The nonlinearity maybe comes from a bug that has been fixed in the 0.2.7.1 branch only; sensors that reach more than 10 units beyond the rim return wrong measurements ( too large ).
Are the errors still present on the CVS test server 2? That would be bad, because it's already running more or less the 0.2.7.1 code from the branch.
CVS test server
-
Jonathan
- A Brave Victim
- Posts: 3391
- Joined: Thu Feb 03, 2005 12:50 am
- Location: Not really lurking anymore
Yesterday LaTeX went through all walls and through the rim. I didn't lose contact with the server.z-man wrote:Are the errors still present on the CVS test server 2? That would be bad, because it's already running more or less the 0.2.7.1 code from the branch.
My improved chatbot (don't expect much from barely 20 lines of stateless code) sometimes tries to drive through a wall because it doesn't see it (I see what the chatbot sees on the HUD), but then boom.
I'll try to do it again.
Thanks! That's definitely the same bug as in the other thread. I'll apply the fixes to the CVS server in an hour or so. The distance to the sides is shown to be zero by your sensors because you're actually still inside a very narrow channel ( probably really of zero width ) and the sensors know that. your_mon found out how to escape the channel: set cycle_wall_length to 10 when you are outside. this will cause the walls bordering the channel to disappear and you are free.
Great. The stab bug has been replaced by a crash. I'm investigating while CVS 2 is down.
Edit: preliminary result: Jonathans cycle got destroyed. Not killed, but utterly and completely destroyed. Deleted from memory. < Insert parrot sketch script >. Although it was referenced by the its walls that were still in memory. I've sent another server instance into an endless loop. There's layers upon layers of bugs, you fix the bugs in the top layer and the lower, even uglier bugs start to appear. That's what I call fun
EditEdit: The cycle was deleted because its reference counter overflowed. Due to, well not really bugs, but inefficiencies in the grid code, thousands of wall segments get created, each increasing this reference counting variable of the cycle. Guess what: it's a signed short, so after 32000 wall segments are created, the cycle gets destroyed. I'll make the reference counter a full int and let the cycle commit suicide when the refcount reaches a configurable threshold.
Edit: preliminary result: Jonathans cycle got destroyed. Not killed, but utterly and completely destroyed. Deleted from memory. < Insert parrot sketch script >. Although it was referenced by the its walls that were still in memory. I've sent another server instance into an endless loop. There's layers upon layers of bugs, you fix the bugs in the top layer and the lower, even uglier bugs start to appear. That's what I call fun
EditEdit: The cycle was deleted because its reference counter overflowed. Due to, well not really bugs, but inefficiencies in the grid code, thousands of wall segments get created, each increasing this reference counting variable of the cycle. Guess what: it's a signed short, so after 32000 wall segments are created, the cycle gets destroyed. I'll make the reference counter a full int and let the cycle commit suicide when the refcount reaches a configurable threshold.
The server is running on ADSL at my home and the shitty provider cuts the line every 24 hours or so. When dialing in again after some seconds, I get a new IP and so all connections are lost. No disruption of any kind appears here on the logs.Jonathan wrote:It was the same as now: unreachable on the server list after it timed out while playing. Oops!
On second investigation, the server seems to be caught in a nasty loop. It really is not responding currently.
On third investigation, I should not have added a function that has a worst case O(2^N) complexity. Thanks, Jonathan!
Edit: CVS 1 is running again with the code from RC5.
-
Jonathan
- A Brave Victim
- Posts: 3391
- Joined: Thu Feb 03, 2005 12:50 am
- Location: Not really lurking anymore
My chatbot is really productive... Sometimes I just leave it on a test server for a while because it can do many crazy things. In this case it turned on every tick because it got partly stuck and its turns are delayed (still need to look at that, but it should never do that). No 1052 fps for me, but enough to accidently DOS a server it seems.
Edit: OK, I can get close to 1000 fps when turning off some rendering on this laptop, and over 1000 fps between rounds. But when playing online things slow down to 200-300 fps on CVS 1 (as usual on OS X - see nemo's post somewhere).
Sometimes I just leave it on a test server for a while because it can do many crazy things. In this case it turned on every tick because it got partly stuck and its turns are delayed (still need to look at that, but it should never do that). No 1052 fps for me, but enough to accidently DOS a server it seems.Edit: OK, I can get close to 1000 fps when turning off some rendering on this laptop, and over 1000 fps between rounds. But when playing online things slow down to 200-300 fps on CVS 1 (as usual on OS X - see nemo's post somewhere).
It's back!
Named "Z-Man's 0.2.8 CVS /Test Server/ (NO FUN!)" violently color coded (sorry, users of old clients). It's running current CVS from the 0.2.8 branch in optimized mnode and is set up to record everything and notify me of problems per mail. I'll update the source every once in a while, but not automatically as this could ruin the recordings.
Bug reports will have maximum usefulness if you set your client to record the action as well and attach the result.
Custom connect info:
server: l04.thp.uni-koeln.de (That's "lima zero four" at the beginning for the font-challenged)
port: 4538
Named "Z-Man's 0.2.8 CVS /Test Server/ (NO FUN!)" violently color coded (sorry, users of old clients). It's running current CVS from the 0.2.8 branch in optimized mnode and is set up to record everything and notify me of problems per mail. I'll update the source every once in a while, but not automatically as this could ruin the recordings.
Bug reports will have maximum usefulness if you set your client to record the action as well and attach the result.
Custom connect info:
server: l04.thp.uni-koeln.de (That's "lima zero four" at the beginning for the font-challenged)
port: 4538