BUG cries on Bugfarm Fortress and Sumo

For things that have to do with those crazy test servers... and yeah. By request of z-man, and, of course, you gotta obey...

Moderator: Z-Man

Post Reply
User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Code: Select all

[0] dookatee |DpG| core dumped ~}SD{~Upacreek for 2 points.
[0] ~}SD{~Upacreek core dumped dookatee |DpG| for 2 points.
[3] ~}SD{~Upacreek: lol
[4] dookatee |DpG|: was that possible?
[4] dookatee |DpG|: was that a bug?
[3] ~}SD{~Upacreek: wild want me to def?
[3] ~}SD{~Upacreek: no i was ahead ;)
[1] !LOL!WiLdCaT: yea idk what i did
[1] !LOL!WiLdCaT: u d
[3] ~}SD{~Upacreek: lol k
[4] dookatee |DpG|: BUG: Z-man it seemed like one of those we're on both sides of the wall
There are may possibilities to double suicide even at close range, exlposions have a finite expansion speed, for example. What exactly happened?

Code: Select all

[10] gArIsImO: how did I die? if I didn't even grind BUG
Packet loss or lag spike. The server received a turn command, but only after it had declared you dead.

Code: Select all

[5] wrtlprnft |mbc|: BUG dookatee |DpG| thinks i went through a wall last round
Well, you'd be the authority for that :) There is the well known problem that old clients that don't support the trail rubber shrinking see dead players' walls a bit longer than they actually are, looks like dokatee is using one. I'd fix it by resetting the rubber of dead cycles back to zero so the wall shrinking gets disabled, but once a cycle is dead, syncs are unreliable.

This one also looks like the same thing:

Code: Select all

[0] Hay-Lin renamed to Z-Girl.
[0] Z-Girl committed suicide.
[9] Z-Girl: BUG
[9] Z-Girl: i went through a ******* dead wall

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

00Saen was doning something strange this night. His client sent totally nonsensical turn commands. They neither contained a turn nor a brake status change. The server interpreted that as packet loss and waited a bit for missed packets. Does anyone know what he did? If it's not a hacked client, then it's a client bug. Rain, goodtv, Wildcat and Chuck were online at the same time.

User avatar
Rain
Round Winner
Posts: 300
Joined: Sat Apr 15, 2006 2:59 pm
Location: a random empty server playing with bots

Post by Rain »

°°Saen is °°Hay-Lin, so "she" came there because I asked to show me the bug she had. She told to me that the bike did a lot of turn without touch any commands. So I took a recording of it. I suppose it is a client bug. I wrote the report and attached the recording here: http://forums.armagetronad.net/viewtopic.php?t=6021

Greetings,

rain

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

There have been a couple of cries about the following problem: Two small teams are fighting. One of the teams suddenly quits. The other team is left on the field and has to suicide, and worse, the quitting team's score is remembered when the teams are balanced the next round.

So now, when the team count goes from 1 to 2 or 2 to 1 on any server (also useful on last team standing servers), a new round and a new match are started. A tiny problem is that if you leave with "/teamleave", then the change will be detected too late, and the player who is left alone will play a quick round that immediately ends.

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Code: Select all

[6] SL(meriton): BUG: my client did not receive updates for about 1.5 seconds
If it was a real network problem, nothing I can do. I fixed a bug that caused you not to get updates on enemy cycles while they just go straight, though, that may have been it.

Code: Select all

[2] [nrx]: BUG: Death even with tail vanish
Turns out the prediction for the tail end movement used to compensate simulation times of the two cycles was going the wrong way. Fixed now.

Code: Select all

[1] ~Cloud~: BUG odd team behavior with xell
A bit more specific, please?

Code: Select all

[9] 2020: BUG ?
[8] Zop: BUG. collision between 2020 and me... both died
I'm looking at that one right now, but the previous cries of that kind were caused by the initially surviving player not having enough rubber to survive until the explosion reached him.

Code: Select all

[0] madmax core dumped ~*SP*~whatup for 2 points.
[0] Lag slide for ~*sp*~whatup: (-8.06232,0), rubberSpeedFactor 0
[8] Kamakazi: WHA????
[12] ~*SP*~whatup: ?
[12] ~*SP*~whatup: thats bs
[8] Kamakazi: TOTAL BUG
[12] ~*SP*~whatup: BUG
[9] !L0L!Pyr0TeK: that was bs
[8] Kamakazi: ;p
[5] madmax: what bug?
Death can be delayed quite a bit by lag and packet loss compensation. The lag slide message tells me that here, whatup's death was delayed at least 1/3 of a second, probably more. And clients mispredict collisions at times.

