careo-eventmachine 0.12.5.1

data/lib/protocols/buftok.rb

@@ -0,0 +1,127 @@
+# BufferedTokenizer - Statefully split input data by a specifiable token
+#
+# Authors:: Tony Arcieri, Martin Emde
+#
+#----------------------------------------------------------------------------
+#
+# Copyright (C) 2006-07 by Tony Arcieri and Martin Emde
+# 
+# Distributed under the Ruby license (http://www.ruby-lang.org/en/LICENSE.txt)
+#
+#---------------------------------------------------------------------------
+#
+
+# (C)2006 Tony Arcieri, Martin Emde
+# Distributed under the Ruby license (http://www.ruby-lang.org/en/LICENSE.txt)
+
+# BufferedTokenizer takes a delimiter upon instantiation, or acts line-based
+# by default.  It allows input to be spoon-fed from some outside source which
+# receives arbitrary length datagrams which may-or-may-not contain the token
+# by which entities are delimited.
+
+class BufferedTokenizer
+  # New BufferedTokenizers will operate on lines delimited by "\n" by default
+  # or allow you to specify any delimiter token you so choose, which will then
+  # be used by String#split to tokenize the input data
+  def initialize(delimiter = "\n", size_limit = nil)
+    # Store the specified delimiter
+    @delimiter = delimiter
+
+    # Store the specified size limitation
+    @size_limit = size_limit
+
+    # The input buffer is stored as an array.  This is by far the most efficient
+    # approach given language constraints (in C a linked list would be a more
+    # appropriate data structure).  Segments of input data are stored in a list
+    # which is only joined when a token is reached, substantially reducing the
+    # number of objects required for the operation.
+    @input = []
+
+    # Size of the input buffer
+    @input_size = 0
+  end
+
+  # Extract takes an arbitrary string of input data and returns an array of
+  # tokenized entities, provided there were any available to extract.  This
+  # makes for easy processing of datagrams using a pattern like:
+  #
+  #   tokenizer.extract(data).map { |entity| Decode(entity) }.each do ...
+  def extract(data)
+    # Extract token-delimited entities from the input string with the split command.
+    # There's a bit of craftiness here with the -1 parameter.  Normally split would
+    # behave no differently regardless of if the token lies at the very end of the 
+    # input buffer or not (i.e. a literal edge case)  Specifying -1 forces split to
+    # return "" in this case, meaning that the last entry in the list represents a
+    # new segment of data where the token has not been encountered
+    entities = data.split @delimiter, -1
+
+    # Check to see if the buffer has exceeded capacity, if we're imposing a limit
+    if @size_limit
+      raise 'input buffer full' if @input_size + entities.first.size > @size_limit
+      @input_size += entities.first.size
+    end
+    
+    # Move the first entry in the resulting array into the input buffer.  It represents
+    # the last segment of a token-delimited entity unless it's the only entry in the list.
+    @input << entities.shift
+
+    # If the resulting array from the split is empty, the token was not encountered
+    # (not even at the end of the buffer).  Since we've encountered no token-delimited
+    # entities this go-around, return an empty array.
+    return [] if entities.empty?
+
+    # At this point, we've hit a token, or potentially multiple tokens.  Now we can bring
+    # together all the data we've buffered from earlier calls without hitting a token,
+    # and add it to our list of discovered entities.
+    entities.unshift @input.join
+
+=begin
+	# Note added by FC, 10Jul07. This paragraph contains a regression. It breaks
+	# empty tokens. Think of the empty line that delimits an HTTP header. It will have
+	# two "\n" delimiters in a row, and this code mishandles the resulting empty token.
+	# It someone figures out how to fix the problem, we can re-enable this code branch.
+    # Multi-character token support.
+    # Split any tokens that were incomplete on the last iteration buf complete now.
+    entities.map! do |e|
+      e.split @delimiter, -1
+    end
+    # Flatten the resulting array.  This has the side effect of removing the empty
+    # entry at the end that was produced by passing -1 to split.  Add it again if
+    # necessary.
