feat(lifetime)!: debug info and wait_no_references

[TestingPlant]

wait_no_references allows the packet bump allocator to wait until there are no more references before clearing the bump allocator instead of aborting if there are any references. Debug information is also added.

[github-actions]

Benchmark Results for general

ray_intersection/aabb_size_0.1                     [  34.1 ns ...  34.1 ns ]      +0.05%
ray_intersection/aabb_size_1                       [  33.4 ns ...  33.5 ns ]      +0.13%
ray_intersection/aabb_size_10                      [  24.2 ns ...  24.3 ns ]      +0.17%
ray_intersection/ray_distance_1                    [  13.7 ns ...  13.8 ns ]      +0.47%
ray_intersection/ray_distance_5                    [  13.8 ns ...  13.8 ns ]      -0.24%
ray_intersection/ray_distance_20                   [  13.8 ns ...  13.8 ns ]      -0.09%
overlap/no_overlap                                 [  24.8 ns ...  24.8 ns ]      -0.03%
overlap/partial_overlap                            [  25.1 ns ...  25.0 ns ]      -0.42%
overlap/full_containment                           [  22.5 ns ...  22.5 ns ]      -0.15%
point_containment/inside                           [   4.9 ns ...   4.9 ns ]      -0.00%
point_containment/outside                          [   4.4 ns ...   4.4 ns ]      +0.08%
point_containment/boundary                         [   5.0 ns ...   5.0 ns ]      +0.13%

Comparing to 4c402d0

[github-actions]

Benchmark Results for general

ray_intersection/aabb_size_0.1                     [  34.0 ns ...  34.0 ns ]      -0.08%
ray_intersection/aabb_size_1                       [  33.8 ns ...  33.8 ns ]      -0.11%
ray_intersection/aabb_size_10                      [  25.9 ns ...  25.9 ns ]      +0.10%
ray_intersection/ray_distance_1                    [  13.6 ns ...  13.5 ns ]      -0.44%
ray_intersection/ray_distance_5                    [  13.6 ns ...  13.6 ns ]      +0.22%
ray_intersection/ray_distance_20                   [  13.6 ns ...  13.6 ns ]      -0.12%
overlap/no_overlap                                 [  24.8 ns ...  24.8 ns ]      -0.16%
overlap/partial_overlap                            [  24.8 ns ...  24.9 ns ]      +0.08%
overlap/full_containment                           [  22.5 ns ...  22.6 ns ]      +0.24%
point_containment/inside                           [   5.5 ns ...   5.4 ns ]      -0.42%
point_containment/outside                          [   4.9 ns ...   4.9 ns ]      +0.63%
point_containment/boundary                         [   5.0 ns ...   5.0 ns ]      -0.36%

Comparing to 4c402d0

[codecov]

Codecov Report

Attention: Patch coverage is 26.73267% with 74 lines in your changes missing coverage. Please review.

Project coverage is 20.87%. Comparing base (4c402d0) to head (3a0bedc).

Files with missing lines	Patch %	Lines
crates/hyperion-utils/src/lifetime.rs	26.00%	73 Missing and 1 partial ⚠️

@@            Coverage Diff             @@
##             main     #888      +/-   ##
==========================================
+ Coverage   20.85%   20.87%   +0.02%     
==========================================
  Files         161      161              
  Lines       16867    16931      +64     
  Branches      468      473       +5     
==========================================
+ Hits         3518     3535      +17     
- Misses      13285    13332      +47     
  Partials       64       64              

Files with missing lines	Coverage Δ
crates/hyperion/src/net/mod.rs	6.91% <100.00%> (ø)
crates/hyperion-utils/src/lifetime.rs	18.28% <26.00%> (+4.77%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[andrewgazelka]

can you go a little more into this?

in which scenarios would all the references not be 0 by the end of the gametick? I could see this causing issues if people accidentally hold onto references of say one event and then many bumps (massive memory) is never cleared

[TestingPlant]

in which scenarios would all the references not be 0 by the end of the gametick?

See #861 as an example of where this would occur. When events such as ChatMessage are processed with a multi_threaded system, the event processing threads may still be running and holding references while the clear_bump system runs and calls assert_no_references.

I could see this causing issues if people accidentally hold onto references of say one event and then many bumps (massive memory) is never cleared

This doesn't create multiple bump allocators. It waits until all references are cleared before clearing the bump allocator. If a reference is leaked, then wait_no_references will emit warnings every 5 seconds including debug information about which reference was leaked.

[andrewgazelka]

in which scenarios would all the references not be 0 by the end of the gametick?

See #861 as an example of where this would occur. When events such as ChatMessage are processed with a multi_threaded system, the event processing threads may still be running and holding references while the clear_bump system runs and calls assert_no_references.

I could see this causing issues if people accidentally hold onto references of say one event and then many bumps (massive memory) is never cleared

This doesn't create multiple bump allocators. It waits until all references are cleared before clearing the bump allocator. If a reference is leaked, then wait_no_references will emit warnings every 5 seconds including debug information about which reference was leaked.

ideally there should be a sync point before assert_no_references though where it "reads" all event related things. I think we should prio fixing sync points over this.

[TestingPlant]

ideally there should be a sync point before assert_no_references though where it "reads" all event related things

wait_no_references is a sync point; the thread is parked until all references are dropped, meaning that all events with references have also finished processing. Once the last reference is dropped, the thread calling wait_no_references is unparked.

[andrewgazelka]

ideally there should be a sync point before assert_no_references though where it "reads" all event related things

wait_no_references is a sync point; the thread is parked until all references are dropped, meaning that all events with references have also finished processing. Once the last reference is dropped, the thread calling wait_no_references is unparked.

what I am saying though is why not use flecs for this natively instead of having our own custom logic? flecs has sync points natively.