|hostname|hostname:file||The name of the remote host, and the name of the queue file of articles destined for that host. The hostname may be an internet address in dotted format (e.g. 10.2.0.78, [10.0.0.78]). If the hostname is given without an associated file, it is assumed that the hostname is also the name of the queue file. If the separator is "::" instead of ":", it is assumed that the remote host speaks DECNET, instead of the default, IP/TCP.|
Toggles reporting of transfer statistics (how many articles we
offered them, how many they accepted, etc).
Default is ON.
Toggles DEBUG output on stderr.
This can be used to see exactly what the two systems are saying to
each other, except for the actual article text.
Default is OFF.
Toggles requeuing of failed articles.
A failed article is an article that we (client) offer them (remote server),
they accept, we transmit, and then they report that they "failed"
or dropped the article (i.e. inews(1) on the remote returned non-zero).
If we have requeuing set, we save the list of articles that they
failed on, and rewrite the queue file with them, so that they get
reoffered the next time we initiate transmission to them.
Default is ON.
|-a||This flag says that the next queue file on the command line isnt a queue file, but is a single netnews article to be transmitted to the remote in a single operation.|
|NOTE: this option causes nntpxmit to exit immediately after this transfer is done (regardless of whatever else is on the command line), and to exit with a code indicating whether the articles was successfully accepted by the remote server (zero exit for success, non-zero for failure).|
|The next options set the underlying transport protocol that nntpxmit uses. The NNTP specification assumes a TCP-style transport protocol underlies it (i.e. a reliable, flow-controlled, full-duplex byte stream). Nntpxmit assumes that after doing some magic to get a descriptor, it can do read(2) and write(2) calls (and use stdio) to move data and check for errors. By default, nntpxmit will use IP/TCP (DoD Internet Protocol suite).|
|-T||Sets transport protocol to IP/TCP for all remaining transfers (unless reset by other transport flags). Default transport.|
|-D||Sets transport protocol to DECNET for all remaining transfers (unless reset by other transport flags). NOTE: using "::" as the hostname/queue filename separator has the same effect.|
|-F||This says that the hostname is a file descriptor number, already open to a remote server (with some reliable protocol underneath) that was passed to nntpxmit through a fork(2).|
Nntpxmit implements an interactive ihave/sendme transmission system. Roughly, the protocol is
If the communcation link should fail (and nntpxmit detects it through a system call error return), nntpxmit will rewrite the queue file with the article filenames of the articles that it did not transmit (that is, we dont retransmit stuff weve already successfully sent and gotten back an positive confirmation that they got it).
1. open the article, fetch out the message-id (required on all netnews articles), and send the command IHAVE <message-id> to the remote. 2. The remote will then say either "Ive seen it already" or "please send that article to me." 3. If the response was negative, nntpxmit loops back to step 1 and offers the next article (until queue file EOF). Otherwise, nntpxmit will send the article, using SMTP [RFC821] text transmission conventions (i.e. CRLF line terminators, and dot escaping). 4. Nntpxmit waits for the remote to say whether the article was successfully accepted or not. If the answer is negative and requeuing of failed articles is enabled, nntpxmit will queue this articles filename to be written back to the queue file at the end of the session with this remote.
Erik E. Fair
RFC977 - Network News Transfer Protocol (NNTP),
RFC850 - USENET Article Format standard,
RFC821 - Simple Mail Transfer Protocol (SMTP),
Always requeuing failed articles can lead to beating the remote to death with a list of articles that he cant accept for come structural reason. How many of these have to pile up before you should declare that something is seriously wrong with the remote system and stop trying?
While nntpxmit will lock a queue file (your version of UNIX permitting) against multiple invocations of itself, there is no locking with inews(1), which is what writes the queue files in the first place. Therefore, never use nntpxmit on the queue files that inews(1) writes, because two processes writing into the same file without some kind of cooperation will almost certainly trash the file; move them to some other name that inews(1) knows nothing about, so that you wont lose articles to races between inews and nntpxmit.
Adding inews(1) compatible locking to the C code would be much more trouble than its worth, and violates the KISS principle besides.