+    if (entities[-1] == [])
+      entities.flatten! << []
+    else
+      entities.flatten!
+    end
+=end
+
+    # Now that we've hit a token, joined the input buffer and added it to the entities
+    # list, we can go ahead and clear the input buffer.  All of the segments that were
+    # stored before the join can now be garbage collected.
+    @input.clear
+    
+    # The last entity in the list is not token delimited, however, thanks to the -1
+    # passed to split.  It represents the beginning of a new list of as-yet-untokenized  
+    # data, so we add it to the start of the list.
+    @input << entities.pop
+    
+    # Set the new input buffer size, provided we're keeping track
+    @input_size = @input.first.size if @size_limit
+
+    # Now we're left with the list of extracted token-delimited entities we wanted
+    # in the first place.  Hooray!
+    entities
+  end
+  
+  # Flush the contents of the input buffer, i.e. return the input buffer even though
+  # a token has not yet been encountered
+  def flush
+    buffer = @input.join
+    @input.clear
+    buffer
+  end
+
+  def empty?
+    @input.empty?
+  end
+end

data/lib/protocols/header_and_content.rb

@@ -0,0 +1,129 @@
+# $Id$
+#
+# Author:: Francis Cianfrocca (gmail: blackhedd)
+# Homepage::  http://rubyeventmachine.com
+# Date:: 15 Nov 2006
+# 
+# See EventMachine and EventMachine::Connection for documentation and
+# usage examples.
+#
+#----------------------------------------------------------------------------
+#
+# Copyright (C) 2006-07 by Francis Cianfrocca. All Rights Reserved.
+# Gmail: blackhedd
+# 
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of either: 1) the GNU General Public License
+# as published by the Free Software Foundation; either version 2 of the
+# License, or (at your option) any later version; or 2) Ruby's License.
+# 
+# See the file COPYING for complete licensing information.
+#
+#---------------------------------------------------------------------------
+#
+# 
+
+
+
+module EventMachine
+    module Protocols
+
+	# Originally, this subclassed LineAndTextProtocol, which in
+	# turn relies on BufferedTokenizer, which doesn't gracefully
+	# handle the transitions between lines and binary text.
+	# Changed 13Sep08 by FCianfrocca.
+	class HeaderAndContentProtocol < Connection
+	    include LineText2
+	
+
+	    ContentLengthPattern = /Content-length:\s*(\d+)/i
+
+	    def initialize *args
+		super
+		init_for_request
+	    end
+
+	    def receive_line line
+		case @hc_mode
+		when :discard_blanks
+		    unless line == ""
+			@hc_mode = :headers
+			receive_line line
+		    end
+		when :headers
+		    if line == ""
+			raise "unrecognized state" unless @hc_headers.length > 0
+			if respond_to?(:receive_headers)
+			    receive_headers @hc_headers
+			end
+			# @hc_content_length will be nil, not 0, if there was no content-length header.
+			if @hc_content_length.to_i > 0
+			    set_binary_mode @hc_content_length
+			else
+			    dispatch_request
+			end
+		    else
+			@hc_headers << line
+			if ContentLengthPattern =~ line
+			    # There are some attacks that rely on sending multiple content-length
+			    # headers. This is a crude protection, but needs to become tunable.
+			    raise "extraneous content-length header" if @hc_content_length
+			    @hc_content_length = $1.to_i
+			end
+			if @hc_headers.length == 1 and respond_to?(:receive_first_header_line)
+			    receive_first_header_line line
+			end
+		    end
+		else
+		    raise "internal error, unsupported mode"
+		end
+	    end
+
+	    def receive_binary_data text
+		@hc_content = text
+		dispatch_request
+	    end
+
+	    def dispatch_request
+		if respond_to?(:receive_request)
+		    receive_request @hc_headers, @hc_content
+		end
+		init_for_request
+	    end
+	    private :dispatch_request
+
+	    def init_for_request
+		@hc_mode = :discard_blanks
+		@hc_headers = []
+		# originally was @hc_headers ||= []; @hc_headers.clear to get a performance
+		# boost, but it's counterproductive because a subclassed handler will have to
+		# call dup to use the header array we pass in receive_headers.
