CVS test server

What do you want to see in Armagetron soon? Any new feature ideas? Let's ponder these ground breaking ideas...
Post Reply
meriton
Round Winner
Posts: 256
Joined: Sun Nov 20, 2005 3:33 am

Post by meriton »

Yep, that's the one. (A server restart would have given a different message and wouldn't have occured during a round.)

The error happens at the end of the recording. I am sending it now.

I used a custom build (gcc, windows) with the glancing prototype and mercam, do you need the source as well?
User avatar
Z-Man
God & Project Admin
Posts: 11585
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Keep the sources around, I may need them later. But probably not: your changes don't look like they can break the recording. What version are your sources derived from?

And I'm afraid you have to resend the recording :( Postfix comes with a preconfigured 10MB message size limit, and fetchmail throws away messages exceeding that instead of keeping it on the server like it does for its own size limit. Grrr and sorry, I was unaware of this quirk. I bumped the limits up to 200MB per message.
meriton
Round Winner
Posts: 256
Joined: Sun Nov 20, 2005 3:33 am

Post by meriton »

z-man wrote:Keep the sources around, I may need them later. But probably not: your changes don't look like they can break the recording. What version are your sources derived from?
I synched with CVS on 16.12.2005, 22:50 CET using branch tag Nv0_2_8_beta4_2.
And I'm afraid you have to resend the recording :( Postfix comes with a preconfigured 10MB message size limit, and fetchmail throws away messages exceeding that instead of keeping it on the server like it does for its own size limit. Grrr and sorry, I was unaware of this quirk. I bumped the limits up to 200MB per message.
Yeah, EMail is quirky. Especially since most admins/developers think their list of usecases is exhaustive. The problems I had with EMail ... but that's another story. I am sending it again.

Update: First the mail server didn't respond, and now the mail client has crashed (!). I'll try the webinterface.
Update2: The webinterface accepted the upload with the awesome speed of 1 KB/s. 10 Minutes later, the connection broke. I'll try the mailclient again, after all, it worked last night ...
Update3: My ISP is the culprit. Even if the connection is idle, I have 20% packet loss. On ADSL! Grrr ... I'll try again later.
meriton
Round Winner
Posts: 256
Joined: Sun Nov 20, 2005 3:33 am

Post by meriton »

Did you get the recording?
User avatar
Z-Man
God & Project Admin
Posts: 11585
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Yes, thanks, it arrived safely this time. I'm having the usual trouble playing it back, but I haven't tried running it in Windows yet.
Edit: After a little editing (for some reason, the loading of the master server list appears twice in the recording), it now plays back fine to the error. It appears the server is reusing network objects IDs too fast.
User avatar
Z-Man
God & Project Admin
Posts: 11585
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

I think I've got a fix. It's not perfect, but probably sufficient: freed network object IDs are not recycled immediately anymore, they are put into a queue that has a minimal fill state.

Whenever rubber is currently active, turn commands from the client are executed at exactly the TIME the client says they should be executed. There was some slack regarding this previously. In local experiments, this eliminated all instances where the server would calculate a higher rubber usage (and thus possibly an earlier death) than the client.

Both changes will be in effect after the next restart in a couple of hours. Maybe I'll manually trigger one in a few minutes so I can be online to monitor possible ill side effects immediately.
User avatar
Lucifer
Project Developer
Posts: 8640
Joined: Sun Aug 15, 2004 3:32 pm
Location: Republic of Texas
Contact:

Post by Lucifer »

I've got another feature request. :) This one comes from lag, actually. The ability to disable color codes in player names, so as to eliminate mistakes from looking at just player names in the list of players and trying to associate what teams they're on. I know, it's weak, and maybe you added this when you fixed another, more serious color exploit? Or maybe it's been there for awhile. Anyway, if you can do this, lag would probably appreciate it. :)

(I know, it's weak, there is a column giving team membership and all....)
Image

Be the devil's own, Lucifer's my name.
- Iron Maiden
User avatar
Z-Man
God & Project Admin
Posts: 11585
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Accepted and added for the next server restart. There already was the option to filter all color codes from the clients, but this may be overkill.

Oh, and I forgot: Thanks a million, meriton! For all: if the client disconnection bug happens again on Fortress, just reconnect and complain in the usual way ("BUG! Random disconnect!") even if you don't have a clientside recording so I know whether it is still there.
meriton
Round Winner
Posts: 256
Joined: Sun Nov 20, 2005 3:33 am

Post by meriton »

Code: Select all

for (int i=0; i<1000000; i++) {
    cout << "You're welcome." << endl;
}
:)
User avatar
Z-Man
God & Project Admin
Posts: 11585
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Code: Select all

[4] received logout from 4.
[4] meriton left the game.
[4] meriton leaves Team blue.
[4] Killing user 4, ping 0.129509.
...
[4] New user: 4
[4] meriton entered the game.
[4] meriton plays for Team blue.
...
[4] meriton: BUG: netowork error. i'll send the recording later.
Right, got it. The recording is much smaller this time and requires less guesswork. In fact, my guessed cause was wrong. The real reason for the bug is clientside: the client connects just as some objects get destroyed. Unfortunately, it gets the object destroy message before the object creation message and does not notice that. When it gets the destroy message, it does nothing. When it later gets the creation message, it creates the object. Since it should no longer exist, trouble is ahead when later the object's ID is recycled. Should be easy to fix now.

The previous "fix" was not wasted work, it could lead indeed to the same observed error.

For you all, this means no more disconnection reports unless you're using a client from tomorrow's CVS.
meriton
Round Winner
Posts: 256
Joined: Sun Nov 20, 2005 3:33 am

Post by meriton »

meriton wrote:When connecting tonight, my client displayed cycles outside the grid until the round ended.
I'd like to add to that that the gauges on the HUD can also be affected by a similar bug. In the attached recording, the client thinks really fast cycles (speed = 2300) are around and rescales the speed gauge accordingly. The ridiculos scale persists across the entire session.

Personally I don't care, but in case other do you now have a recording.
Attachments
ridiculos speed gauge scale.aarec.bz2
(38.1 KiB) Downloaded 79 times
User avatar
Lucifer
Project Developer
Posts: 8640
Joined: Sun Aug 15, 2004 3:32 pm
Location: Republic of Texas
Contact:

Post by Lucifer »

I think when the needle hits the end of the gauge, it should just fall off and show no indicator of speed. :)
Image

Be the devil's own, Lucifer's my name.
- Iron Maiden
User avatar
joda.bot
Match Winner
Posts: 421
Joined: Sun Jun 20, 2004 11:00 am
Location: Germany
Contact:

Post by joda.bot »

ok, I've got two other bugs... ehhh just remember one now ...

no recordings yet :)

Somtimes the player names (above the cycles) get yellow ... this depends on the last color of the console as far as I can tell it. At least scrolling the console up and down changed the color between white or yellow.

I'll post the other bug once I remember what it was ... I should have a block at hand while playing.

EDIT: reminds himself not to bother z-man about bugs while playing "Apology" attached
Attachments
armagetron_beta4.2_console_color_leak.tar.bz2
5sec into the game ... shows console color leak 0.2.8beta4.2 client used
(51.48 KiB) Downloaded 84 times
Last edited by joda.bot on Wed Jan 11, 2006 1:19 pm, edited 3 times in total.
User avatar
Lucifer
Project Developer
Posts: 8640
Joined: Sun Aug 15, 2004 3:32 pm
Location: Republic of Texas
Contact:

Post by Lucifer »

Are there still that many bugs that you need a block while playing? ;)

Personally, I prefer to wait until they all crawl out at once, and then jump on my keyboard. My wife tells me to quit wearing my shoes in the house, so I might not be able to do that anymore.
Image

Be the devil's own, Lucifer's my name.
- Iron Maiden
User avatar
2020
Outside Corner Grinder
Posts: 1322
Joined: Thu Dec 29, 2005 9:21 pm
Location: the present, finally

my gratitude

Post by 2020 »

can't really contribute to the tecnical talk around here

just wanted to say
once again
thanks for the gameplay design of fortress

amazing

last night i found playing in the cockpit view
with upto 4v4
ideally 3v3
was sweet

a real rush

and thanks to all the developers for the entire game-design
of armagetron
hold the line
Post Reply