe = UNSAFE)
      unless unsafe.kind_of?(Regexp)
        # perhaps unsafe is String object
        unsafe = Regexp.new("[#{Regexp.quote(unsafe)}]", false, 'N')
      end
      str.gsub(unsafe) do |us|
        tmp = ''
        us.each_byte do |uc|
          tmp << sprintf('%%%02X', uc)
        end
        tmp
      end
    end
    alias encode escape
    #
    # == Synopsis
    #
    #   URI.unescape(str)
    #
    # == Args
    #
    # +str+::
    #   Unescapes the string.
    #
    # == Usage
    #
    #   require 'uri'
    #
    #   enc_uri = URI.escape("http://example.com/?a=\11\15")
    #   p enc_uri
    #   # => "http://example.com/?a=%09%0D"
    #
    #   p URI.unescape(enc_uri)
    #   # => "http://example.com/?a=\t\r"
    #
    def unescape(str)
      str.gsub(ESCAPED) do
        $&[1,2].hex.chr
      end
    end
    alias decode unescape
  end

  include REGEXP
  extend Escape

  @@schemes = {}
  
  #
  # Base class for all URI exceptions.
  #
  class Error < StandardError; end
  #
  # Not a URI.
  #
  class InvalidURIError < Error; end
  #
  # Not a URI component.
  #
  class InvalidComponentError < Error; end
  #
  # URI is valid, bad usage is not.
  #
  class BadURIError < Error; end

  #
  # == Synopsis
  #
  #   URI::split(uri)
  #
  # == Args
  #
  # +uri+::
  #   String with URI.
  #
  # == Description
  #
  # Splits the string on following parts and returns array with result:
  #
  #   * Scheme
  #   * Userinfo
  #   * Host
  #   * Port
  #   * Registry
  #   * Path
  #   * Opaque
  #   * Query
  #   * Fragment
  # 
  # == Usage
  #
  #   require 'uri'
  #
  #   p URI.split("http://www.ruby-lang.org/")
  #   # => ["http", nil, "www.ruby-lang.org", nil, nil, "/", nil, nil, nil]
  #
  def self.split(uri)
    case uri
    when ''
      # null uri

    when ABS_URI
      scheme, opaque, userinfo, host, port, 
        registry, path, query, fragment = $~[1..-1]

      # URI-reference = [ absoluteURI | relativeURI ] [ "#" fragment ]

      # absoluteURI   = scheme ":" ( hier_part | opaque_part )
      # hier_part     = ( net_path | abs_path ) [ "?" query ]
      # opaque_part   = uric_no_slash *uric

      # abs_path      = "/"  path_segments
      # net_path      = "//" authority [ abs_path ]

      # authority     = server | reg_name
      # server        = [ [ userinfo "@" ] hostport ]

      if !scheme
        raise InvalidURIError, 
          "bad URI(absolute but no scheme): #{uri}"
      end
      if !opaque && (!path && (!host && !registry))
        raise InvalidURIError,
          "bad URI(absolute but no path): #{uri}" 
      end

    when REL_URI
      scheme = nil
      opaque = nil

      userinfo, host, port, registry, 
        rel_segment, abs_path, query, fragment = $~[1..-1]
      if rel_segment && abs_path
        path = rel_segment + abs_path
      elsif rel_segment
        path = rel_segment
      elsif abs_path
        path = abs_path
      end

      # URI-reference = [ absoluteURI | relativeURI ] [ "#" fragment ]

      # relativeURI   = ( net_path | abs_path | rel_path ) [ "?" query ]

      # net_path      = "//" authority [ abs_path ]
      # abs_path      = "/"  path_segments
      # rel_path      = rel_segment [ abs_path ]

      # authority     = server | reg_name
      # server        = [ [ userinfo "@" ] hostport ]

    else
      raise InvalidURIError, "bad URI(is not URI?): #{uri}"
    end

    path = '' if !path && !opaque # (see RFC2396 Section 5.2)
    ret = [
      scheme, 
      userinfo, host, port,         # X
      registry,                        # X
      path,                         # Y
      opaque,                        # Y
      query,
      fragment
    ]
    return ret
  end