A general note, if you want to report all kinds of "odd" effects and lags and jumps and stuff that shouldn't be there, please send a clientside recording and the timeindex where the strange thing happens. Serverside recordings are really only useful if you know what you need to look for.

User avatar
Van-hayes
Round Winner
Posts: 398
Joined: Sat Mar 04, 2006 1:15 am
Location: The Maritimes

Post by Van-hayes »

Tonight around 6:30 pm-7:30pm Atlantic, I think that is 21:30 gmt-22:30 GMT(its 7:30pm here when I posted this), there was some weird lag/slides and what I can only describe as Jumps on Bugfarm. Here is a recording of it, there is a big one right at the start.

BTW I was using the name Tionuan, and called out bug a few times I think during that time
Attachments
ArmagetronAdvancedDebugRecording.zip
Jumps
(260.75 KiB) Downloaded 331 times
You've gone too far, turn back!

User avatar
Van-hayes
Round Winner
Posts: 398
Joined: Sat Mar 04, 2006 1:15 am
Location: The Maritimes

Post by Van-hayes »

Image
This happened at I guess 22:27 GMT, Sticky hit my wall but then there was about a 5 second pause, then he suddenly died and his tail went past where he died.The time might be a few minutes off.
You've gone too far, turn back!

mkzelda
On Lightcycle Grid
Posts: 14
Joined: Sat Apr 23, 2005 7:40 pm
Location: Raleigh, NC
Contact:

Post by mkzelda »

I have a recording where *my* bike froze while driving. Someone else in the game said they experienced the same thing. I experienced this two other times in the same night, but this one I actually cried bug and have a recording. You must fastwforward to 1330 seconds. You might want about 10 seconds before that to see it better. The way I experienced the bug in real time differed from the way it appears in the recording, although something wrong is still happening. When it actually happened, I was frozen for a few seconds until I pressed turn. However I was surprised that when I hit turn, I did not jump to some other spot, but really had stopped in place.

This happened about 2-3 days ago. My user name was "Diplo" at the time and I cried "BUG: zman: i stopped moving"

Since the forum limits attachments to 2mb, you can download the recording here.
A funky not a junky but I know where to get it...

mkzelda
On Lightcycle Grid
Posts: 14
Joined: Sat Apr 23, 2005 7:40 pm
Location: Raleigh, NC
Contact:

Post by mkzelda »

I just recorded this of players jumping/bouncing about on bugfarm. I didnt mark any points because there was a fair amount of jumping going on the whole time.
Attachments
tron06.rec.bz2
Having trouble with attachments here for some reason...
(1.39 MiB) Downloaded 267 times
A funky not a junky but I know where to get it...

mkzelda
On Lightcycle Grid
Posts: 14
Joined: Sat Apr 23, 2005 7:40 pm
Location: Raleigh, NC
Contact:

Post by mkzelda »

Yet another lag jumping recording. Watch as webgamer jumps an enormous margin the first round. The second round has a few little jumps.
Attachments
tron07.rec.bz2
(316.36 KiB) Downloaded 246 times
A funky not a junky but I know where to get it...

mkzelda
On Lightcycle Grid
Posts: 14
Joined: Sat Apr 23, 2005 7:40 pm
Location: Raleigh, NC
Contact:

Post by mkzelda »

Here is another recording. At 1900 secs I get an all black screen, no 3,2,1 count and then I'm dead.

A round or two after that we had the biggest lag jump I've ever seen. Johnnney is playing for blue, runs back to def, freezes as he runs into a wall, I leave the area since I have no idea where he is, about 5+ seconds later his wall appears only after I hit it.

Again, the recording is too large for the forum so you can find it here.
A funky not a junky but I know where to get it...

mkzelda
On Lightcycle Grid
Posts: 14
Joined: Sat Apr 23, 2005 7:40 pm
Location: Raleigh, NC
Contact:

Post by mkzelda »

At this point I feel like I'm annoying but I hope to give you something that helps. This recording lasts about 5 minutes. It seems the freezing is limited to certain players, I'd guess client versions. You'll see in this one that "alf n XP" and "|nc|Johneyy" are the freezing clients repeatedly. The initial round has alf frozen. I think the 3rd round has both frozen. The last round in the recording was interesting enough to justify the 5th post in a row :P
Attachments
tron09.rec.bz2
(773.77 KiB) Downloaded 232 times
A funky not a junky but I know where to get it...

