Contact listener guarantees about order of callbacks

[zopsicle]

The documentation for ContactListener reads:

contact listener callbacks are called from multiple threads at the same time

Is it possible that, within an update, a pair of bodies both start colliding and stop colliding, but the contact listener receives the corresponding OnContactAdded and OnContactRemoved calls concurrently or out of order?

[jrouwe]

Is it possible that, within an update, a pair of bodies both start colliding and stop colliding, but the contact listener receives the corresponding OnContactAdded and OnContactRemoved calls concurrently or out of order?

No, during a simulation step (parameter inCollisionSteps in PhysicsSystem::Update) you either get an OnContactAdded/Persisted or you get an OnContactRemoved callback for a contact, never both. If you passed inCollisionSteps > 1 then you can see an 'add' and a 'remove' during the same call to PhysicsSystem::Update but they will be ordered correctly (possibly called from different threads, with a memory barrier between the two calls).

Note that a contact is defined as (body a, subshape a, body b, subshape b), you can e.g. receive an 'added' call for (body a, subshape a, body b, subshape c) and a 'remove' for (body a, subshape a, body b, subshape b) in the same step. Removes are always called at the end of a simulation step so even if logically a body disconnected with subshape b first before connecting to subshape c you will receive add c, remove b in this particular scenario (again the calls can happen from different threads, but there will be a memory barrier).