Comments on Thinking Asynchronously in C++: Unbuffered socket iostreams

Now, it would be nice if we could do async reads f...

2008-12-25T05:13:00.000+11:00

Now, it would be nice if we could do async reads from socket iostreams. The example is nice, but it only covers one-off sends and receives. As far as I can tell, there is no way to get the iostream ease of use for reading from or writing to a socket unless you do it completely synchronously, since the underlying basic_socket_streambuf doesn't support anything but an asynchronous connect().

Such functionality would also simplify one form of async_read(), it seems a bit strange that the first argument is something like an ip::tcp::socket and the second argument is a basic_stream_buf, which is itself a basic_socket.

In fact, I did think about it and decided it shoul...

2008-02-29T23:32:00.000+11:00

In fact, I did think about it and decided it should be on by default. Consequently my revised TR2 proposal (N2175) does specify that the stream should be tied to itself.

However, it seems I forgot to actually implement it! :-/

Just a note to say that I had the same thought tha...

2008-02-28T14:23:00.000+11:00

Just a note to say that I had the same thought that anonymous did, and think it's an entirely reasonable default for the stream to flush its write buffer when a read is attempted. In fact, I was surprised that it didn't :)

Interesting idea, I didn't think of that. Using ti...

2007-02-09T16:27:00.000+11:00

Interesting idea, I didn't think of that. Using tie() is not exactly the same as an unbuffered stream, but it does give the expected behaviour for request-response style protocols (like HTTP, SMTP etc). Perhaps it should also be set by default, hmmm... something to think about.

Is it not possible to tie it to itself (or somethi...

2007-02-02T01:14:00.000+11:00

Is it not possible to tie it to itself (or something similar) so that it automagically flushes on a read operation?