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

Thanks for all the material, that will give me something to chew on.

I actually did play back the diplo freeze serverside recording; it looked like a regular clientside misprediction lag jump from that perspective. But I'll tackle it again after seeing the clientside version.

The 2O2O vs Zop collision was what I thought: first, 2O2O dies, and a split second later, Zop. The distance between them was something like 3.5 meters, the explosion didn't travel fast enough. One unlucky problem with lag compensation is that while all the other wall actions can be easily extrapolated over short times both into the future and the past, so the server knows where a wall is at any time and whether a wall will disappear or not, the same is not true for explosions. They can't be predicted. The result: with high lag compensation values that delay deaths and explosions to later real times, you sometimes crash into bits of wall that should have already been blown away. Nothing can be done about it, apart from disabling all forms of lag compensation.

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

I looked at the serverside recording belonging to Van's screenshot. Van's cycle was stuck between two turns for about two seconds; the stickiness was only broken by his next turn. Feather and Sticky were unlucky to have crossed his path (well, feather hadn't really crossed it yet, but was about to and probably surprised by the wall popping up, it was probably already behind him on his client). Sticky had traveled quite a bit and made some turns, that's why he had to get killed: in game time, Van just was at the collision point earlier. The stickyness bug should be fixed after the restart just now. I'll try to think about something that would clean up Sticky's path after the collision, he shouldn't have been on that side of the wall after all.

User avatar
Z-Man
God & Project Admin
Posts: 11357
Joined: Sun Jan 23, 2005 6:01 pm
Location: Cologne
Contact:

Post by Z-Man »

I'm ignoring those bug cries that sound like the kind of "lag" the bug from the previous post caused, or alternatively can be explained by real network problems. I'll still have a look at the clientside recordings to check whether they fit my theory, of course :) Here's the list:

Code: Select all

[2] madmax --> Teammates: **Defense Needs Help**
[0] Lag slide for titsnass: (-0.000488281,1.50595), rubberSpeedFactor 1
[12] TitsnAss --> Teammates: HOLE IN DEF OR DEF NEEDS HELP
[9] Tionuan: wow
[2] Granting 0.0111713 seconds of lag credit.
[11] =LOL=Chuck --> Teammates: DEF IN TROUBLE!! HELP DEF!!
[0] Lag slide for zop: (-0.000595093,6.2905), rubberSpeedFactor 1
[0] Zop core dumped madmax for 2 points.
[0] Lag slide for madmax: (4.29253,0), rubberSpeedFactor 0
[0] madmax core dumped Zop for 2 points.
[11] =LOL=Chuck --> Teammates: Incoming
[9] Tionuan: BUG
[5] Granting 0.00191402 seconds of lag credit.
[2] madmax: ???
[7] SL (Sasha) --> Teammates:  ! ! !  ATTACK   ! ! !
[5] Granting 0.0011673 seconds of lag credit.
[11] =LOL=Chuck --> Teammates: DEF IN TROUBLE!! HELP DEF!!
[1] Pandemonium: what the hell? i just lagged
[8] Zop: weird
[0] Team blue was awarded 10 points for conquering the enemy base.
[0] Winner: Team blue
[2] madmax: i think we all did

Code: Select all

[7] ~*SP*~whatup: wtf was that
[3] Granting 0.00556946 seconds of lag credit.
[7] ~*SP*~whatup: who saw that
[2] CTDimbiman: aw shit lag
[7] ~*SP*~whatup: BUG ZMAN
[7] ~*SP*~whatup: biggest lag ever

Code: Select all

[1] [nrx]: BUG: Major 'lag' issues

Code: Select all

[4] dookatee |DpG|: BUG Big jumps at round start last time
...
[7] dookatee |DpG|: BUG: WebGamer jumped, gonna post it next round
I'm especially ignoring those, because they are quite unhelpful and lack all information:

Code: Select all

[5] trace(LoL);: this BUG sucks

Code: Select all

[6] ~*SP*~whatup: BUG zman fix your damn connection!
Sure, buy me a server on a real server farm.

