Serializing a C struct in Python and sending over a socket

Question

I'm trying to serializing the following C struct

struct packet
{
    int id;
    unsigned char *ce;
    unsigned char *syms;
};

in Python and send it over a socket. The number of elements pointed by ce and syms are known to be N. Currently I'm doing this way. First, I wrap the struct using ctypes to

class Packet(Structure):
    _fields_ = [("id",  c_int),
                ("ce", POINTER(c_ubyte)),
                ("syms", POINTER(c_ubyte))]

Then, I populate an object of the following class using its fill_data function from a ctypes.POINTER(Packet):

class DataLoad:
    def __init__(self):
        self.id = -1
        self.ce = []
        self.syms = []

    def fill_data(self, pkt_p):
        """ pkt_p is POINTER(Packet)
        """
        self.id = pkt_p.contents.id
        self.ce = []
        for i in range(N):
            self.ce.append(pkt_p.contents.ce[i])
        self.syms = []
        for i in range(N):
            self.syms.append(pkt_p.contents.syms[i])

Finally, I simply use pickle.dumps(DataLoad) to generate a byte stream and send.

This approach works well. However, it seems to be quite slow. One reason I can see is that pickle.dumps bring much overhead. For example, if the C struct is only 1024 bytes, I may have to send almost 4000 bytes for each struct using pickle. Also, packing/populating DataLoad also takes time.

So my question is, do I have other better choices to serialize this C struct in python and send? I would prefer to avoid pickle and populate a separate class instance. Thanks.

Answer 1

Finally I figured out the following way to manually serialize the `Packet' instance without using pickle.

def serialize(pkt_p, size_g, size_p):
    """ Serialize Packet instance
        size_g - number of elements pointed by ce
        size_p - number of elements pointed by syms
        Return a byte stream
    """ 
    pktstr = b''
    pktstr += struct.pack('i', pkt_p.contents.id)
    pktstr += string_at(pkt_p.contents.ce, size_g)
    pktstr += string_at(pkt_p.contents.syms, size_p)
    return pktstr

def deserialize(pkt_p, pktstr, size_g, size_p):
    """ De-serialize pktstr and fill a POINTER(Packet)
    """
    pkt_p.contents.id = struct.unpack('i', pktstr[0:4])[0]
    ce = (c_ubyte * size_g).from_buffer_copy(pktstr[4:4+size_g])
    pkt_p.contents.ce = cast(ce, POINTER(c_ubyte))
    syms = (c_ubyte * size_p).from_buffer_copy(pktstr[-size_p:])
    pkt_p.contents.syms = cast(syms, POINTER(c_ubyte))

The string_at() and the from_buffer_copy() functions are the key.

Answer 2

First off, if you know the number of elements to be N, I would suggest changing your structure to this:

class Packet(Structure):
    _fields_ = [("id", c_int),
                ("ce", c_ubyte * N),
                ("syms", c_ubyte * N)]

Next, if you only want to send the struct data, you don't need to pickle the whole thing. Just send the packet data:

p = Packet()
p.id = 555
...
# cast the struct to a pointer to a char array
pdata = ctypes.cast(ctypes.byref(p), ctypes.POINTER(ctypes.c_char * ctypes.sizeof(p)))
# now you can just save/send the struct data
someSocketObject.send(pdata.contents.raw)

To read the data on the other side:

p = Packet()
raw = someSocketObject.read(ctypes.sizeof(p))
ctypes.memmove(ctypes.pointer(p),raw,ctypes.sizeof(p))