The changes I made last night allow me to run many clients and check the syncing.

Well I think I have tracked down the sync problems between the client and the server. It's not good news. It looks like different clients are producing different floating point results from some maths calls and that the clients are not all using IETF floating point representation like I (wrongly) assumed.

I have several choices for the fix, but none of them small. I'll keep you posted when I make progress...

were you using different compilers?

i know another game wich suffer from diff floating points math wich drives players out of sync; they temporanely solved it by using the same compiler over the different OS

How would the different compilers issue apply to me tho.

I serve and play using the same server/client builds, compiled
using the exact method outlined in the Wiki setup, and it still
gets these 'un-dead / un-sync' errors.

It's even connected via loop-back, so it isn't losing track from
the internet delay.

Should there be some specific sync errors i should be looking for?

Edit: if the round ending without showing that you died counts, it's still happening.
Possibly it would help show the truth, if you added one more channel
to the server-log output; [combat]. Then we could see if the server
is recording it, and the clients not. It currently only shows [info], and
misses out on most of the details of what's going on.

Thrax I agree there are issues there too, but I am hoping that is a smaller issue and I need to focus on the larger one first.

oh, no stress.. there are definately a lot of server-controlled elements
on the players client, so even a few of them out of alignment could make
results vary.
i guess i'm just unclear on the nature of the sync errors you are addressing.

Basically it seems that different clients (due to floating point errors) can generate or load very slightly different versions of the same landscapes. This over time causes shots to miss on one client where they would have hit on the other client etc...

ah, generated landforms. i did see you'd added a few.
Differences between players of a fraction of a centimeter could indeed
make impacts seem odd. If the damages are declared by the server,
then all players still see the results the same.

An update to let you know things are moving on. I've spent the last little while converting the simulation code from using floating point numbers to using fixed point numbers. An initial check seems to show that this has fixed the issues I was seeing.

I am going to complete the process and then do some more testing, hopefuly have something availible in the next couple of days.

This is going to be another protocol update isn't it?

at least this time i'll be expecting it, it's only 2 letters in a mod..

As long as players keep going with your updates, I've already got quite a
few that prefer the 41 and don't like to go back.

Once this is cured, we have a several non-visual items lined up.

Yes, definitely. Initial testing looks good BTW.

An update:-

Still working on it. Every night for the last week infact. The problem is the testing takes so long. I am getting 1-5 sync issues in an eight hour test cycle so its quite hard to track down.

I am currently running a linux server with 3 clients (2 windows and 1 linux) continualy connecting playing and then disconnecting. So I think it is pretty representive of a real scenario (for multiplayer). Probably the most amount of scale testing that has ever been done so I have fixed one or two other issues along the way.

Just found another bug so I will run with that fixed and see what happens....

Sounds like a tough issue.

I've commited the changes so far, so people can try them out. I think there is still one bug in the sync code (at least) but it has only shown itself once.

Let me know how you get on.

Off to update.