Code: Select all

[11] -|ct|- play: i cant play this is teleporting me
[10] TitsnAss: i know!
[10] TitsnAss: i hate this
[11] -|ct|- play: BUG
[11] -|ct|- play: BUG

Code: Select all

[1] SL(Infamous): BUG
(On Sumo, fighting the AI, without further explanation)

Normal mode on those:

Code: Select all

[1] Zop: BUG. when triple turn, third leg disappears temporarily
[1] Zop: BUG. after triple turn, seem to move sideways
The first is a clientside bug that is fixed now. The second one is also clientside, but not fixable totally; I still get it shortly after entering a server when the timer is not synced completely. But after some seconds, it goes away. If it still happens with the next alpha, I'd need a clientside recording.

Code: Select all

[3] °°rain: BUG!
[3] °°rain: BUG: he goes throw a wall
Who?

Two of those:

Code: Select all

[1] WS dookatee DPG: BUG: take a look at that
At what?

Code: Select all

[3] SL(belenus): BUG: you still running this with DEDICATED_FPS 1 Z?
Never was :) Lagfarm is running with 10 fps, fortress was on 20 fps temporarily, but is up to 40 now again so I get a better separation of bug reports. I'll turn it down later again. Before the new network code, it was on 50.

Code: Select all

[11] madmax: BUG: and round eleven without a tie at 100 or something like it
Yeah, there was a tie, a round winner was declared, and while the round was still going, the tie was broken, so the next round was really short. The round after that, too, that definitely is a bug. I'll have a look when I have time.

Code: Select all

[11] =LOL=Chuck: NO WAY!
[11] =LOL=Chuck: BUG: Trail was fully gone, but i still died
On your client, maybe, but the server will just tell me that the wall was just vanishing. I need clientside recordings for those.

Code: Select all

[2] WS dookatee DPG: BUG: dunno, round ended as soon as it started despite teamswtich correct
Should be fixed now. Wrong order of function calls somewhere.

Code: Select all

[3] M&Ms: BUG
[3] M&Ms: BUG: sasha crashed but his trail continued
and

Code: Select all

[9] Tionuan: Bug: Pandemonium was killed by TitsnAss  but his tail went further than where he was killed
Yes, rubberless kills did that. Should be fixed now, what happens now is this:

Note: the automated BUG cry refers to Bacteria's death, not root's.

Code: Select all

[0] gArIsImO core dumped |AST| BaCtErIa for 2 points.
[0] Possible BUG, new code: removing future walls of the cylce that just got killed mercilessly.
[0] !lol! pimp core dumped CT|root for 2 points.
[0] Lag slide for !lol!_legit: (0.109283,-0.0057373), rubberSpeedFactor 0.188816
[6] WS dooaktee DPG: hah
[0] -Delta- core dumped ]i[ 2020 for 2 points.
[7] CT|root: what
[6] WS dooaktee DPG: wd golds
[3] ]i[ 2020: ?
[7] CT|root: that was not right
[5] -Delta-: LOL BUG!
[9] Granting 0.0264559 seconds of lag credit.
[6] WS dooaktee DPG: way to split
[9] Granting 0.0338259 seconds of lag credit.
[6] WS dooaktee DPG: :D
[0] Lag slide for !lol!.durka: (-0.239853,0.000396729), rubberSpeedFactor 0.427652
[0] Lag slide for !lol!_legit: (-0.17012,3.05176e-05), rubberSpeedFactor 1
[0] Lag slide for !lol!_legit: (-7.62939e-05,-0.158539), rubberSpeedFactor 1
[0] Kamakazi core dumped gArIsImO for 2 points.
[9] Granting 0.00749207 seconds of lag credit.
[2] gArIsImO: shit
[4] New user: 4
[8] |AST| BaCtErIa: No shit durka :P
[0] Lag slide for !lol!wildcat: (0.105194,-0.000152588), rubberSpeedFactor 1
[2] gArIsImO: wtf happened...
That question interests me, too. The proximity of the automated cry and root's apparent strange death can't be a coincidence. Did the happen close together? Maybe the vanishing of Bacteria's walls had some odd side effect?

There's still a long list of other cries, but I think the serverside recording is enough once I have time to look at them.

Post Reply