The simulation loop design

[ohlidalp]

I have a very large redesign of RoR in my head (devmaster9 is just a start), and I'd like every further git-commit to be a step towards that design. I'm using a GitHub issue rather than RoR forum because my employer blocks RoR website, and I need to stay in contact.

I'm writing this down to help myself calm down in the current coding hurricane and to sync up with other devs' ideas and visions. This doesn't mean I want to re-shuffle anything, just to explain why I focus on some things and ignore others.

Terminology

• Task (capital T) = a single job in thread pool.
• Sim (capital S) = All gameplay state - data representing physics and other simulations. They should be updated only on constant time step (PHYSICS_DT), but for historical reasons many updates are done on per-frame basis with elapsed time as parameter. To see why it's wrong, read https://gafferongames.com/post/fix_your_timestep/
• SimBuffer = a copy of game state (For example node positions, drivetrain state, player vehicle pointer etc...)
• Flex = deformable meshes (flexbodies, wheels, cab...)

Devmaster9

For each rendered frame:

1. Sync Sim Tasks (threadpool)
2. Legacy per-frame Sim + OGRE updates:
   1. Local player inputs (GUI, keyboard, mouse, gamecontrollers...)
   2. Network inputs
   3. Scripting
   4. Camera
   5. Stuff... :smile:
3. Update SimBuffers
4. Update simulation (Start Sim Tasks)
5. Update OGRE scene and render:
   1. Start Flex tasks
   2. Update non-Flex OGRE scene elements, including GUI
   3. Sync Flex tasks, update related OGRE scene elements
   4. Invoke rendering

Ideal

For each simulation step (PHYSICS_DT):

1. Update player inputs (GUI, keyboard etc...) - Will enable recording/replaying macros and exact benchmarking
2. Update network inputs (UDP protocol) - Will enable networked collisions.
3. Invoke scripting (SIMULATION* hooks, see https://github.com/RigsOfRods/rigs-of-rods/issues/1558#issuecomment-420234219)
4. Update simulation
5. If it's time to render, update SimBuffers and launch all OGRE scene updates + scripting (FRAME hook) + rendering on background.

Softbody data handling.

• See sticky forum thread Development/Enter the codebase - the main spot for my rants.
• We need to unify referencing nodes to uint16_t - currently we mostly use int or node_t* which is incorrect (we can't have more than 2^16 nodes) and has limitations - you can't divide data to multiple arrays (such as ar_nodes + simbuf_nodes). See also https://github.com/only-a-ptr/rigs-of-rods/blob/dev-master-9/source/main/datatypes/node_t.h#L31-L34
• This review needs to be looked into: https://github.com/only-a-ptr/rigs-of-rods/pull/20 - summary: we need an unified mechanism for retrieving position and orientation of the actor.
• Threading: https://github.com/only-a-ptr/rigs-of-rods/pull/22#issuecomment-423994007

Feedback needed

This is what I want to archieve in the long run. Does it make sense? What steps (and in what order) will get us there?

[ohlidalp]

Closing as partially done and generally out of date. The remaining points were moved to #2981.