+
+		@hc_content_length = nil
+		@hc_content = ""
+	    end
+	    private :init_for_request
+
+	    # Basically a convenience method. We might create a subclass that does this
+	    # automatically. But it's such a performance killer.
+	    def headers_2_hash hdrs
+		self.class.headers_2_hash hdrs
+	    end
+
+	    class << self
+		def headers_2_hash hdrs
+		    hash = {}
+		    hdrs.each {|h|
+			if /\A([^\s:]+)\s*:\s*/ =~ h
+			    tail = $'.dup
+			    hash[ $1.downcase.gsub(/-/,"_").intern ] = tail
+			end
+		    }
+		    hash
+		end
+	    end
+
+	end
+    end
+end
+

data/lib/protocols/httpcli2.rb

@@ -0,0 +1,794 @@
+# $Id$
+#
+# Author:: Francis Cianfrocca (gmail: blackhedd)
+# Homepage::  http://rubyeventmachine.com
+# Date:: 16 July 2006
+# 
+# See EventMachine and EventMachine::Connection for documentation and
+# usage examples.
+#
+#----------------------------------------------------------------------------
+#
+# Copyright (C) 2006-07 by Francis Cianfrocca. All Rights Reserved.
+# Gmail: blackhedd
+# 
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of either: 1) the GNU General Public License
+# as published by the Free Software Foundation; either version 2 of the
+# License, or (at your option) any later version; or 2) Ruby's License.
+# 
+# See the file COPYING for complete licensing information.
+#
+#---------------------------------------------------------------------------
+#
+# 
+
+
+
+module EventMachine
+module Protocols
+
+  # = Example
+  #
+  #  
+  #  EM.run{
+  #    include EM::Protocols
+  #    conn = HttpClient2.connect 'google.com', 80
+  #  
+  #    req = conn.get('/')
+  #    req.callback{
+  #      p(req.content)
+  #    }
+	class HttpClient2 < Connection
+		include LineText2
+
+
+		class Request
+			include Deferrable
+
+			attr_reader :version
+			attr_reader :status
+			attr_reader :header_lines
+			attr_reader :headers
+			attr_reader :content
+			attr_reader :internal_error
+
+			def initialize conn, args
+				@conn = conn
+				@args = args
+				@header_lines = []
+				@headers = {}
+				@blanks = 0
+			end
+
+			def send_request
+				az = @args[:authorization] and az = "Authorization: #{az}\r\n"
+
+				r = [
+					"#{@args[:verb]} #{@args[:uri]} HTTP/#{@args[:version] || "1.1"}\r\n",
+					"Host: #{@args[:host_header] || "_"}\r\n",
+					az || "",
+					"\r\n"
+				]
+				@conn.send_data r.join
+			end
+
+
+			#--
+			#
+			def receive_line ln
+				if @chunk_trailer
+					receive_chunk_trailer(ln)
+				elsif @chunking
+					receive_chunk_header(ln)
+				else
+					receive_header_line(ln)
+				end
+			end
+
+			#--
+			#
+			def receive_chunk_trailer ln
+				if ln.length == 0
+					@conn.pop_request
+					succeed
+				else
+					p "Received chunk trailer line"
+				end
+			end
+
+			#--
+			# Allow up to ten blank lines before we get a real response line.
+			# Allow no more than 100 lines in the header.
+			#
+			def receive_header_line ln
+				if ln.length == 0
+					if @header_lines.length > 0
+						process_header
+					else
+						@blanks += 1
+						if @blanks > 10
+							@conn.close_connection
+						end
+					end
+				else
+					@header_lines << ln
+					if @header_lines.length > 100
+						@internal_error = :bad_header
+						@conn.close_connection
+					end
+				end
+			end
+
+			#--
+			# Cf RFC 2616 pgh 3.6.1 for the format of HTTP chunks.
