Creating a custom ODBC driver

Another option: Instead of creating a ODBC driver, implement a back end that talks the wire protocol that another database (Postgresql or MySQL for instance) uses.

Your users can then download and use for instance the Postgresql ODBC driver.

Exactly what back-end database you choose to emulate should probably depend the most on how well the wire protocol format is documented.

Both Postgres and MySQL has decent documentation for their client-server protocols.

A simple Python 2.7 example of a server backend that understands parts of the Postgresql wire protocol is below. The example script creates a server that listens to port 9876. I can use the command psql -h localhost -p 9876 to connect to the server. Any query executed will return a result set with columns abc and def and two rows, all values NULL.

Reading the Postgresql docs and using something like wireshark to inspect real protocol traffic would make it pretty simple to implement a Postgresql-compatible back end.

import SocketServer
import struct

def char_to_hex(char):
    retval = hex(ord(char))
    if len(retval) == 4:
        return retval[-2:]
    else:
        assert len(retval) == 3
        return "0" + retval[-1]

def str_to_hex(inputstr):
    return " ".join(char_to_hex(char) for char in inputstr)

class Handler(SocketServer.BaseRequestHandler):
    def handle(self):
        print "handle()"
        self.read_SSLRequest()
        self.send_to_socket("N")

        self.read_StartupMessage()
        self.send_AuthenticationClearText()
        self.read_PasswordMessage()
        self.send_AuthenticationOK()
        self.send_ReadyForQuery()
        self.read_Query()
        self.send_queryresult()

    def send_queryresult(self):
        fieldnames = ['abc', 'def']
        HEADERFORMAT = "!cih"
        fields = ''.join(self.fieldname_msg(name) for name in fieldnames)
        rdheader = struct.pack(HEADERFORMAT, 'T', struct.calcsize(HEADERFORMAT) - 1 + len(fields), len(fieldnames))
        self.send_to_socket(rdheader + fields)

        rows = [[1, 2], [3, 4]]
        DRHEADER = "!cih"
        for row in rows:
            dr_data = struct.pack("!ii", -1, -1)
            dr_header = struct.pack(DRHEADER, 'D', struct.calcsize(DRHEADER) - 1 + len(dr_data), 2)
            self.send_to_socket(dr_header + dr_data)

        self.send_CommandComplete()
        self.send_ReadyForQuery()

    def send_CommandComplete(self):
        HFMT = "!ci"
        msg = "SELECT 2\x00"
        self.send_to_socket(struct.pack(HFMT, "C", struct.calcsize(HFMT) - 1 + len(msg)) + msg)

    def fieldname_msg(self, name):
        tableid = 0
        columnid = 0
        datatypeid = 23
        datatypesize = 4
        typemodifier = -1
        format_code = 0 # 0=text 1=binary
        return name + "\x00" + struct.pack("!ihihih", tableid, columnid, datatypeid, datatypesize, typemodifier, format_code)

    def read_socket(self):
        print "Trying recv..."
        data = self.request.recv(1024)
        print "Received {} bytes: {}".format(len(data), repr(data))
        print "Hex: {}".format(str_to_hex(data))
        return data

    def send_to_socket(self, data):
        print "Sending {} bytes: {}".format(len(data), repr(data))
        print "Hex: {}".format(str_to_hex(data))
        return self.request.sendall(data)

    def read_Query(self):
        data = self.read_socket()
        msgident, msglen = struct.unpack("!ci", data[0:5])
        assert msgident == "Q"
        print data[5:]


    def send_ReadyForQuery(self):
        self.send_to_socket(struct.pack("!cic", 'Z', 5, 'I'))

    def read_PasswordMessage(self):
        data = self.read_socket()
        b, msglen = struct.unpack("!ci", data[0:5])
        assert b == "p"
        print "Password: {}".format(data[5:])


    def read_SSLRequest(self):
        data = self.read_socket()
        msglen, sslcode = struct.unpack("!ii", data)
        assert msglen == 8
        assert sslcode == 80877103

    def read_StartupMessage(self):
        data = self.read_socket()
        msglen, protoversion = struct.unpack("!ii", data[0:8])
        print "msglen: {}, protoversion: {}".format(msglen, protoversion)
        assert msglen == len(data)
        parameters_string = data[8:]
        print parameters_string.split('\x00')

    def send_AuthenticationOK(self):
        self.send_to_socket(struct.pack("!cii", 'R', 8, 0))

    def send_AuthenticationClearText(self):
        self.send_to_socket(struct.pack("!cii", 'R', 8, 3))

if __name__ == "__main__":
    server = SocketServer.TCPServer(("localhost", 9876), Handler)
    try:
        server.serve_forever()
    except:
        server.shutdown()

Example command line psql session:

[~]
$ psql -h localhost -p 9876
Password:
psql (9.1.6, server 0.0.0)
WARNING: psql version 9.1, server version 0.0.
         Some psql features might not work.
Type "help" for help.

codeape=> Select;
 abc | def
-----+-----
     |
     |
(2 rows)

codeape=>

An ODBC driver that speaks the Postgresql protocol should work as well (but I have not tried it yet).


ODBC drivers are very complex - the decision to write one should not be taken lightly. Reviewing existing open source drivers are a good approach for examples but most have shortcommings you may not want to emulate :) APIs are the same regardless of OS platform. FreeTDS for MSSQL/Sybase has one of the better open source ODBC Driver implementations I've seen.

If you control the application you can get away with implementing what may be just a very small subset of the spec in a reasonable amount of time. To use in a general purpose environment can require quite a bit more effort to get right. Off the top of my head in addition to simply implementing dozens of wrapper calls you will also have to implement:

  • Metadata access functions
  • ODBC specific query syntax parsing
  • SQLSTATE Error message mappings
  • Multibyte/Character set marshalling
  • ODBC version 2,3 support - error messages/function mappings
  • Cursors
  • DM configuration UI for managing the datasource