programs hangs during socket interaction

Question

I have two programs, sendfile.py and recvfile.py that are supposed to interact to send a file across the network. They communicate over TCP sockets. The communication is supposed to go something like this:

sender =====filename=====> receiver

sender <===== 'ok' ======= receiver
               or
sender <===== 'no' ======= receiver

if ok:
sender ====== file ======> receiver 

I've got

The sender and receiver code is here:

Sender:

import sys
from jmm_sockets import *

if len(sys.argv) != 4:
    print "Usage:", sys.argv[0], "<host> <port> <filename>"
    sys.exit(1)

s = getClientSocket(sys.argv[1], int(sys.argv[2]))

try:
    f = open(sys.argv[3])
except IOError, msg:
    print "couldn't open file"
    sys.exit(1)

# send filename
s.send(sys.argv[3])

# receive 'ok'
buffer = None
response = str()
while 1:
    buffer = s.recv(1)
    if buffer == '':
        break
    else:
        response = response + buffer
if response == 'ok':
    print 'receiver acknowledged receipt of filename'
    # send file
    s.send(f.read())
elif response == 'no':
    print "receiver doesn't want the file"

# cleanup
f.close()
s.close()

Receiver:

from jmm_sockets import *

s = getServerSocket(None, 16001)
conn, addr = s.accept()


buffer = None
filename = str()

# receive filename
while 1:
    buffer = conn.recv(1)
    if buffer == '':
        break
    else:
        filename = filename + buffer
print "sender wants to send", filename, "is that ok?"
user_choice = raw_input("ok/no: ")

if user_choice == 'ok':
    # send ok
    conn.send('ok')
    #receive file
    data = str()
    while 1:
        buffer = conn.recv(1)
        if buffer=='':
            break
        else:
            data = data + buffer
            print data
else:
    conn.send('no')
conn.close()

I'm sure I'm missing something here in the sorts of a deadlock, but don't know what it is.

Answer 1

With blocking sockets, which are the default and I assume are what you're using (can't be sure since you're using a mysterious module jmm_sockets), the recv method is blocking -- it will not return an empty string when it has "nothing more to return for the moment", as you seem to assume.

You could work around this, for example, by sending an explicit terminator character (that must never occur within a filename), e.g. '\xff', after the actual string you want to send, and waiting for it at the other end as the indication that all the string has now been received.

Answer 2

TCP is a streaming protocol. It has no concept of message boundaries. For a blocking socket, recv(n) will return a zero-length string only when the sender has closed the socket or explicitly called shutdown(SHUT_WR). Otherwise it can return a string from one to n bytes in length, and will block until it has at least one byte to return.

It is up to you to design a protocol to determine when you have a complete message. A few ways are:

1. Use a fixed-length message.
2. Send a fixed-length message indicating the total message length, followed by the variable portion of the message.
3. Send the message, followed by a unique termination message that will never occur in the message.

Another issue you may face is that send() is not guaranteed to send all the data. The return value indicates how many bytes were actually sent, and it is the sender's responsibility to keep calling send with the remaining message bytes until they are all sent. You may rather use the sendall() method.