+			#
+			def receive_chunk_header ln
+				if ln.length > 0
+					chunksize = ln.to_i(16)
+					if chunksize > 0
+						@conn.set_text_mode(ln.to_i(16))
+					else
+						@content = @content ? @content.join : ''
+						@chunk_trailer = true
+					end
+				else
+					# We correctly come here after each chunk gets read.
+					# p "Got A BLANK chunk line"
+				end
+
+			end
+
+
+			#--
+			# We get a single chunk. Append it to the incoming content and switch back to line mode.
+			#
+			def receive_chunked_text text
+				# p "RECEIVED #{text.length} CHUNK"
+				(@content ||= []) << text
+			end
+
+
+			#--
+			# TODO, inefficient how we're handling this. Part of it is done so as to
+			# make sure we don't have problems in detecting chunked-encoding, content-length,
+			# etc.
+			#
+			#
+			HttpResponseRE = /\AHTTP\/(1.[01]) ([\d]{3})/i
+			ClenRE = /\AContent-length:\s*(\d+)/i
+			ChunkedRE = /\ATransfer-encoding:\s*chunked/i
+			ColonRE = /\:\s*/
+
+			def process_header
+				unless @header_lines.first =~ HttpResponseRE
+					@conn.close_connection
+					@internal_error = :bad_request
+				end
+				@version = $1.dup
+				@status = $2.dup.to_i
+
+				clen = nil
+				chunks = nil
+				@header_lines.each_with_index do |e,ix|
+					if ix > 0
+						hdr,val = e.split(ColonRE,2)
+						(@headers[hdr.downcase] ||= []) << val
+					end
+
+					if clen == nil and e =~ ClenRE
+						clen = $1.dup.to_i
+					end
+					if e =~ ChunkedRE
+						chunks = true
+					end
+				end
+
+				if clen
+					@conn.set_text_mode clen
+				elsif chunks
+					@chunking = true
+				else
+					# Chunked transfer, multipart, or end-of-connection.
+					# For end-of-connection, we need to go the unbind
+					# method and suppress its desire to fail us.
+					p "NO CLEN"
+					p @args[:uri]
+					p @header_lines
+					@internal_error = :unsupported_clen
+					@conn.close_connection
+				end
+			end
+			private :process_header
+
+
+			def receive_text text
+				@chunking ? receive_chunked_text(text) : receive_sized_text(text)
+			end
+
+			#--
+			# At the present time, we only handle contents that have a length
+			# specified by the content-length header.
+			#
+			def receive_sized_text text
+				@content = text
+				@conn.pop_request
+				succeed
+			end
+		end
+
+		# Make a connection to a remote HTTP server.
+		# Can take either a pair of arguments (which will be interpreted as
+		# a hostname/ip-address and a port), or a hash.
+		# If the arguments are a hash, then supported values include:
+		#  :host => a hostname or ip-address;
+		#  :port => a port number
+		#--
+		# TODO, support optional encryption arguments like :ssl
+		def self.connect *args
+			if args.length == 2
+				args = {:host=>args[0], :port=>args[1]}
+			else
+				args = args.first
+			end
+
+			h,prt,ssl = args[:host], Integer(args[:port]), (args[:tls] || args[:ssl])
+			conn = EM.connect( h, prt, self )
+      conn.start_tls if ssl
+			conn.set_default_host_header( h, prt, ssl )
+			conn
+		end
+
+
+		#--
+		# Compute and remember a string to be used as the host header in HTTP requests
+		# unless the user overrides it with an argument to #request.
+		#
+		def set_default_host_header host, port, ssl
+			if (ssl and port != 443) or (!ssl and port != 80)
+				@host_header = "#{host}:#{port}"
+			else
+				@host_header = host
+			end
+		end
+
+
+		def post_init
+			super
+			@connected = EM::DefaultDeferrable.new
+		end
+
+		def connection_completed
+			super
+			@connected.succeed
+		end
+
+		#--
+		# All pending requests, if any, must fail.
