Skip to main content

March

Monday 2nd​

  • Investigating potential issue with rail invalidation not rewinding far enough at high time speed
    • Adding extra information to the inspector for data pages
    • Doesn't seem to be a real bug
  • Cleaning up February notes
  • Modifying orbit rail picking to be done mostly in parallel jobs, instead of the main thread.
    • Probably slower for a small number of picks, but scales much better for large numbers.
  • Added live-integration based extrapolation back into rail sampler (temporarily removed last month while developing live integration system)

Tuesday 3rd​

  • Drafting "what is kessler" for Discord
  • Considering sensor/tracks refactor
    • Current system:
      • Relationships:
        • Sensor Platform (E) -> Sensor (E)
        • Sensor Platform (E) -> TrackCollection (C) -> Track (E)
        • Sensor Platform (E) <- Track -> Target (E)
      • System updates sensor(s), sensor updates track(s)
        • This is a problem, sensor updates do a lot of spatial querying (expensive) and chase a lot of entity relationships (expensive)
    • New idea:
      • Rather than actually scanning the sky for tracks, create a new track for every detectable object in the universe when it is created. Only remove tracks when the target entity is destroyed.
      • Attach SensorTrackable to things
      • Implement SensorTrackLifecycleSystem
      • Implement SensorTrackable on destroy logic

Wednesday 4th​

  • AFK

Thursday 5th​

  • Implementing new life cycle for sensor tracks
    • When a new Trackable entity is created, create a new Track entity for every TrackCollection
    • When a new TrackCollection is created, create a new Track entity for every Trackable and add it to the collection.
    • When a Trackable is destroyed, remove it from all TrackCollections
    • When a TrackCollection is destroyed, destroy all Tracks that it owns
  • Added new way to get deferred entities after they're created in Myriad.ECS
  • Fixing bug in Myriad.ECS collect queries
  • Track entities are not getting created
    • Trackables and platforms are created on the same frame, so neither creation logic runs!
  • Phased array radar cannot track things (because position deviation of new tracks is infinity)
    • Expected - need to add broadband scanning
  • Need to add a max cap on deviation values now that they're no longer culled when large

Friday 6th​

  • Converting deviation decay systems to jobs
  • Converting sensor cross reference systems to jobs
    • Not doing this, they all reference the same data as each other so there's less parallelism available without a refactor
  • Improving TypedEntityReference in Myriad.ECS (faster, simpler)

Monday 9th​

  • Improving visualisation in RADAR test scene
  • Creating components for tracks, for the new sensor updating system
  • Created system to initialise tracks with sensor info
    • Need to calculate time factor for tracking
  • Created system for RADAR Scan
    • Implement this
    • Line of Sight check
      • Shared system between all sensor systems for kepler body positions?
  • Created system for RADAR Track
    • Implement this
  • todo:
    • Tracking beam size
    • Tracking duration
    • Line-of-Sight Obstacles
    • Remove unused fields of PhasedArrayRadar

Tuesday 10th​

  • Updating RADAR track system to decide which tracks it should track
    • Overall tracking is slower, the more targets there are
  • Creating RADAR scanning
    • Count contacts
    • Decide which ones to track
    • Determine beam angle and SNR based on worst case assumptions about track position/range etc
  • todo:
    • Line of sight
    • Remove unused fields of PhasedArrayRadar
    • Set threshold values (perfect/infinite error)
  • Profiling
    • GetRadarCrossSection: 5.26ms
      • This can be cached!
    • CalculateSNR: 5.31ms
    • AdjustTrackingResolutionSNR: 2.37ms
    • UpdateDeviations: 1.83ms

Wednesday 11th​

  • Creating system to calculate all kepler positions every frame, sensors systems can use this to check LoS
  • Caching RCS (radar cross section) per-track
  • Pre-calculating some values in RADAR scanner and re-using them every frame
  • Jobbify RADAR scan
    • Can't do this with current structure, list (components) contains a list (of sensors)
  • Created system to do LoS checks once, all sensors can use this
    • RADAR track
    • RADAR scan

