|
Boost : |
From: George A. Heintzelman (georgeh_at_[hidden])
Date: 2001-07-26 15:10:51
I also have no time to help with this, but let me talk about something
connected to any sockets library, and that is non-blocking IO.
I want an interface where I don't have to make local buffers, and can,
once I've determined that the data is there to be read, treat the
socket transparently like a stream.
As an example of what I've been vaguely thinking about being able to do:
#include <boost/socket.hpp>
int main() {
boost::socket<char> server("some_server", 1234);
// Elide catching of failures to connect
int size = 0;
while (1) {
if (size) {
if (server.chars_avail(size)) {
// Read in the message we expect to have & process it.
}
} else if (server.chars_avail(2)) {
char hi, lo;
server >> hi >> lo;
size = hi >> 8 + lo;
}
// Process other tasks, or sleep for some length of time.
}
}
Some things which I think would be very useful in the kinds of tasks
I've run into in this domain:
1) Being able to find the first occurance of a particular character
in the incoming stream, without consuming all the characters preceding
it, and without blocking in the case where it is not there to be read
yet. This is useful for specific-character-delimited protocols.
2) Being able to 'mark' and rewind a stream to that mark on failure.
This is needed for protocols where there is no specific delimiter but
you know when the message is over as you get through it. This can also
be useful for character-delimited protocols where you expect incomplete
messages to be rare but occasionally pop up.
3) Being able to find out if at least N characters are available to
be read. This is important for message-length-in-a-header protocols.
4) Otherwise being able to treat the socket as a C++ stream, so that
operator>> and <<'s can be used internally, with only the highest-level
piece having to be able to figure out whether a message is complete or
not.
I think some derivations from basic_streambuf and basic_[io]{1-2}stream
would make these things possible.
This is really disjoint from the work on sockets per se, but since it
is a related task in most uses of sockets, I thought I would mention it
here.
George Heintzelman
georgeh_at_[hidden]
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk