The last errors rather look like you added/removed objects twice, could it be you fixed something like that on the way too? Raytest, right… That bugger might basically cause all kinds of calls…
Waiting on rayTest / shape synchronization causes lag in the rendering thread, which defeats the whole purpose of wanting them multithreaded in the first place. You mentioned native bullet shouldn’t have these problems, correct? Is this as simple as just using the native libraries instead?
@phr00t said:
Waiting on rayTest / shape synchronization causes lag in the rendering thread, which defeats the whole purpose of wanting them multithreaded in the first place. You mentioned native bullet shouldn't have these problems, correct? Is this as simple as just using the native libraries instead?
..and living with its bugs, yeah. Native Bullet isn't finished.
@phr00t said:
Welp.. guess I'm going to give it a go. I suppose finding bugs and reporting them will help move the libraries along? :D
Sure but most collision stuff needs to be rewritten anyway because its a contribution, so I don't know how much single details really lead to solutions. But coding and building the natives yourself using our project and NetBeans is very easy too ;) Final move to Native Bullet is designated for 3.1.
@phr00t said:
How "bad" is it? :P Is there some nasty known bugs, or just likely a bunch of unknown ones?
Just a heads up if you're playing with rayTests, last I checked (a few days ago) native bullet reverses the hit fraction, so you'll get "far" reported as "near" and the reverse. Might save you a bit of frustration to know :)
@jmaasing said:
Just a heads up if you're playing with rayTests, last I checked (a few days ago) native bullet reverses the hit fraction, so you'll get "far" reported as "near" and the reverse. Might save you a bit of frustration to know :)
This is corrected in the sources but idk to what binaries it really went because the windows build for native bullet on our server is broken due to mingw issues :( It should be fixed in android and macos at least in RC2
normen said:
This is corrected in the sources but idk to what binaries it really went because the windows build for native bullet on our server is broken due to mingw issues :( It should be fixed in android and macos at least in RC2
Oh, I'd better check again then, I'm on RC2-nightly linux. It ought to work then. Sorry for the false alarm.
@normen said:
.. I'd rather look for a solution for the jbullet implementation to be clear :)
I would too... I got the exceptions to stop, but at the cost of practically all benefits I was trying to achieve from multithreading. It really helps to be able to handle heavy physics calculations in one thread, while doing quick physics stuff in another... so far, it seems like jbullet just can't do this.
I remember reading about the rayTest reversal stuff earlier.. I can wrap the rayTest calls and switch the hitfraction value if needed, thanks for the heads-up. I'll let you know if it is still swapped now.
EDIT: Yes, rayTest hitfraction is still swapped on Linux 64 (if it is fixed for MacOSX, then my code won't work on MacOSX if I "fix" the swap for Linux 64?) :/