  #
  # == Synopsis
  #
  #   URI::parse(uri_str)
  #
  # == Args
  #
  # +uri_str+::
  #   String with URI.
  #
  # == Description
  #
  # Creates one of the URI's subclasses instance from the string.
  #  
  # == Raises
  #
  # URI::InvalidURIError
  #   Raised if URI given is not a correct one.
  #
  # == Usage
  #
  #   require 'uri'
  #
  #   uri = URI.parse("http://www.ruby-lang.org/")
  #   p uri
  #   # => #<URI::HTTP:0x202281be URL:http://www.ruby-lang.org/>
  #   p uri.scheme 
  #   # => "http" 
  #   p uri.host 
  #   # => "www.ruby-lang.org" 
  # 
  def self.parse(uri)
    scheme, userinfo, host, port, 
      registry, path, opaque, query, fragment = self.split(uri)

    if scheme && @@schemes.include?(scheme.upcase)
      @@schemes[scheme.upcase].new(scheme, userinfo, host, port, 
                                   registry, path, opaque, query, 
                                   fragment)
    else
      Generic.new(scheme, userinfo, host, port, 
                  registry, path, opaque, query, 
                  fragment)
    end
  end

  #
  # == Synopsis
  #
  #   URI::join(str[, str, ...])
  #
  # == Args
  #
  # +str+::
  #   String(s) to work with
  #
  # == Description
  #
  # Joins URIs.
  #
  # == Usage
  #
  #   require 'uri'
  #
  #   p URI.join("http://localhost/","main.rbx")
  #   # => #<URI::HTTP:0x2022ac02 URL:http://localhost/main.rbx>
  #
  def self.join(*str)
    u = self.parse(str[0])
    str[1 .. -1].each do |x|
      u = u.merge(x)
    end
    u
  end

  #
  # == Synopsis
  #
  #   URI::extract(str[, schemes][,&blk])
  #
  # == Args
  #
  # +str+:: 
  #   String to extract URIs from.
  # +schemes+::
  #   Limit URI matching to a specific schemes.
  #
  # == Description
  #
  # Extracts URIs from a string. If block given, iterates through all matched URIs.
  # Returns nil if block given or array with matches.
  #
  # == Usage
  #
  #   require "uri"
  #
  #   URI.extract("text here http://foo.example.org/bla and here mailto:test@example.com and here also.")
  #   # => ["http://foo.example.com/bla", "mailto:test@example.com"]
  #
  def self.extract(str, schemes = nil, &block)
    if block_given?
      str.scan(regexp(schemes)) { yield $& }
      nil
    else
      result = []
      str.scan(regexp(schemes)) { result.push $& }
      result
    end
  end

  #
  # == Synopsis
  #
  #   URI::regexp([match_schemes])
  #
  # == Args
  #
  # +match_schemes+:: 
  #   Array of schemes. If given, resulting regexp matches to URIs
  #   whose scheme is one of the match_schemes.
  # 
  # == Description
  # Returns a Regexp object which matches to URI-like strings.
  # The Regexp object returned by this method includes arbitrary
  # number of capture group (parentheses).  Never rely on it's number.
  # 
  # == Usage
  #
  #   require 'uri'
  #
  #   # extract first URI from html_string
  #   html_string.slice(URI.regexp)
  # 
  #   # remove ftp URIs
  #   html_string.sub(URI.regexp(['ftp'])
  # 
  #   # You should not rely on the number of parentheses
  #   html_string.scan(URI.regexp) do |*matches|
  #     p $&
  #   end
  #
  def self.regexp(schemes = nil)
    unless schemes
      ABS_URI_REF
    else
      /(?=#{Regexp.union(*schemes)}:)#{PATTERN::X_ABS_URI}/xn
    end
  end

end

module Kernel
  # alias for URI.parse.
  #
  # This method is introduced at 1.8.2.
  def URI(uri_str) # :doc:
    URI.parse(uri_str)
  end
  module_function :URI
end

Youez - 2016 - github.com/yon3zu
LinuXploit