On this page
Saturday 4th Swept collision test prototypes Monday 6th Updating packages Fixing some project rot: Investigating up-sampling settings (Unity STP) Enabling GPU resident drawer Updated Sonity Fixed some build errors in new version Imported Mirror Constructed a quick menu (host/join/exit) Installed Multiplayer Play Mode Created menu flowmenu -> lobby -> gameplay Lobby sync stuffCreating a UI "card" for each player in a way that supports late joining Things not spawning on clientsMaybe something to do with ready states? It was something to do with ready states. Setup BasicAuthenticator with username/passwordOnly using password for lobbies Tuesday 7th Working player metadata (initially names) Working out why VS debugger won't attachOr it attaches, but breakpoints don't work This has been an intermittent problem for a while Refresh action fixed it this time Back to names for playersSaving name Splitting out to separate component, general pattern for player metadata in lobbyCan't use generics in a load of places related to Mirror, making factoring out a generic data object impossible. This may become a problem... Refactored again in separate partsLobby Data ObjectContains the actual data, and the SyncVar/RPC/Commands necessary to sync it One per player, separate component for each bit of data Lobby UI PanelOne per player Component for each "data binding", two way binding reads and writes to the networked data object Top level components which walks the tree, telling each component to setup bindings Wednesday 8th Split data objects into base classes (can't make it generic, so I have to duplicate it one per type). Added BaseLobbyPlayerBooleanData and LobbyPlayerReadyData Creating UI elements for ready indicator Setting up host start button to disable itself until everyone is ready Setup bindings for collections of objectsUsed this to show number of ready players on start button Modifying player list UI ready for teams Discussed naming on Discord Refactored various network utility behaviours into more atomic behaviours and relocated them into main project Removed mostly useless sender parameter from UI event stuff Thursday 9th Refactoring lobby player data into more general lobby data Designing new player list UI which supports teams Syncing team data Binding team data to UIShowing name Showing max count per team Showing empty slots Changing team list when a scenario is selected Added test scenarios Showing scenario name in lobby UI Friday 10th Removed scenario test code Added description field to scenarios Fixed some network events being invoked twice on the host Moving players cards into team panels Experimenting with another team panel UI design Removed concept of team slots, it's a lot of unnecessary UI complexity Adding buttons to empty slots to join teams Moving players between teams when button is clickedFixing empty slot (de)spawning as players move teams Moving players to team zero and clearing ready state when changing scenario Playtest Impromptu docking sim v2 playtestRendering issues, mostly skybox flickering (possible due to VA screen?) Saturday 11th Testing multiplayer connectivity on LAN Fixed connectivity (probably) Possible issue with cards not despawning when players disconnect Investigated Steamworks transport backend for Mirror Monday 13th Tuesday 14th Investigating details of Mirror scene loading Added a UI element to switch network backends Syncing all scenario settings, storing them in scene data so they're accessible in the gameplay scene Setting up a script which reads scenario data and configures scene loading scripts Removed audio listeners in all planet content scenes, prevents warning about multi listeners during loading Added other game scene stuff (camera rig, postprocessing volume, symbol renderer) Started building loading overlay (blocking view of game until loading is complete) Sketched out some scripts for the networked loading processAlso ensuring this process can work without a server, so I can load the scene directly for testing in editor Wednesday 15th Syncing player data from server to clientsConsider refactoring this into an ECS system later Setting player to ready as soon as the game scene is loaded (even before planets are streamed in) Investigating using Resources/StreamingResources to store campaign/scenario files Working out a data format for campaigns and scenarios Debugging scene objects not spawning in game sceneNeed to request a player object before doing things in scene Spawning an empty player object Building out more scenario serialization formatRapidly remembering why XML sucks to use Fiddling with XML serialisation of colours todo: Thursday 16th Investigating System.Text.Json for Unity Unity.SerializationJson.NETModified lobby to use scenarios loaded from file Building UI for time speed controllerShowing:Current time Pause Button for each speed Binding time display text in ECS, experimenting with how to drive UI in ECS paradigm Updated Myriad.ECS to return query count from delegate queries, consistent with all other query types. Working out a rough data model for representing remote players in the ECSNeed this to store time speed requests somewhere, so I can query them in the UI bindings Building system for setting tooltip to list of players requesting this speed Friday 17th Saturday 18th Updated Myriad.ECS package in main project Added option to enable "auto destruct" (i.e. destroy GO when entity is destroyed) Added optional reverse destruct (i.e. destroy entity when GO is destroyed) Binding UI with autodestruct both ways Binding player objects with autodestruct one way (GO -> entity) Using tagged disposable to store UI event bindings in ECS Sunday 19th Building a better entity/GameObject destruct system, with a flags enum determining mode Discovered a bug in Myriad, leaving entities in an invalid state if a phantom entity is modified and deleted in the same command Fixed that bug Monday 20th Updated Myriad/Unity binding package to use the version with the bugfix Added some basic stats to the inspector which could help track down similar issues in the future Fixed some other minor inspector bugs Fixed issue with skybox reflections on smooth surfaces (e.g. Earth water) Debugging issues in docking sim caused by new binding stuffSystems aren't running because setup isn't finished Setup isn't finished because it's waiting for the binding system to do something Fixed