Thursday 12th​

  • Investigating GitHub CI runner for Unity projects
  • Debugging issue with resource disposal in job queries
    • Job gates take a copy of the handle, so it's being disposed twice (on both copies)
    • Needs future work to improve this
  • Testing obstacles/LOS with RADAR
    • Wrong range calculation
    • LoS check is inverted
    • LoS should be line segment not infinite ray
    • comparing SNR and SNR(dB) is wrong
    • Properly handling zero range (clamp to 1mm minimum)
    • Fixed RADAR tracking time factor

Friday 13th​

  • More work on CI
    • Access token for private packages
    • Hanging forever while running edit mode tests?
  • Improved sensor track inspector (fixing perf issues with lots of tracks)
  • Setting up test coverage reporting
  • Tests for various helpers/extensions:
    • Colour
    • Span
    • Stats
      • Variance is wrong (incorrect weight)
      • StdDev is wrong (same reason)
    • Find Min/Max
    • List
    • Rail serialisation
    • Quaternions
      • IsNaN checks check is incorrect
    • Enum
    • Plane
    • ULong
    • v128/v256
    • TimeSpan

Monday 16th​

  • Updating editor version to 6.3.11f1
  • More tests!
    • long3
      • Fixed off-by-one bounds checking, actual memory safety error!
    • BoundsMinMax
    • BitList128
    • BitList256
    • FixedSizeStackList
    • Metre3
    • LineSegment3D
      • Sphere intersection test broken
      • Line closest point not clamping to segments properly
    • BoundsSphere
    • Double3Extensions
    • NativeHashMapExtensions
    • NativeArrayExtensionsTests
    • RootFinder
      • Inverse interpolation checks the bounds of the guess incorrectly, sometimes guessing a worse value
    • NativeSliceExtensions
    • NativeHashMapExtensions
    • NativeMultiHashMapExtensions
    • SimpleRandom
    • NormalizedRay3D
      • Double normalisation (not a bug, just inefficient)
    • UnityMathematics.Random.Extensions
    • System.Random.Extensions
    • Up to 13.2% total coverage
    • ColorConverter
      • Falling back to default colour (magenta) due to incorrect formatting
    • JSON SingleOrArrayConverter
    • InterceptCalculator
      • Failing to calculate intercept during projectile boost phase
        • Running out of iterations in the root finder
        • Reduced max-time, no point searching for intercepts so far into the future
  • Designing orbital intercept solver
    • Writing a lambert solver, going to need this to solve intercepts in keplerian space

Tuesday 17th​

Wednesday 18th​

  • Tests
    • Octree
    • Parallel Octree
    • Ray3D
      • Fixed NaN for parallel/zero length rays
    • ExponentialMovingAverage
    • SweptBoundingSphere
      • Incorrect AABB
    • DualRailZipper
  • More menu work

Thursday 19th​

  • Tests will continue until morale improves
    • SceneUnit
    • SceneUnit3
    • SingleExtensions
    • NetworkWriter/NetworkReader extensions
    • JobExtensions
    • GameObjectExtensions
    • BoundsCone
    • long3 operators
    • Root finder fuzzing
      • Fixed bug in handling negative ranges which caused it to diverge away from solution!
    • Particle swarm optimisation
    • 22.6% coverage
  • Creating scene for testing orbital gunnery
    • Systems for shooter and defender
    • Tag components
    • Refactoring system/scenario loading
  • Adding some new capabilities to campaign/scenario loading
    • "Parallel" orbit generation
    • Vector generators

Friday 20th​

  • Fixing RootFinder bug exposed yesterday
  • Gunnery test scene
    • Setting up system to shoot at defender
    • Spawning bullets - need to setup position sync components for them!
    • Bullets are not attached to rails - rail sampler query is not picking them up for some reason
  • Refactoring NBody setup
    • Split into 3 components: NBody, LiveNBody, RailNBody
    • New setup struct - forces choices of one of the three integrator types (rail, live, none)
    • New editors for the new components
    • Update integrator systems to use new components
    • Fix all call sites
  • Fixed thread safety issue in Myriad.ECS query jobs

Monday 23rd​

  • Investigating why test bullets are not having their position set
    • Recent changes for extrapolation in rail system added a requirement that Nbodies have the capability to make burns (needed for extrapolation). Bullets don't have this!
  • Added 1 second of extrapolation to all rails when first created, gives time for position systems to warm up
  • Added activation to scene objects, so they can be disabled until the position is first set
  • Investigating parallelising FindClosestVisualProximitySystem
    • Needs access to Entity which isn't possible in a job query
    • Exposing EntityId array in Myriad.ECS
      • todo: use this in job queries

