Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!elroy.jpl.nasa.gov!decwrl!nsc!voder!blia!ted
From: ted@blia.sharebase.com (Ted Marshall)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Problems with TCP, read(), and write()
Message-ID: <13790@blia.sharebase.com>
Date: 4 Apr 91 00:26:58 GMT
References: <1991Mar30.174657.28200@wpi.WPI.EDU>
Organization: ShareBase Corp, Los Gatos, CA
Lines: 26


The problem is that TCP is a pure stream protocol that does not preserve
record boundries. Thus, if something delays the processing on some of the
bytes from one of your write()s, your read can easily complete with only the
beginning of what we written.

Since you know that all "records" are 32 bytes, there is a real easy solution:
if your read() completes with less than 32 bytes, add that count to your
buffer pointer, subtract the count from your buffer size, and repeat the read.
Continue this loop until your buffer is full. For example:

	ptr = &buffer[0];
	cnt = sizeof buffer;
	do
	{
		i = read(d, ptr, cnt);
		if (i <= 0)
			<error-processing>
		ptr += i;
		cnt -= i;
	} while (cnt > 0);

-- 
Ted Marshall                                       ted@airplane.sharebase.com
ShareBase Corp., 14600 Winchester Blvd, Los Gatos, Ca 95030     (408)378-7000
The opinions expressed above are those of the poster and not his employer.