+		# We might come here without ever passing through connection_completed
+		# in case we can't connect to the server. We'll also get here when the
+		# connection closes (either because the server closes it, or we close it
+		# due to detecting an internal error or security violation).
+		# In either case, run down all pending requests, if any, and signal failure
+		# on them.
+		#
+		# Set and remember a flag (@closed) so we can immediately fail any
+		# subsequent requests.
+		#
+		def unbind
+			super
+			@closed = true
+			(@requests || []).each {|r| r.fail}
+		end
+
+
+		def get args
+			if args.is_a?(String)
+				args = {:uri=>args}
+			end
+			args[:verb] = "GET"
+			request args
+		end
+
+		def post args
+			if args.is_a?(String)
+				args = {:uri=>args}
+			end
+			args[:verb] = "POST"
+			request args
+		end
+
+		def request args
+			args[:host_header] = @host_header unless args.has_key?(:host_header)
+			args[:authorization] = @authorization unless args.has_key?(:authorization)
+			r = Request.new self, args
+			if @closed
+				r.fail
+			else
+				(@requests ||= []).unshift r
+				@connected.callback {r.send_request}
+			end
+			r
+		end
+
+		def receive_line ln
+			if req = @requests.last
+				req.receive_line ln
+			else
+				p "??????????"
+				p ln
+			end
+
+		end
+		def receive_binary_data text
+			@requests.last.receive_text text
+		end
+
+		#--
+		# Called by a Request object when it completes.
+		#
+		def pop_request
+			@requests.pop
+		end
+	end
+
+
+=begin
+	class HttpClient2x < Connection
+		include LineText2
+
+		# TODO: Make this behave appropriate in case a #connect fails.
+		# Currently, this produces no errors.
+
+		# Make a connection to a remote HTTP server.
+		# Can take either a pair of arguments (which will be interpreted as
+		# a hostname/ip-address and a port), or a hash.
+		# If the arguments are a hash, then supported values include:
+		#  :host => a hostname or ip-address;
+		#  :port => a port number
+		#--
+		# TODO, support optional encryption arguments like :ssl
+		def self.connect *args
+			if args.length == 2
+				args = {:host=>args[0], :port=>args[1]}
+			else
+				args = args.first
+			end
+
+			h,prt = args[:host],Integer(args[:port])
+			EM.connect( h, prt, self, h, prt )
+		end
+
+
+		#--
+		# Sugars a connection that makes a single request and then
+		# closes the connection. Matches the behavior and the arguments
+		# of the original implementation of class HttpClient.
+		#
+		# Intended primarily for back compatibility, but the idiom
+		# is probably useful so it's not deprecated.
+		# We return a Deferrable, as did the original implementation.
+		#
+		# Because we're improving the way we deal with errors and exceptions
+		# (specifically, HTTP response codes other than 2xx will trigger the
+		# errback rather than the callback), this may break some existing code.
+		#
+		def self.request args
+			c = connect args
+		end
+
+		#--
+		# Requests can be pipelined. When we get a request, add it to the
+		# front of a queue as an array. The last element of the @requests
+		# array is always the oldest request received. Each element of the
+		# @requests array is a two-element array consisting of a hash with
+		# the original caller's arguments, and an initially-empty Ostruct
+		# containing the data we retrieve from the server's response.
+		# Maintain the instance variable @current_response, which is the response
+		# of the oldest pending request. That's just to make other code a little
+		# easier. If the variable doesn't exist when we come here, we're
+		# obviously the first request being made on the connection.
+		#
+		# The reason for keeping this method private (and requiring use of the
+		# convenience methods #get, #post, #head, etc) is to avoid the small
+		# performance penalty of canonicalizing the verb.