Tuesday 24th#​

  • Updating Myriad+Unity integration to expose entities to job queries
  • Updating FindClosestVisualProximitySystem to use job for some of the work
  • Attaching gun (using bullet chain system) to test ship
    • System to assign targets to guns
    • System to points guns at targets and shoot
      • Setting up gun rotation local transform
      • Optimising bullet link culling

Wednesday 25th​

  • Refactoring some more bullet culling systems
  • Encountered exception testing at high time speed
    • IndexOutOfRangeException in Myriad.ECS.Worlds.Chunks.Chunk.AddEntity
    • Adding tests to cover this
      • Reproduced the error - full chunks are not handled properly in some cases
      • Fixed
      • Also optimised Archetype.HasComponents checks while working on this
        • Replaced set membership check with array lookup
  • Created basic system to point guns at assigned targets
    • Shooting the wrong way?
      • Need to use WorldTransform to determine gun direction for aiming

Thursday 26th​

  • Implementing gun aiming properly
    • It works!
  • Switching job query scheduler to use finer grained scheduling (per archetype+component, instead of just per archetype)
    • Updating safety system
    • Updating all query types
    • Updating Unity+Myriad package

Friday 27th​

  • Improving bullet rendering
    • Shorter
    • Brighter
    • Additive blending
    • Replaced texture with a soft curve
    • Randomised brightness
      • Removed this, don't have a stable per-bullet ID to use in the shader for randomisation
  • Setting up test projectiles with collisions
    • Target needs:
      • BroadphaseSphere
      • NarrowphaseCollider
        • Need to tag collider leaves with entity ID
          • Made this optional - use the root entity if the leaf is not tagged with an entity
  • Optimising collision handling system
    • Use typed reference for BulletNarrowPhaseHit?
      • LinkEntity
      • ColliderId
      • No point, hit is only used once so caching doesn't help!
  • Fixed damage model lookup (no longer fetching armour damage model from the bullet!)
  • Setting up damage model
    • Target needs ArmourComponents.DamageModel
    • Bullet needs BulletDamageContainer
      • Attach to gun, it will be auto copied to bullets
  • Narrowphase collision handling is incredibly expensive
    • Looks like it's the job completion that's taking time
      • Maybe due to the use of a ParallelQueue (it's safe, but still has thread contention to maintain an ordering guarantee that I don't actually need)
      • Investigating writing a custom NativeBag parallel collection
        • Maybe I can use NativeStream?
        • Using NativeArray<UnsafeList>
        • No change D:
      • innerLoopbatchCount was set way too high. Problem solved.
      • This could be expanded to a pipeline with more parallelism
        • Job 1 processes pairs into individual bullet rays
        • Job 2 goes parallel over individual rays

Monday 30th​

  • Added temp code to delete entities as soon as they take a hit from a bullet
  • Setting up kinematic chain for turrets (base rotates to azimuth, barrel rotates to elevation)
    • Temp component for rotating base to target yaw
    • Temp component for rotating gun cradle to target elevation
    • Port temp stuff to permanent system
  • Messing with some ideas for rendering flares from phased array lasers

Tuesday 31st​

  • Finished up temp systems for turret pointing
  • Added a new system resource cleanup to Myriad.ECS
    • Using this in bullet system group
  • Ported turret rotation code to a proper system
  • Researching assignment problem (optimally assigning PDCs to incoming targets)
    • Algorithms:
      • Hungarian algorithm is optimal, but O(n^3)
      • Greedy assignment (assign best gun to best target, repeat) is fast but non optimal
      • Auction Algorithm is fast and near optimal
    • Some extra complexity:
      • Multiple targets (jobs), more than number of PDCs (workers)
      • Costs change constantly (every bullet that is fired reduces the danger, since that target has a probability to be destroyed)
      • Multiple guns could attack one target
      • Some guns might want to target groups of close objects (e.g. flak)
      • Some guns may attack multiple targets, but get less efficient for every one they attack simultaneously (e.g. phased array laser)
  • Building system to link together gun targeting and turret kinematic chain