2,602 Commits over 639 Days - 0.17cph!
Merge: from fix_boat_no_sail
- Bugfix for boats and vehicles getting stuck in open areas, effects not playing, building appearing slow
Tests: unit and perf tests, built and sailed a boat on craggy
Bugfix: fix player boats getting stuck while sailing (and other weird bugs)
- Network subscription now creates discovered cells
Was caused by overly aggressive network group lazy creation. If they are created within player subscription range, player doesn't get subscribed to them automatically, which caused all kinds of issues. This'll hurt perf over time, but no worse than current perf on Release - I'll optimize this later
Tests: unit tests, perf tests and built a boat and sailed across craggy
Merge: from stableobjectcache_taa_nulls
- Update: Replaced StableObjectCache with StableObjectArray, updated TriggerParent, Projectiel batching and server player processing to use it
- Bugfix: fix for "Adding null to TransformAccessArray" warning
Tests: unit tests, 2p session on Craggy with jumping on boat, shooting, conencting-disconnecting
Update(tests): add validation of StableIndexLookup to StressTest
Tests: ran unit tests
Clean: delete StableObjectCache, no longer in use
- fixed up const references that I missed
Tests: compiles
Update: add missing NativeDisableParallelForRestriction to BasePlayerJobs.RecacheTransforms
Tests: none, trivial change
Update: rewrite projectile batching to use StableObjectArray instead of StableObjectCache
- internal statics are now initialized and cleaned up on client connect/disconenct (+ scripts reload)
- had to sprinkle container access checks due to indirection
This was the last user of StableObjectCache, will rip it out next
Tests: ran unit tests, 2p craggy test with shooting (confirmed fly by effects, ricoches still happen)
Update: rewriten TriggerParent to use StableObjectArray instead of StableArrayCache
- ran codegen
- made DisposeInternalState run on assembly reload
Tests: built a boat, jumped on off - no issues (except the boat not moving, but a separate bug)
Bugfix: ensure PlayerTransformsAccess stays in sync with PlayerCache when players are removed
- also removed safety if as it's obsolete after recent change to cleanup player before shutting down the server
Tests: had 2nd client connect-disconnect to me on Craggy (though technically it's not enough, I'm relying on unit test logic)
Update: StableObjectArray gets GetIndexForSyncRemove to support removing from linked collections
- used it to fix TestTransformAccessArraySync
Need to update actual game code since it's bugged
Tests: ran unit test
Update(tests): add unit test that validates StableObjectArray is in sync with TransformAccessArray
Tests: ran the test - they're not, gotta fix it tomorrow
Bugfix(editor): cleanup connected players before we shutdown server when stopping playing in editor
Tests: started and stopped play mode
Bugfix: if shorevector caching is disabled, then make sure we generate it
Prev version skipped if it found the cache files on disk even though it didn't load them
Tests: booted into craggy
Bugfix: properly dispose of BasePlayer.PlayerCache
- also invoke DisposeInternalState on assembly reload
Tests: started and stopped playing on Craggy
Clean: rename BasePlayer.StableIndex -> BasePlayer.ActivePlayerInd
Eventually we'll have SleeperInd and etc, so doing this change early
Tests: compile all modes pases
Clean: delete PlayerCache and it's unit tests
Tests: compiles
Bugfix(editor): rengerate client-side shorevector cache if it's size changed
Tests: connected to a server running craggy
Update: StableObjectArray rewrite - version 2
- Replaces PlayerCache on the server in all the relevant paths
Still need to do clientside projectiles stable cache before ripping it out
Tests: booted into Craggy, ran around. Need to test 2p, but being stopped by some shore vector out of bounds access, hoping sync fixes it
Bugfix: StableObjectArray.StableIndexLookup was returning unbounded array
Tests: booted into craggy and ran around
Update: useful bits from the first failed StableObjectArray conversion
It ended up being too pervasive for my liking, going to try a simpler approach
Tests: compiles as part of rewrite, but none otherwise
Update: StableObjectArray now exposes stable index lookup map (needed for transform access array jobs)
- moved NativeArrayEx to Facepunch.UnityEngine plugin, enabled unsafe code, added Unity.Collections dep
Tests: none, code is in the middle of breaking rewrite
Clean: dead test code removal
Tests: compiles
Update: conversion to StableObjectArray - step 2, WaterLevel tests
Tests: ran WaterLevel unit and perf tests
Update: conversion to StableObjectArray - step 1, AntiHack unit tests
Tests: ran antihack unit and perf tests
Update: implement StableObjectArray.Repack
- update unit tests to better track consistency of the state
- expand StressTest to validate Repack
Going to start replacing StableObjectCache in code tomorrow. Also gotta stop unit test from allocating, it takes too long
Tests: ran StableObjectArrayTests
Update(tests): add StressTest unit test
Next up going to try to figure out how to support Repacking
Tests: ran unit tests
New: StableObjectArray, a pending replacement for StableObjectCache
- added basic tests, needs more
Follows TransformAccessArray API, provides similar stability guarantees as StableObjectCache but requires to be interacted with differently.
