57 lines
4.8 KiB
Markdown
57 lines
4.8 KiB
Markdown
# Code review — NEO-93 (E5M2-07)
|
|
|
|
**Date:** 2026-05-27
|
|
**Scope:** Branch `NEO-93-npc-behavior-state-machine-lazy-tick` vs `origin/main` — NPC runtime state machine, lazy tick advance, tests, and docs
|
|
**Base:** `origin/main`
|
|
|
|
**Follow-up:** Code review suggestions addressed (backlog in-scope/AC alignment, AAA test layout, single-call multi-phase test, plan test count, windup→recover write simplification).
|
|
|
|
## Verdict
|
|
|
|
**Approve with nits**
|
|
|
|
## Summary
|
|
|
|
This branch delivers the E5M2-07 server spine: `INpcRuntimeStateStore`, thread-safe in-memory rows, and `NpcRuntimeOperations.AdvanceAll` driving the prototype **`idle` → `aggro` → `telegraph_windup` → `attack_execute` → `recover`** loop from catalog timings and `IThreatStateStore` holders. Kickoff decisions are honored — holder-only aggro sync, no HTTP wire, instant `attack_execute` stub, comment-only telemetry hooks, and dev fixture reset alongside HP + threat. Tests cover the acceptance paths (cooldown, windup, full melee cycle, single-call multi-phase advance, holder clear, archetype spot checks, delta cap) plus store and fixture integration; **22** focused tests passed locally.
|
|
|
|
## Documentation checked
|
|
|
|
| Path | Result |
|
|
|------|--------|
|
|
| `docs/plans/NEO-93-implementation-plan.md` | **Matches** — store, ops, state machine, fixture, DI, tests, and reconciliation section align with the diff. |
|
|
| `docs/plans/E5M2-prototype-backlog.md` (E5M2-07) | **Matches** — AC checkboxes (including full-chain), landed note, and engine-only in-scope bullet align with kickoff and implementation. |
|
|
| `docs/decomposition/modules/E5_M2_NpcAiAndBehaviorProfiles.md` | **Matches** — E5M2-07 landed note and runtime state machine section present. |
|
|
| `docs/decomposition/modules/documentation_and_implementation_alignment.md` | **Matches** — E5.M2 row includes **NEO-93 landed** note and plan/README links. |
|
|
| `docs/decomposition/modules/client_server_authority.md` | **Matches** — server-owned behavior state; no client projection in this slice (NEO-94 deferred). |
|
|
| `server/README.md` | **Matches** — NPC runtime section documents states, lazy advance, holder sync, fixture reset, and NEO-94 deferral. |
|
|
|
|
## Blocking issues
|
|
|
|
None.
|
|
|
|
## Suggestions
|
|
|
|
1. ~~**Align E5M2-07 backlog “In scope” bullet** — Change “called from snapshot GET handler” to **engine + DI only in NEO-93; NEO-94 invokes `AdvanceAll` from `GET /game/world/npc-runtime-snapshot`** so the backlog matches the adopted kickoff decision and the implementation plan.~~ **Done.** E5M2-07 in-scope bullet updated.
|
|
2. ~~**Add full-chain AC to E5M2-07 backlog** — The implementation plan checks “full transition chain reachable”; add the same checkbox to the backlog AC list for traceability with Linear/E5M2-07.~~ **Done.** Fourth AC checkbox added to E5M2-07 backlog.
|
|
3. ~~**AAA layout in new tests** — Several `NpcRuntimeOperationsTests` methods call `TryGet` in **Act** (e.g. `AdvanceAll_ShouldCompleteMeleeWindupDuration_BeforeAttackExecute` performs a mid-Act read and a second `Advance`). Move verification reads to **Assert** per [csharp-style](../../.cursor/rules/csharp-style.md#unit-and-integration-tests-arrange-act-assert). Non-blocking but worth fixing before NEO-94 adds more tests on this file.~~ **Done.** `TryGet` moved to Assert; windup test split into two facts.
|
|
4. ~~**Optional: single-call multi-phase test** — Add one test that advances from aggro entry to second telegraph in **one** `AdvanceAll` call (e.g. `T0` → `T0+7.5`) to lock the inner `while (cursor < windowEnd)` multi-step loop independently of sequential polls.~~ **Done.** `AdvanceAll_ShouldRunFullMeleeCycle_ToSecondTelegraph_InSingleAdvanceCall` added.
|
|
|
|
## Nits
|
|
|
|
- ~~Nit: `TelegraphWindup` case writes `AttackExecute` then immediately overwrites with `Recover`; the intermediate write is never observable — harmless but could be a single `Recover` write until NEO-95 needs a visible execute phase.~~ **Done.** Windup completion writes `Recover` directly with comment for logical `attack_execute`.
|
|
- ~~Nit: Plan reconciliation cites **17** NEO-93-focused tests; local filter run reports **18** (theory counts as two cases) — trivial count drift.~~ **Done.** Plan reconciliation updated to **18**.
|
|
- Nit: `LastAdvancedUtc` is an unlocked property on the store (same pattern as typical session singletons); acceptable for prototype poll-driven advance.
|
|
|
|
## Verification
|
|
|
|
```bash
|
|
# NEO-93-focused tests (passed 2026-05-27)
|
|
dotnet test server/NeonSprawl.Server.Tests/NeonSprawl.Server.Tests.csproj \
|
|
--filter "FullyQualifiedName~NpcRuntime|FullyQualifiedName~InMemoryNpcRuntime|FullyQualifiedName~CombatTargetFixtureApiTests"
|
|
|
|
# Full server suite before merge
|
|
dotnet test server/NeonSprawl.Server.Tests/NeonSprawl.Server.Tests.csproj
|
|
```
|
|
|
|
Manual: no Bruno changes (no HTTP wire). Confirm dev fixture still resets HP + aggro + runtime rows between Bruno combat runs.
|