Resent-Date: Tue, 5 Dec 1995 16:42:53 -0500 (EST)
Subject: Re: LETS DO TS
To: ircd2-patches@kei.com
Date: Tue, 5 Dec 1995 22:42:20 +0100 (MET)
Resent-Message-Id: <"0A3wk.0.n2.SpBnm"@kei.com>
Resent-From: ircd2-patches@kei.com
Precedence: list
Resent-Sender: ircd2-patches-request@kei.com

Master of Puppets said:

> Can we make a TS (patch) that is compatible with non-TS servers? I'd
> like to see something on the table within the next few weeks at least.

I'd be willing to code such a patch within the next few weeks. It will 
do TS on nicks (for collisions) and on channel creation time (for 
chanops after a split), and nothing more. I'll make it a patch on the 
current standard ircd (2.8.21), and see if much work is needed to make 
it work with +CS and other common ircd performance patches.. once I have 
something working I'll just run it on a test server and it can be linked 
to other test servers, some running the patches and some not, with bots 
introducing fake users, and so on, to see how it stands.

Craig A. Huegen said:

> But it *IS* broken.  We don't want to see what happened with uoknor 
> again. You need to make sure that it will run well and not cause 
> unexpected problems with non-TS stuff.  It has been proven in the past 
> to cause such problems.
> Actually, if I remember right, it was just the undernet's TS code 
> slapped into a normal server.  There was nothing in the patch to only 
> send timestamps to TS-capable servers, or just basically somehow phase 
> TS in.

That's precisely why we need a new patch written to work with EFnet
servers, and to change as little as possible in them.

Brian Tackett said:

> With the current way that TS works, I don't see any easy way off the top 
> of my head to make TS work with non-TS servers, unless you use automatic 
> protocol checking to determine whether the server sending a mode is a TS 
> server or not. But that's a whole other story. 

But not an overly complicated one. Have the server send a special 
command, like "TSVER" with some parameters to be determined, between 
PASS and SERVER. Non-TS servers just ignore this (and they don't even 
send anythign to the IrcOPS because the server hasn't registered as one 
yet, at that point). TS servers know which of the servers that are 
linked to them have the TS patch and which don't, and act accordingly.

The whole compatibility code can be #ifdef'ed and not compiled in later 
if the move is successful (or just when all of a server's uplinks have 
TS too).

Jim Benson said:

> Because TS *is* an either/or situation - you go all TS, or you do not go 
> TS at all. Things in life that are overly comlicated are usually wrong. 
> This is actually rather simple - fix EFnet or not. Fix it woth TS or 
> not.

But it can be introduced gradually. Asking everyone to change to another 
entirely incompatible version at the same time won't work, just think of 
EU or of .jp. Putting out a version that can talk to TS and non-TS 
servers, and having people who think it's a good idea switch to it, and 
then if the decision is made, having the rest switch to it too, that can 
work.

And the best incentive to switch to TS, once major hubs run it, will be 
that the non-TS servers will be getting all the clonebotters :-)

Brian Tackett said:

> As I see it here, the issue is not whether TS works or does not work. 
> The issue is whether TS can be phased in with non-TS server, in such a 
> way that the net's stability isn't destroyed. TS has been deployed on 
> Undernet IRC server for a couple of years now, and has shown no signs of 
> breaking as the number of clients tops 5k or so. Will it work with 
> 15,000? I don't know, but you are very correct in pointing out that the 
> answer lies in a test net. Get a dozen or so servers, put TS on half, 
> non-TS on the other half, write some protocol detection code, and then 
> load each server with a few thousand dummy clients. See what happens, 
> observe the results, and revise the code accordingly.

That needs to be done before any TS-patched server runs on EFnet,
obviously. *Then* we can follow Jim Benson's idea.

So........

Here's my proposal for the TS patch:

1. servers that run TS patches announce it when connecting to another
   server or when one connects to them.

2. they keep a timestamp for each nick and one for each channel, 
   validating the date of creation of the nick and the date of creation of 
   the channel

3. they use this information to bounce modes and kill only one nick
   out of the 2 conflicting ones, only when the offending mode change
   or nick introduction is coming from another TS-aware server.

This in particular means that this patch would not attempt to avoid 
desynchs caused by two valid ops deopping (or kicking) each other 
through the lag - that kind of test is much more complicated to do, and 
if we avoid all the desynchs caused by people trying to hack, it's 
already a huge improvement. The case in which two valid ops deop each 
other for some reason can be left as it is now; they still have ops each 
on half the net, and they can fix it by reopping each other.

The main problem is dealing with clocks that are not set to the same 
exact time, and deciding whether the TS's that get sent are universal 
(always kept the same) or are adjusted by each server according to the 
time differences between directly connected servers.

Either way, this should be possible to get to work without any kind of 
central clock, mainly because the TS comparisons will be between rather 
different values almost all the time (a channel created hours ago vs. 
one created a minute ago, or the same with nicks), so a few seconds
of desynchronization wouldn't really hurt the protocol.

I'll be looking at this more closely, and at the Undernet code, and post 
again when I have something more definite... I'd like to see some talk 
on here about the actual protocol to be used between TS servers, so be
sure to complain if you see something wrong with my idea :-)

-- 
e-mail: roger.espel.llima@ens.fr                     IRC: orabidoo

"You don't possess me, don't impress me, just upset my mind
 Can't instruct me or conduct me, just use up my time" -King Crimson