Python网络编程：asynchat握手

Question

我正在用Python的asynchat库来实现一个网络协议。在连接的时候，我需要发送一个命令，然后服务器会回复我一个会话。

我遇到的主要问题是，我需要等到收到会话的回复，但我不太确定该怎么做。是不是应该在连接设置的时候用socket.recv？这样做合适吗？

Answer 1

在用异步技术编写网络应用时，你需要通过记录你的状态来“等待”，然后让主循环继续运行。过一段时间后，你等待的数据会变得可用，主循环会通知你这个消息，然后你可以把新数据和记录的状态结合起来，完成你正在做的任务。根据具体的任务，你可能需要多次经历这个过程，才能真正完成你的工作。

这些想法基本上是相同的，不管你使用的是哪种异步系统。不过，Twisted 是一种比 asynchat 更优秀的系统，所以我不打算解释 asynchat 的细节。相反，这里有一个使用 Twisted 的例子，展示了你所询问的那种操作：

from twisted.internet.defer import Deferred
from twisted.internet.protocol import Protocol, Factory
from twisted.internet.endpoints import TCP4ClientEndpoint
from twisted.internet import reactor

# Stream-oriented connections like TCP are handled by an instance 
# of a Protocol subclass
class SomeKindOfClient(Protocol):

     # When a new connection is established, the first thing that
     # happens is this method is called.
     def connectionMade(self):
         # self.transport is set by the superclass, and lets us 
         # send data over the connection
         self.transport.write("GREETING")

         # a Deferred is a generic, composable API for specifying
         # callbacks
         self.greetingComplete = Deferred()

         # Here's some local state
         self._buffer = ""

     # Whenever bytes arrive on the TCP connection, they're passed
     # to this method
     def dataReceived(self, bytes):

         # Incorportate the network event data into our local state.
         #  This kind of buffering is always necessary with TCP, because
         # there's no guarantees about how many bytes will be delivered
         # at once (except that it will be at least 1), regardless of
         # the size of the send() the peer did.
         self._buffer += bytes

         # Figure out if we're done - let's say the server response is 32
         # bytes of something
         if len(self._buffer) >= 32:
             # Deliver it to whomever is waiting, by way of the Deferred
             # object
             greeting, self._buffer = self._buffer[:32], self._buffer[32:]
             complete = self.greetingComplete
             self.greetingComplete = None
             complete.callback(greeting)

         # Otherwise we'll keep waiting until dataReceived is called again
         # and we have enough bytes.


# One of the normal ways to create a new client connection
f = Factory()
f.protocol = SomeKindOfClient
e = TCP4ClientEndpoint(reactor, "somehost", 1234)

# Connect returns one of those Deferreds - letting us specify a function
# to call when the connection is established.  The implementation of
# connect is also doing basically the same kind of thing as you're asking
# about.
d = e.connect(f)

# Execution continues to this point before the connection has been
# established.  Define a function to use as a callback when the connection
# does get established.
def connected(proto):
    # proto is an instance of SomeKindOfClient.  It has the 
    # greetingComplete attribute, which we'll attach a callback to so we
    # can "wait" for the greeting to be complete.
    d = proto.greetingComplete

    def gotGreeting(greeting):
        # Note that this is really the core of the answer.  This function
        # is called *only* once the protocol has decided it has received
        # some necessary data from the server.  If you were waiting for a
        # session identifier of some sort, this is where you might get it
        # and be able to proceed with the remainder of your application
        # logic.
        print "Greeting arrived", repr(greeting)

    # addCallback is how you hook a callback up to a Deferred - now
    # gotGreeting will be called when d "fires" - ie, when its callback
    # method is invoked by the dataReceived implementation above.
    d.addCallback(gotGreeting)

# And do the same kind of thing to the Deferred we got from 
# TCP4ClientEndpoint.connect
d.addCallback(connected)

# Start the main loop so network events can be processed
reactor.run()

要查看这个是如何工作的，你可以启动一个简单的服务器（比如 nc -l 1234），然后让客户端连接到它。你会看到问候信息到达，并且你可以发送一些字节回去。一旦你发送了30个字节，客户端会打印出来（然后就会一直停在那里，因为我在这个协议中没有实现进一步的逻辑）。