+		#
+		def request args
+			d = EventMachine::DefaultDeferrable.new
+
+			if @closed
+				d.fail
+				return d
+			end
+
+			o = OpenStruct.new
+			o.deferrable = d
+			(@requests ||= []).unshift [args, o]
+			@current_response ||= @requests.last.last
+			@connected.callback {
+				az = args[:authorization] and az = "Authorization: #{az}\r\n"
+
+				r = [
+					"#{args[:verb]} #{args[:uri]} HTTP/#{args[:version] || "1.1"}\r\n",
+					"Host: #{args[:host_header] || @host_header}\r\n",
+					az || "",
+					"\r\n"
+				]
+				p r
+				send_data r.join
+			}
+			o.deferrable
+		end
+		private :request
+
+		def get args
+			if args.is_a?(String)
+				args = {:uri=>args}
+			end
+			args[:verb] = "GET"
+			request args
+		end
+
+		def initialize host, port
+			super
+			@host_header = "#{host}:#{port}"
+		end
+		def post_init
+			super
+			@connected = EM::DefaultDeferrable.new
+		end
+
+
+		def connection_completed
+			super
+			@connected.succeed
+		end
+
+		#--
+		# Make sure to throw away any leftover incoming data if we've
+		# been closed due to recognizing an error.
+		#
+		# Generate an internal error if we get an unreasonable number of
+		# header lines. It could be malicious.
+		#
+		def receive_line ln
+			p ln
+			return if @closed
+
+			if ln.length > 0
+				(@current_response.headers ||= []).push ln
+				abort_connection if @current_response.headers.length > 100
+			else
+				process_received_headers
+			end
+		end
+
+		#--
+		# We come here when we've seen all the headers for a particular request.
+		# What we do next depends on the response line (which should be the
+		# first line in the header set), and whether there is content to read.
+		# We may transition into a text-reading state to read content, or
+		# we may abort the connection, or we may go right back into parsing
+		# responses for the next response in the chain.
+		#
+		# We make an ASSUMPTION that the first line is an HTTP response.
+		# Anything else produces an error that aborts the connection.
+		# This may not be enough, because it may be that responses to pipelined
+		# requests will come with a blank-line delimiter.
+		#
+		# Any non-2xx response will be treated as a fatal error, and abort the
+		# connection. We will set up the status and other response parameters.
+		# TODO: we will want to properly support 1xx responses, which some versions
+		# of IIS copiously generate.
+		# TODO: We need to give the option of not aborting the connection with certain
+		# non-200 responses, in order to work with NTLM and other authentication
+		# schemes that work at the level of individual connections.
+		#
+		# Some error responses will get sugarings. For example, we'll return the
+		# Location header in the response in case of a 301/302 response.
+		#
+		# Possible dispositions here:
+		# 1) No content to read (either content-length is zero or it's a HEAD request);
+		# 2) Switch to text mode to read a specific number of bytes;
+		# 3) Read a chunked or multipart response;
+		# 4) Read till the server closes the connection.
+		#
+		# Our reponse to the client can be either to wait till all the content
+		# has been read and then to signal caller's deferrable, or else to signal
+		# it when we finish the processing the headers and then expect the caller
+		# to have given us a block to call as the content comes in. And of course
+		# the latter gets stickier with chunks and multiparts.
+		#
+		HttpResponseRE = /\AHTTP\/(1.[01]) ([\d]{3})/i
+		ClenRE = /\AContent-length:\s*(\d+)/i
+		def process_received_headers
+			abort_connection unless @current_response.headers.first =~ HttpResponseRE
+			@current_response.version = $1.dup
+			st = $2.dup
+			@current_response.status = st.to_i
+			abort_connection unless st[0,1] == "2"
+
+			clen = nil
+			@current_response.headers.each do |e|
+				if clen == nil and e =~ ClenRE
+					clen = $1.dup.to_i
+				end
+			end
+
+			if clen
+				set_text_mode clen
+			end
+		end
+		private :process_received_headers
+
+
+		def receive_binary_data text
+			@current_response.content = text
+			@current_response.deferrable.succeed @current_response
+			@requests.pop
+			@current_response = (@requests.last || []).last
+			set_line_mode
+		end
+
+
+
+		# We've received either a server error or an internal error.
+		# Close the connection and abort any pending requests.
