🚀 go-pugleaf

In article <1993Jun16.103920.9820@ludd.luth.se> peppar@ludd.luth.se (Peter Peppar Parnes) writes:
   >>Sorry, I found another problem last night. It looks hard to fix and I
   >>am thinking how to treat it.
   >
   >What's the problem?

The RPCWriter class negotiates transfer mode and endian with
its peer at the opening of a connection. The negotiation protocol is
quite simple, that is, sending characters representing its modes
to its peer. Since there are two modes to be negotiated
in this negotiation, a RPCWriter sends two characters to the peer.
The coding of iostreamb::negotiate() looks like:

	stream << localMode; stream.flush();
	stream >> remoteMode;	
	// check binary transfer mode

	stream << localMode; stream.flush();  // (A)
	stream >> remoteMode;			    // (B)
	// check endian

After sending a character, flushing of the output stream is required
in order to flush out the character from the stream. 
        // see comment of iostreamb::negotiate() for detail.
The problem is that the flushing at (A) may discard a pending
character in the input stream.  The discarded character is, of course,
required for the next input at (B).  If this situation occurs, the (B)
will never return.

An idea was getting rid of discarding of pending characters from
flush(). // see rpcbuf::sync().
I tried this, but communication became unstable.
Another idea is use of write() at (A) instead of stream. But this is
not elegant.

I found another problem today. Sending float or double loses a byte.
Sigh...

@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
HIRANO Satoshi                  Electrotechnical Laboratory, Japan
				hirano@etl.go.jp
The future is Fluid.