by moving binding system to a separate group that is always running, even before setup is complete Cleaned up MyriadEntity binding (storing less redundant things, returning a nullable which integrates better with C# nullability checking) Fixed breakage in main project from those changes Improved planet loading - all planet objects are deactivated in their respective scenes and are only activated once loading is complete Fixed broken skybox reflections on water (reflection map data was baked at some point) Made a start on time sync systems (finding the minimum request across all players) Updated Space Graphics Toolkit Tuesday 21st Worked out general flows of user inputs to game actions through network Started implementing that for time sync system Implemented inspector for EventBusService listing all event buses Wednesday 22nd Hooking up the CMD/RPC flow for time speed requests Spawning player data entities in game scene Time speed requests work Tweaking time speed UI slightlyDifferent colours for local and remote requests List of players requesting the speed (shown on hover) also shows the time speed for that button Cleaning up ready status/player spawn logic Thursday 23rd Improved Map/Reduce queries in Myriad.ECS Considering how to apply CPU backpressure to time speed requestsAs integration rails get shorter (below some limit), time should slow down to allow the integrator to catch up When a rail gets too short (less than 30 seconds at the current playback speed) time speed is set to 1x Added an indicator to the UI when a time speed request is coming from a backpressure override Friday 24th Adding in a button to spawn random nbodies, attempting to deliberately overload the integratorExtracting code from other ad-hoc places that spawn nbodies, starting refactoring into a set of helpers Spawning rails with negative rail length does a pretty good job at overloading the integrator. Oops Profiling with lots of nbodiesLots of unexpected allocations in WorldHost.Update Every NBodyLineRenderer costs 1,038,476 vertices, that's too manyIt's also not visibleNot a bug, just need to send a SetRailRelativeOrigin event before they become visible Monday 27th Directly using the integrated data in NativeArray, instead of copyingThe copy was necessary before the changes on Friday, to copy from Unity NativeArray to normal C# array Worked out some more of the work schedule for the next few months, organised some issues/tags Moved some unfinished items from the multiplayer sprint to backlog issues ready for a second multiplayer sprint later Designing an LOD system for NBodyLineRenderer to reduce vertex countUpdating vertex shader to add LOD support Simplified pixel shader Refactoring line renderer to pick an appropriately sized mesh, instead of always using the largest Investigating how to pick LOD levelFor each bounding sphere: determine screen size, pick largest, some function to map from size -> LOD level Create a Unity job to find the largest apparent size of bounding spheres, triggering it every frame if the job is not already running Ignoring scene camera when finding camera for LOD calculation Tweaked values mapping from apparent size to LOD level Tuesday 28th Ignoring parts of orbit line which are off screen for LOD calculation Tweaked min and max LOD values Testing with 100 orbits at 21600x realtime (6 hours/second)RailIntegrator main thread time is spiking each time it runsSynchronousIntegration is running, slowing down the main thread, this is probably happening more often than necessary.Disabled synchronous integration entirely, it should only be used in extremely bad cases and arguably just makes things worse by slowing the main thread even more! ConvertPagedRailToRelative main thread time is also spikingThis isn't in a job yet, no surprise this is slowing us down BoundsMinMax.Create is the biggest cost, which is a surprise!Still a lot of verticesStarted converting lines from 4 vertices per point to to 3 Modified line generator script to produce 3 vertices per point Fixed up NBodyOrbitLine assuming 4 verts, in just one place SetWorldPositionFromRail could maybe do with some speedups900us for 163 entities (5.5us each) This is mostly just the plain cost of calculating kinematic interpolation Wednesday 29th Converting ConvertPagedRailToRelative to use jobsModifying PagedRail to store a list of jobs which are currently referencing it, protecting the arrays it holds from being deallocated while a job is in flightTurns out this isn't needed, we can just copy the arrays and modify them in place for the relative rail. Rewritten relative paged rail conversion so it works in place on existing arrays, this makes it a lot easier to do in a job and removes the need for the PagedRail to track job dependencies Profiling relative rail code Considering design for "partial" queriesPass in a "Cursor" object which tracks how far through the query execution got Query can be interrupted (e.g. max entity count limit) Executing query again (next frame) using the cursor resumes from there There's a lot of work that could be split up like this Apply burst attributes to relative rail conversionBurst doesn't like some type punning tricks with Span Rewritten to do the same with NativeArray.Reinterpret doesn't work either! Thursday 30th Applying BurstCompile to more code Rewriting ConvertPagedRailToRelative to use a parallel-for job for offsetting, and a chained job for bounds generation Updated SetWorldPositionFromRail to do less work most of the time:Near the start point, just predict forward Near the end point, just predict backward In the middle 10%, do both and smoothly interpolate between them Reviewed all jobs in the project, applied BurstCompile where possible Rewritten SIMD code in BoundsMinMax.Create to work with burst Implemented Cursor queries in Myriad.ECSExecution of a query has a budget, when that budget is exceeded execution is stopped and the cursor is set to the current position. Executing again with the same cursor resumes at approximately the same position. This way any world that needs doing regularly, but not instantly can be spread over multiple frames. Friday 31st General code cleanupMoving some events into the events namespace Cleaned up extension methods Cleaned up some JSON converters Testing assets:UGUI Super TreeView Seems good! Added to FUI project, done lots of code cleanup (all fairly minor) Started working out styling with FUI