Author |
Message |
Tillie
Joined: 2012-02-09 21:01:50 Posts: 284
|
Happened for me again, too. But could make the prim jump back to previous position by selecting it and then click undo in the top menu. Forgot what exactly I clicked before, will keep an eye on it next time.
|
2012-11-19 18:48:35 |
|
|
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5545
|
I need the logs (the new warnings I implemented in the latest releases are more explicit about where the NaN exception happens) and, if possible, what angles the prim was being rotated to when it happened.
|
2012-11-19 19:28:19 |
|
|
ErikaThorkveld
Joined: 2011-12-13 14:11:38 Posts: 186
|
I don't have any logs to provide since I'm on a Mac and the latest version is not available there… Can someone please reproduce the problem and post the logs here? This bug is becoming extremely annoying. I'm basically fearing to start editing anything now, because I just can't spend a single building or editing session without the bug happening at some point. For now, I didn't lose anything luckily, but that might happen: the coordinates 0, 0, 0 in a region are often quite inaccessible, and if something gets down there, I may not be able to ever get it back…
|
2012-11-28 21:34:32 |
|
|
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5545
|
I could never reproduce this bug here under either Linux or Windows... Perhaps is it specific to the Mac builds (fast math optimization that causes bad roundups and NaNs ?)...
|
2012-11-29 02:55:24 |
|
|
ErikaThorkveld
Joined: 2011-12-13 14:11:38 Posts: 186
|
Got it! Log attached. The editing that made the poseball jump to the position 0, 0, 0 starts around 21:45. Lots of warnings about a non finite number after that indeed…
|
2012-12-03 21:56:56 |
|
|
Ibrew Meads
Joined: 2010-03-14 21:12:58 Posts: 86
|
Alas it's happened on Linux too. My friend had it happen on 32 bit Ubuntu and I've had it happen on 64 bit Debian. Unfortunetly when we discovered this was what happened, and recovered our items from near <0,0,0> several days later, we'd long since lost the logs. I can't even be sure of the server version since it may have changed since then. I'll keep a watch for it and send logs if I can get it to happen again. Sorry I can't be more help yet, but it's aparently not isolated to the Macs.
|
2012-12-05 02:21:19 |
|
|
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5545
|
The only logical solution I came up to is that the "gimbal lock fix" patch that went into v1.26.4.36 is causing these NaNs... I'll revert that patch in next releases.
|
2012-12-05 09:14:25 |
|
|
Henri Beauchamp
Joined: 2009-03-17 18:42:51 Posts: 5545
|
Please, folks, let me know if v1.26.4.42 and v1.26.5.22 got this bug fixed...
|
2012-12-14 09:15:42 |
|
|
Ibrew Meads
Joined: 2010-03-14 21:12:58 Posts: 86
|
It looks like it is fixed, Henri. I was able to get it to happen consistently in 1.26.5.21 by creating a box (in a skybox at 1600m, not sure if that matters), sitting on it and then rotating it in small increments pressing the control key and dragging the green circle. I'll send a log if you still want one. I couldn't get it to trigger the bug with 1.26.5.22. If it happens again in 1.26.7.0 I'll let you know. Otherwise I think you got it fixed.
|
2012-12-15 20:22:22 |
|
|
ErikaThorkveld
Joined: 2011-12-13 14:11:38 Posts: 186
|
Urgh… Bug is back: I wasn't sure so far, but this time, I did see it. Using version 1.26.12(9) on the Mac, and I had again an object jumping to coordinates 0,0,0 while I was rotating it using the red/green/blue circles. It was a posestand I was sitting on. It was after quite a long session though, so I fear the full log won't fit. Here is what I think is the meaningful part: The lines around this in the log don't seem to be related, so I guess they won't bring anything. Tell me if I should get a larger extract of the log though.
|
2014-08-06 19:44:59 |
|
|