Tcl Source Code

There is a bug in the tcludp routine udpInput().

When a Tcl_DriverInputProc() returns the value 0,
the core I/O parts of Tcl take that to mean a signal
of EOF on the channel.  udpInput() is returning 0
in several places when it does not intend that meaning.

    if (statePtr->doread == 0) {
        statePtr->doread = 1;  /* next time we want to behave normally */
        *errorCode = EAGAIN;   /* pretend that we would block */
        UDPTRACE("Pretend we would block\n");
        return 0;
    }

That should be "return -1" instead.

    if (bufSize == 0) {
        return 0;
    }

That's arguably also wrong, but harmless since the Tcl
I/O core will never pass in a value of bufSize==0.

    if (packets == NULL) {
        UDPTRACE("packets is NULL\n");
        return 0;
    }

That should probably be "return -1".  If this is an
attempt to signal that the channel is blocked, then
there should also be

        *errorCode = EAGAIN;

and finally....

    if (bytesRead > -1) {
        return bytesRead;
    }

That might also return 0, whenever recvfrom()
returns 0.  If those circumstances correspond to
EOF, then all is fine.  Otherwise the above should
be preceded by something like

    if (bytesRead == 0 && ThisIsNotEof()) {
        *errorCode = EAGAIN;
        return -1;
    }

With those fixes, I think tclUdp will work again
with current dev sources of Tcl and will also
continue to work with all the deployed releases
of Tcl.

Is there an active upstream to send these changes to?