Tests: ran unit tests
Merge: from unitask_reliable_taskpool
- Optim: UniTask internal pooling now reliably caches items
Tests: unit and perf tests, ran around Craggy with Jobs 3
Optim: ensure UniTask.TaskPool reliably caches items
- copied from repo at 88737fb5
Original version was written in wait-free fashion which permitted only 1 participant while failing others. Rewritten to be lock-free, which allows to guarantee pooling
Tests: ran unit and perf tests. stuck breakpoints to confirm that repeated perf test invocation didn't have allocations. ran around craggy with jobs 3
Merge: from remove_netgroup_prealloc
- Optim: NetGroup objects are now lazy allocated, saves 150MB
Tests: ran unit tests, ran around on craggy
Update: cleanup ClientVisibilityProvider when disconnecting from the server
Tests: disconencted from local server
Update: dispose netgroups when NetVisGrid gets disabled
Tests: booted into craggy and stopped - no errors
Clean: remove Provider.Init method
Tests: ran UpdateSubscriptionConsistency test, booted into craggy
Optim: Get rid of NetGroup pre-allocation, use thread-safe lazy allocation
Comparing memory snapshots, we go from 2.1mil NetGroups down to 131, saving 140MB
Tests: booted to Craggy and ran around. ran UpdateSubscriptionsConsistency unit test.
Clean: move group initialization from PopulateCells to GetGroup
Tests: ran around on craggy
Clean: remove Visibility.Manager.groups
- also remove all provider == null checks
Tests: compiles
Update: add ClientVisibilityProvider - mostly a dud Provider, most methods do nothing
- Upodated all places to use ClientVisibilityProvider instead of null
This allows to remove groups from Network.Visiblity.Manager
Tests: booted into craggy, ran GameTraceTests
Update: instead of iterating all netgroups, add Provider.ForEach(layer, callback) method
- get rid of Visibility.Manager.Groups accessor
This removes public access to groups dictionary, so almost there to remove it
Tests: booted to craggy, opened deep sea and ran deepsea.printentitycount
Update: Visibility.Provider now can create network groups in a thread-safe fashion
- organized hardcoded groups into their own collection
- Network.Visibility.Manager now defers to provider's GetGroup and TryGetGroup
Not sure if I want to support NetworkVisibilityGrid on CLIENT or not, but will try to clean it up in a bit
Tests: Booted and ran around on craggy. compiled for all modes separately
Merge: from useplayerupdatejobs_purge
- Clean: Removal of UsePlayerUpdateJobs 0 and 1 code
- Optim: RelationshipManager now uses cached server occlusion results instead of running new ones
- Bugfix: stop NPCs/Bots writing tick history, corrupting internal memory
Tests: booted a server from a save and connected to it
Bugfix: avoid invalid tick transformation from NPCs/bots that would corrupt array header
- added an assert to GetPlayerTickIterator, only place where it's not checked by default
Tests: booted a standalone server with a save - was able to connect and run around
Update: replace all unsafe usage with spans, should throw if I go out of bounds
Tests: compiles
