Fixed vs variable timestep

Just curious, are you using a fixed or variable timestep in your "update" routines ?

This has been a thorny issue for me, and what the recent work i was doing was all about.

The world updates at a variable rate, basically as fast as possible. If a lot of water is being updated, say when soemone breaches a lake from a cave then updates will happen slower.

I use a priority system to try to manage this.

Player initiated actions (digging, building etc) update with high priority.

New chunks being added over the horizon because the player moves are a low priority but some work time is reserved to process these.

World updates are middle priority but again, some time is reserved to process these also.

I see

I was facing this kind of problem lately with my own voxel renderer.
(Was doing like your are doing, updating linked to the render loop).

While everything was working properly, a faced strange "physic" bug on a slower machine : My collision was not working properly (Fast objects moving through walls). Very difficult to reproduce because it was linked to the execution time of the render loop varying too much between iteration ! What are you doing to prevent this ?

Until I read this paper :

http://gafferongames.com/game-physics/fix-your-timestep/

So now my render loop looks like this :

Code:


//very simplified loop
while()
{
FixedStepUpdate() // Only executed at fixed time step
Update()              // Once per iteration for stuff that needs to be updated as fast as possible
     
Draw(Interpolation value) // Draw with states interpolated between previous and current frame

}


But it asked me to review in depth my core engine !

I do something similar if i've picked you up right.

I monitor the frame rate and adjust the distance moved each update to compensate. So it always takes you a certain amount of time to run 10 blocks distance regardless of the frame rate.

That's variable fixed time your are using, the problem with this methods is mainly "unforseen freeze".

Let's say that the user launch another application while your program is running, it leads to a "freeze" of 1 second because of high CPU usage.

If you use a variable delta time in your equation (Current update - Previous Update time), in this case it could lead to unpredictible results in physic simulation.

Example with collision : The 1 second freeze could lead to have object missing collision with other objects. And this could be very difficult to debug because it is linked to a situation difficult to reproduce. The link added in my previous message describe clearly this situation.