I'm sending this through LSTSRV-L@CLVM since the DEARN server is down and I
want the US to get this this weekend.
Regarding the STANFORD mail problem, this has been fixed in my test version
as I said (you probably didn't get the mail until a long time had elapsed due
to the links being down). The links have been down a long time here in France,
and I found with delight that some of my PUT PEERS NAMES commands had been
held at FRHEC11. I had thought I had individually updated those servers which
needed individual update before sending the global PUT, but obviously this was
not the case.
Considering the number of problems that have been solved in 1.5g, I am going
to distribute my 1.5g version as is, without the corresponding docs. I would
have preferred to wait until the docs were ready but it seems that we need to
synch out our versions ASAP.
The DISTRIBUTE problem was due to the following facts:
1) There was an entry in the old PEERS NAMES file saying that BITNIC's server
was NEWLIST not LISTSERV. If there had been no entry, this would not have
been a problem, but there was an entry with a different userid and the
servers thought that LISTSERV was Ricky's version.
2) Ibid. some sites have NOTIFY OFF in effect for [log in to unmask] *** PLEASE
PLEASE PLEASE *** remove this thing ASAP, a lot of error mail is being
suppressed (ie sent to /dev/null) because of that.
3) I had stupidly decided to protect 'old' servers (< 1.5d) of a security
problem with DISTRIBUTE. A side effect of the corresponding code is that
no entry in PEERS NAMES --> DISTRIBUTE disabled. A corollary is that no
:version tag --> same effect.
Another side effect is that server not found in PEERS NAMES --> invalid
file origin, with the same corollary about the missing :version tag.
To the best of my knowledge there are still 3 sites with an old version of
LISTSERV. I am going to remove all this BS from the 1.5g code and if the
aforesaid sites don't update their server, it's too bad, they may allow
Joe users to add people to their lists. I don't think it's normal that 95%
of people get in so much trouble because of 5% of people who don't update
their code. Besides I can't keep maintaining a server which replies
"Unknown command" when I tell it "RELEASE" to know its release number :-)
Sorry if I sound so negative but it's really becoming a pain.
4) The CUNYVM entry was full of errors. It's partly the fault of the person
who prepared it, and partly mine for not checking the entry before placing
it online. There was no :version tag, there were a few typos, and the
hardware machine was said to be a 3090-4 -- as I said earlier, this one did
not cause any problem but confirms the fact that the entry was prepared in a
hurry and was messed up.
In 1.5g the processing will be slightly altered:
- No entry in PEERS NAMES file: the server is assumed to be a 1.5f server
instead of 1.4-. DISTRIBUTE jobs are never routed to this server of course
(since it's not listed in PEERS NAMES), unless it's a PUT command. This may
cause security problems for people running a release < 1.5b but as I said,
I can't jeopardize the DISTRIBUTE command just because of 2-3 servers.
However, DISTRIBUTE jobs from such a server are accepted PROVIDED that the
userid of the server is LISTSERV. If it's not LISTSERV, jobs will be
rejected until the entry appears in PEERS NAMES, for obvious security
reasons.
- Missing :version tag: same as above, except that since the server does
appear in PEERS NAMES it will receive DISTRIBUTE jobs as if it had a
":version.1.5f" tag. That's the same fix as "no entry" above in fact.
- :version tag indicating a release < 1.5: this means that the DISTRIBUTE
feature is disabled at this node. Other servers won't send any job to such a
server, and if one is received it will be rejected. However, jobs from this
server will still be accepted (in case the local staff re-enable DISTRIBUTE)
- I will make sure to check all entries before including them in the official
PEERS NAMES file. I will also make sure to change my password to something
else. I don't want to suppress the password display from the console log
because (1) the command is displayed by higher-level routines BEFORE it is
scanned, and (2) the password can be useful to diagnose problems. Now
everybody knows that I like Rigoletto :-(
---> About this problem: I would like to ask ALL postmasters to send only
those tags they have updated when they send me an updated PEERS NAMES
entry for their sites. If you want to delete a tag, just send it back
with no value, eg ":remark1.". This will make it much easier for me to
check the thing.
The PEERS NAMES should now be synched, except the SUVM one (they installed
1.5f and the NAMES file from the shipment overrode theirs). I had sent
individual updates to those servers which needed it, except one which got held
at FRHEC11, and then a global update "from the other side" (sent to
LISTSERV@TAMVM1). This seems to have updated all the servers without problem,
but I'd suggest waiting until 1.5g is installed to use DISTRIBUTE again. With
the changes in the code it will probably become MUCH more stable.
Eric
|