+		#--
+		# When should we call close_connection? It will cause #unbind
+		# to be fired. Should the user expect to see #unbind before
+		# we call #receive_http_error, or the other way around?
+		#
+		# Set instance variable @closed. That's used to inhibit further
+		# processing of any inbound data after an error has been recognized.
+		#
+		# We shouldn't have to worry about any leftover outbound data,
+		# because we call close_connection (not close_connection_after_writing).
+		# That ensures that any pipelined requests received after an error
+		# DO NOT get streamed out to the server on this connection.
+		# Very important. TODO, write a unit-test to establish that behavior.
+		#
+		def abort_connection
+			close_connection
+			@closed = true
+			@current_response.deferrable.fail( @current_response )
+		end
+
+
+		#------------------------
+		# Below here are user-overridable methods.
+
+	end
+=end
+end
+end
+
+
+=begin
+module EventMachine
+module Protocols
+
+class HttpClient < Connection
+  include EventMachine::Deferrable
+
+
+    MaxPostContentLength = 20 * 1024 * 1024
+
+  # USAGE SAMPLE:
+  #
+  # EventMachine.run {
+  #   http = EventMachine::Protocols::HttpClient.request(
+  #     :host => server,
+  #     :port => 80,
+  #     :request => "/index.html",
+  #     :query_string => "parm1=value1&parm2=value2"
+  #   )
+  #   http.callback {|response|
+  #     puts response[:status]
+  #     puts response[:headers]
+  #     puts response[:content]
+  #   }
+  # }
+  #
+
+  # TODO:
+  # Add streaming so we can support enormous POSTs. Current max is 20meg.
+  # Timeout for connections that run too long or hang somewhere in the middle.
+  # Persistent connections (HTTP/1.1), may need a associated delegate object.
+  # DNS: Some way to cache DNS lookups for hostnames we connect to. Ruby's
+  # DNS lookups are unbelievably slow.
+  # HEAD requests.
+  # Chunked transfer encoding.
+  # Convenience methods for requests. get, post, url, etc.
+  # SSL.
+  # Handle status codes like 304, 100, etc.
+  # Refactor this code so that protocol errors all get handled one way (an exception?),
+  # instead of sprinkling set_deferred_status :failed calls everywhere.
+
+  def self.request( args = {} )
+    args[:port] ||= 80
+    EventMachine.connect( args[:host], args[:port], self ) {|c|
+      # According to the docs, we will get here AFTER post_init is called.
+      c.instance_eval {@args = args}
+    }
+  end
+
+  def post_init
+    @start_time = Time.now
+    @data = ""
+    @read_state = :base
+  end
+
+  # We send the request when we get a connection.
+  # AND, we set an instance variable to indicate we passed through here.
+  # That allows #unbind to know whether there was a successful connection.
+  # NB: This naive technique won't work when we have to support multiple
+  # requests on a single connection.
+  def connection_completed
+    @connected = true
+    send_request @args
+  end
+
+  def send_request args
+    args[:verb] ||= args[:method] # Support :method as an alternative to :verb.
+    args[:verb] ||= :get # IS THIS A GOOD IDEA, to default to GET if nothing was specified?
+
+    verb = args[:verb].to_s.upcase
+    unless ["GET", "POST", "PUT", "DELETE", "HEAD"].include?(verb)
+      set_deferred_status :failed, {:status => 0} # TODO, not signalling the error type
+      return # NOTE THE EARLY RETURN, we're not sending any data.
+    end
+
+    request = args[:request] || "/"
+    unless request[0,1] == "/"
+      request = "/" + request
+    end
+
+    qs = args[:query_string] || ""
+    if qs.length > 0 and qs[0,1] != '?'
+      qs = "?" + qs
+    end
+
+    # Allow an override for the host header if it's not the connect-string.
+    host = args[:host_header] || args[:host] || "_"
+    # For now, ALWAYS tuck in the port string, although we may want to omit it if it's the default.
+    port = args[:port]
+
+    # POST items.
