fix some quirks in World.step when attempting to account for variable framerate #18
Conversation
…d of exhausting maxSubStep every frame
…eady increase in internalStep
|
@Platane if you know your way around cannon as you seem to do, would you want to have github access? if this lib can grow faster it would be awesome and i think it needs more hands. |
|
@drcmda I am not very familiar with the technical aspect of the engine and solver unfortunately. But I can certainly give some of my time and do my best :) |
| @@ -407,14 +405,16 @@ export class World extends EventTarget { | |||
| substeps++ | |||
| } | |||
|
|
|||
| const t = (this.accumulator % dt) / dt | |||
| // prevent the accumulator to build up delay to catch up. The logic being: if the step did not catch up at this frame, it is unlikely to catch up in the next one. Even if it does, it will result in a speed up simulation which is arguably worse that staying behind the real time. | |||
| this.accumulator = Math.max(this.accumulator, dt) | |||
There was a problem hiding this comment.
so how does this behave diifferently than setting this.accumulator = this.accumulator % dt? Just a question, can't really wrap my head around what would happen differently 😅
There was a problem hiding this comment.
That's a good question!
To my understanding, the accumulator is useful to catch up a frame when for example the simulation runs at 50fps and the step is 1/60ms. The accumulator growing greater than dt is an unexpected behaviour.
afaik, once we agree to discard some simulation time, it does not matter much if we carry the modulo or not.
I arbitrary decide to use a hard limit because of the semantic, as using a modulo might be interpreted as being important in the interpolation after ( which is not, the interpolation does not make any sense in case the accumulator gets truncate )
…step. which allows to pass timeSinceLastCalled=0 and have a noop
|
Closing until we can resolve the simulation speed and freezing body issues, see #16 |
Limit accumulator growth as proposed in schteppe#392
Also fix what seems to be a mistake where this.time is increase in
world.stepand at the same time inworld.internalStep. I am unsure about the implications, this value was used to handle sleep status in bodies.This fix unfortunately does not address #16 as I was still able to reproduce the midair freezing cubes. I notice it seems to happens less frequently though. The investigation continues :)