Move more functions off of ScheduleGraph

[ItsDoot]

Objective

• Part of Extensible Schedule Graphs #20115
• Reduce ScheduleGraph implementation surface and extract reusable parts.

Solution

• Upgraded and exposed Dag<N> as a fully-fledged directed acyclic graph type.
  • The validity of the cached topological order is now tracked by a dirty flag. All modifications to the graph (via DerefMut or Dag::graph_mut) cause the DAG to be marked as dirty.
  • Added Dag::toposort: if the DAG is dirty, computes topological order, caches it, and marks the DAG as clean. If already clean, returns the cached topological order. We now also reuse the previous toposort Vec allocation.
  • Added Dag::get_toposort: can be used to access the topological order with &self, with the stipulation that it returns an Option, and None if the DAG is dirty.
  • Replaced check_graph with Dag::analyze, and made it publicly accessible.
  • Added Dag::remove_redundant_edges, which uses the output of Dag::analyze.
• Renamed CheckGraphResults to DagAnalysis.
  • Added DagAnalysis::check_for_redundant_edges, replacing ScheduleGraph::optionally_check_hierarchy_conflicts.
  • Added DagAnalysis::check_for_cross_dependencies, replacing ScheduleGraph::check_for_cross_dependencies. It now takes two full DagAnalysis for comparison.
  • Added DagAnalysis::check_for_cross_intersection, replacing ScheduleGraph::check_order_but_intersect.
• Added DagGroups to encapsulate the HashMap<SystemSetKey, Vec<SystemKey>> with additional capabilities:
  • DagGroups::flatten and DagGroups::flatten_undirected handle the graph reduction operations previously performed by functions on ScheduleGraph.
• Added ConflictingSystems to encapsulate Vec<(SystemKey, SystemKey, ComponentId)> with additional capabilities and type safety.

See the included migration guide for breaking changes.

Testing

• Ran examples
• Added new tests for Dag, DagAnalysis, and DagGroups functionality.

Future work

• Consider replacing HashSet<SystemKey> with a FixedBitSet-like type for better performance.

[hymm]

Comments on the code are all minor. I'd like to see a run of the schedule_build benchmarks as a sanity check before approving.

I'm not sure how many other changes you have planned, but we should maybe make a pr to update bevy_mod_debug_dump to main when you're done to make sure we haven't broken anything serious for that crate.

[ItsDoot]

After that latest commit (which was making the _no_constraints benches perform badly), here's the benchmarks:

group                                          pr                                     main
-----                                          ------                                 ------
build_schedule/1000_schedule                   1.00   992.2±17.74ms        ? ?/sec    1.04  1028.5±18.70ms        ? ?/sec
build_schedule/1000_schedule_no_constraints    1.05     19.8±0.15ms        ? ?/sec    1.00     19.0±0.12ms        ? ?/sec
build_schedule/100_schedule                    1.00      6.1±0.03ms        ? ?/sec    1.01      6.2±0.04ms        ? ?/sec
build_schedule/100_schedule_no_constraints     1.02   380.9±11.24µs        ? ?/sec    1.00   372.3±14.00µs        ? ?/sec
build_schedule/500_schedule                    1.00    196.8±4.25ms        ? ?/sec    1.03    201.9±4.39ms        ? ?/sec
build_schedule/500_schedule_no_constraints     1.04      5.4±0.15ms        ? ?/sec    1.00      5.2±0.15ms        ? ?/sec
run_empty_schedule/MultiThreaded               1.05      7.0±0.33ns        ? ?/sec    1.00      6.6±0.38ns        ? ?/sec
run_empty_schedule/SingleThreaded              1.41     13.3±0.78ns        ? ?/sec    1.00      9.4±0.74ns        ? ?/sec
schedule/base                                  1.00     20.4±1.73µs        ? ?/sec    1.03     21.0±3.68µs        ? ?/sec

EDIT: Yea, that SingleThreaded bench is just noise; here's another run:

group                                          pr                                     main
-----                                          ------                                 ------
build_schedule/1000_schedule                   1.00   991.8±15.46ms        ? ?/sec    1.04  1028.5±18.70ms        ? ?/sec
build_schedule/1000_schedule_no_constraints    1.05     19.9±0.12ms        ? ?/sec    1.00     19.0±0.12ms        ? ?/sec
build_schedule/100_schedule                    1.00      6.1±0.10ms        ? ?/sec    1.01      6.2±0.04ms        ? ?/sec
build_schedule/100_schedule_no_constraints     1.02   380.2±10.87µs        ? ?/sec    1.00   372.3±14.00µs        ? ?/sec
build_schedule/500_schedule                    1.00    200.3±3.90ms        ? ?/sec    1.01    201.9±4.39ms        ? ?/sec
build_schedule/500_schedule_no_constraints     1.04      5.4±0.17ms        ? ?/sec    1.00      5.2±0.15ms        ? ?/sec
run_empty_schedule/MultiThreaded               1.05      6.9±0.34ns        ? ?/sec    1.00      6.6±0.38ns        ? ?/sec
run_empty_schedule/SingleThreaded              1.00      9.4±0.62ns        ? ?/sec    1.01      9.4±0.74ns        ? ?/sec
schedule/base                                  1.00     20.6±2.32µs        ? ?/sec    1.02     21.0±3.68µs        ? ?/sec

[hymm]

Out of curiousity how much slower was the HashSet?

[ItsDoot]

Out of curiousity how much slower was the HashSet?

Just a tad /s

group                                          pr                                      main
-----                                          -----                                  ------
build_schedule/1000_schedule                   1.06  1094.7±15.67ms        ? ?/sec    1.00  1028.5±18.70ms        ? ?/sec
build_schedule/1000_schedule_no_constraints    2.67     50.7±0.37ms        ? ?/sec    1.00     19.0±0.12ms        ? ?/sec
build_schedule/100_schedule                    1.07      6.6±0.09ms        ? ?/sec    1.00      6.2±0.04ms        ? ?/sec
build_schedule/100_schedule_no_constraints     1.51   562.2±25.96µs        ? ?/sec    1.00   372.3±14.00µs        ? ?/sec
build_schedule/500_schedule                    1.06    214.4±2.44ms        ? ?/sec    1.00    201.9±4.39ms        ? ?/sec
build_schedule/500_schedule_no_constraints     2.27     11.9±0.04ms        ? ?/sec    1.00      5.2±0.15ms        ? ?/sec
run_empty_schedule/MultiThreaded               1.07      7.1±0.32ns        ? ?/sec    1.00      6.6±0.38ns        ? ?/sec
run_empty_schedule/SingleThreaded              1.08     10.2±0.66ns        ? ?/sec    1.00      9.4±0.74ns        ? ?/sec
schedule/base                                  1.00     20.9±3.19µs        ? ?/sec    1.01     21.0±3.68µs        ? ?/sec