OpenBSD support thread

belenus · Post by **belenus** » Fri Jan 19, 2007 5:30 pm

[can be deleted]

belenus · Post by **belenus** » Sat Jan 20, 2007 5:09 pm

Made a port for 0.3.0 and 0.2.8.2.1 but when I play on that server I get weird graphical explosion outputs. Tried the server on two different machines.

Recording attached.

Packages for OpenBSD 4.0 available!

Website · Post by **Z-Man** » Mon Jan 22, 2007 12:30 am

Sure it's related to the server? The first strange explosions appear when you fumble in the menu, it may have to to more with that (from my point of view). Shot in the dark fix suggestion: increase the GAME_TIMEOUT setting. I haven't played the recording back in a debugger yet, but those things can happen when cycle deletion messages for last round's cycles arrive at the client when the new round is already initialized.
And of course, we already had our fair share of bugs with similar effects, but so far, they affected all platforms equally.

belenus · Post by **belenus** » Mon Jan 22, 2007 7:51 am

z-man wrote:Sure it's related to the server? The first strange explosions appear when you fumble in the menu, it may have to to more with that (from my point of view). Shot in the dark fix suggestion: increase the GAME_TIMEOUT setting. I haven't played the recording back in a debugger yet, but those things can happen when cycle deletion messages for last round's cycles arrive at the client when the new round is already initialized.
And of course, we already had our fair share of bugs with similar effects, but so far, they affected all platforms equally.

I've run it on 3 boxes now, two are notebooks in the local LAN which shows the weird explosions and the third is a dedicated server which runs just fine without errors.

Website · Post by **Z-Man** » Mon Jan 22, 2007 11:41 am

Hmm, I think I'll need a serverside recording. I debugged the clientside one, and indeed, the explosions are triggered when the cycles are destroyed on server command. That command usually should come during the grid deconstruction phase and the explosions should never be displayed, but in your case, they come in during the following grid construction phase. If increasing GAME_TIMEOUT does not help, it's a bug in the server code, the server should wait until all clients have received all important messages before progressing from the deconstruction to the construction phase.

The serverside recording, ideally, should not contain AIs. They like to break the recording from floating point inaccuracies. Just use additional clients in chatbot mode

And I'd also need the clientside recording to match the serverside recording, I'll probably need to bounce between the two a lot.

I'll probably have to somehow eliminate the explosion that gets triggered in this situation anyway; it's what causes the explosion sound at round end in 0.3.0.

belenus · Post by **belenus** » Mon Jan 22, 2007 3:28 pm

Recordings with default GAME_TIMEOUT... in a LAN.

Windows Client: 0.2.8.2.1
Server: 0.2.8.2.1 OpenBSD package

Website · Post by **Z-Man** » Mon Jan 22, 2007 4:37 pm

Heh, this seemed familiar for a good reason

It's already fixed largely in the branch. I added some more safety checks. Time to get 0.2.8.3 finished.

What happened is:
- All walls are cleared
- All cycles are killed
- The cycles that are still alive drop their wall
- the wall keeps referencing the cycle and keeping it alive for a bit longer, just enough that the destruction message is sent out late
- the client creates the explosion on arrival of the destruction message, which is too late.

The fix is to simply swap the cycle kill/wall clear order. Here's a patch against 0.2.8.2.1 for you, I hope it works.

belenus · Post by **belenus** » Mon Jan 22, 2007 5:20 pm

The fix is to simply swap the cycle kill/wall clear order. Here's a patch against 0.2.8.2.1 for you, I hope it works.

Patch applied without issues but had to change it a little. (Just need diff -u output from inside the armagetronad folder.)

Made a new package but can't test it right now.

Ahh btw, if you search the LAN with a client... how does that work?
Just had a stupid Windows firewall at work signal a portscan because of that...

Website · Post by **Z-Man** » Mon Jan 22, 2007 5:34 pm

belenus wrote:Ahh btw, if you search the LAN with a client... how does that work?

With a portscan

Query messages are broadcast via UDP to ports 4534 to 4540 to find multiple servers running on different ports on the same machine. If you see a way to avoid this that works in all situations, I'm open to suggestions. If you just want to avoid your firewall, you can limit the scanned range in the network setup.

belenus · Post by **belenus** » Mon Jan 22, 2007 5:41 pm

z-man wrote:If you see a way to avoid this that works in all situations, I'm open to suggestions.

The server could just broadcast itself (can be turned ON and OFF, default should be OFF) and the client listens for those broadcasts. If the client catches a broadcast it connects to the server and fetches detailed information.

Website · Post by **Z-Man** » Mon Jan 22, 2007 5:57 pm

That reverses the problem

Now you can't have several clients on the same machine, because the clients have to open a well defined listening port. I admit that is a more far-fetched scenario in most cases, but for debugging, I sometimes run as many clients as my machines can handle.
Plus, the overall network load is increased by the constant broadcast (a very minor problem that would be tolerable).

belenus · Post by **belenus** » Mon Jan 22, 2007 7:15 pm

z-man wrote:That reverses the problem Now you can't have several clients on the same machine, because the clients have to open a well defined listening port. I admit that is a more far-fetched scenario in most cases, but for debugging, I sometimes run as many clients as my machines can handle.
Plus, the overall network load is increased by the constant broadcast (a very minor problem that would be tolerable).

Sure it would reverse the problem... but it is much better having ONE server broadcast that it is there than having multiple clients portscan a whole local subnet.

belenus · Post by **belenus** » Mon Jan 22, 2007 7:23 pm

The patch worked, looks clean now. Could I get one for 0.3.0 too or is it the same? =)

Website · Post by **Z-Man** » Mon Jan 22, 2007 7:38 pm

Yes, it works on 0.3.0 as well.

belenus · Post by **belenus** » Mon Jan 22, 2007 8:23 pm

z-man wrote:Yes, it works on 0.3.0 as well.

Patches applied, port and packages updated.

Seems to work now but you got an idea why I got huge lag when playing in a local network but connection to a real server is ok?