+    postcontenttype = args[:contenttype] || "application/octet-stream"
+    postcontent = args[:content] || ""
+    raise "oversized content in HTTP POST" if postcontent.length > MaxPostContentLength
+
+    # ESSENTIAL for the request's line-endings to be CRLF, not LF. Some servers misbehave otherwise.
+    # TODO: We ASSUME the caller wants to send a 1.1 request. May not be a good assumption.
+    req = [
+      "#{verb} #{request}#{qs} HTTP/1.1",
+      "Host: #{host}:#{port}",
+      "User-agent: Ruby EventMachine",
+    ]
+
+    if verb == "POST" || verb == "PUT"
+      req << "Content-type: #{postcontenttype}"
+      req << "Content-length: #{postcontent.length}"
+    end
+
+    # TODO, this cookie handler assumes it's getting a single, semicolon-delimited string.
+    # Eventually we will want to deal intelligently with arrays and hashes.
+    if args[:cookie]
+      req << "Cookie: #{args[:cookie]}"
+    end
+
+    req << ""
+    reqstring = req.map {|l| "#{l}\r\n"}.join
+    send_data reqstring
+
+    if verb == "POST" || verb == "PUT"
+      send_data postcontent
+    end
+  end
+
+
+  def receive_data data
+    while data and data.length > 0
+      case @read_state
+      when :base
+        # Perform any per-request initialization here and don't consume any data.
+        @data = ""
+        @headers = []
+        @content_length = nil # not zero
+        @content = ""
+        @status = nil
+        @read_state = :header
+      when :header
+        ary = data.split( /\r?\n/m, 2 )
+        if ary.length == 2
+          data = ary.last
+          if ary.first == ""
+	      if @content_length and @content_length > 0
+		  @read_state = :content
+	      else
+		  dispatch_response
+		  @read_state = :base
+	      end
+          else
+            @headers << ary.first
+            if @headers.length == 1
+              parse_response_line
+            elsif ary.first =~ /\Acontent-length:\s*/i
+              # Only take the FIRST content-length header that appears,
+              # which we can distinguish because @content_length is nil.
+              # TODO, it's actually a fatal error if there is more than one
+              # content-length header, because the caller is presumptively
+              # a bad guy. (There is an exploit that depends on multiple
+              # content-length headers.)
+              @content_length ||= $'.to_i
+            end
+          end
+        else
+          @data << data
+          data = ""
+        end
+      when :content
+        # If there was no content-length header, we have to wait until the connection
+        # closes. Everything we get until that point is content.
+        # TODO: Must impose a content-size limit, and also must implement chunking.
+        # Also, must support either temporary files for large content, or calling
+        # a content-consumer block supplied by the user.
+        if @content_length
+          bytes_needed = @content_length - @content.length
+          @content += data[0, bytes_needed]
+          data = data[bytes_needed..-1] || ""
+          if @content_length == @content.length
+            dispatch_response
+            @read_state = :base
+          end
+        else
+          @content << data
+          data = ""
+        end
+      end
+    end
+  end
+
+
+  # We get called here when we have received an HTTP response line.
+  # It's an opportunity to throw an exception or trigger other exceptional
+  # handling.
+  def parse_response_line
+    if @headers.first =~ /\AHTTP\/1\.[01] ([\d]{3})/
+      @status = $1.to_i
+    else
+      set_deferred_status :failed, {
+        :status => 0 # crappy way of signifying an unrecognized response. TODO, find a better way to do this.
+      }
+      close_connection
+    end
+  end
+  private :parse_response_line
+
+  def dispatch_response
+    @read_state = :base
+    set_deferred_status :succeeded, {
+      :content => @content,
+      :headers => @headers,
+      :status => @status
+    }
+    # TODO, we close the connection for now, but this is wrong for persistent clients.
+    close_connection
+  end
+
+  def unbind
+    if !@connected
+      set_deferred_status :failed, {:status => 0} # YECCCCH. Find a better way to signal no-connect/network error.
+    elsif (@read_state == :content and @content_length == nil)
+      dispatch_response
+    end
+  end
+end
+
+
+end
+end
+
+=end
+