fix(renderer): drain threaded loads on shutdown

Reviewed-on: #6

coordination/claims/M00-QAR-CAPTURE-SHUTDOWN-001.md

@@ -0,0 +1,78 @@
+# M00-QAR-CAPTURE-SHUTDOWN-001 — Capture shutdown drain
+
+<!-- OPENWC_CLAIM:M00-QAR-CAPTURE-SHUTDOWN-001:sindo-main-codex:2026-07-14 -->
+
+## Ownership
+
+- Target: M00
+- Program: QAR
+- Owner/Agent ID: sindo-main-codex
+- Branch: `work/sindo-main-codex/m00-capture-shutdown`
+- Lease expires UTC: 2026-07-14
+- Integrator: milestone integrator
+
+## Outcome
+
+Eliminate the reproducible capture-tool ObjectDB leak by completing owned threaded resource requests and draining deferred scene deletion before process exit.
+
+## Non-goals
+
+- Refactoring renderer resource ownership.
+- Hiding genuine RID or worker-task leaks.
+- Changing capture images, metrics, or streaming behavior.
+
+## Paths
+
+- Exclusive: capture tool shutdown sequence
+- Shared/hotspots: renderer baseline documentation
+- Generated/ignored: verbose Godot logs
+
+## Contracts and data
+
+- Public API/events: none
+- Schema/format version: unchanged
+- Migration/compatibility: none
+- Consumers: M00 baseline runner
+
+## Dependencies
+
+- Requires: reproducible verbose leak diagnostic
+- Blocks: clean M00 capture shutdown evidence
+- External state: none
+
+## Verification
+
+- Commands: identical verbose dry-run before/after, M00 dry-run, repository gates
+- Fixtures: ADT-boundary filtered capture
+- Fidelity evidence: no visual behavior change
+- Performance budget: blocking resource completion is restricted to shutdown; two SceneTree drain frames only
+
+## Documentation deliverables
+
+- Inline public API docs: shutdown rationale comment
+- Module specification: recovery/known-risk update
+- Data-flow diagram: unchanged
+- Sequence/state/dependency diagrams: shutdown sequence documented in baseline
+- Source map/status updates: baseline evidence
+
+## Simplicity and naming
+
+- Important names introduced: none
+- Simplest considered solution: await two SceneTree frames after `queue_free` and finish registered in-flight ResourceLoader requests during loader shutdown
+- Rejected complexity/abstractions: resource registry or manual child traversal
+- Unavoidable complexity and justification: deferred deletion requires frame drain
+- Measured optimization evidence: not applicable
+
+## Status
+
+- State: ready-for-review
+- Done: one leaked RefCounted reproduced; isolation attributed it to `StreamingWorldLoader`; active threaded tile request was confirmed as the retained object; all owned threaded resource registries now finish in-progress requests before clearing; two identical verbose ADT-boundary runs and the full M00 dry-run completed without ObjectDB leak diagnostics
+- Next: integrator review and merge
+- Blocked by:
+
+## Handoff
+
+- Commit: this work-package commit
+- Results: verbose filtered capture returned one report and zero `Leaked instance`/`ObjectDB instances leaked` lines on two consecutive runs; `run_render_baseline.ps1 -DryRun` passed all gates and seven checkpoints
+- Remaining risks:
+- Documentation updated: `docs/modules/world-renderer.md` shutdown sequence, ownership contract, recovery table and known-risk status

coordination/claims/M00-QAR-FOV-SWEEP-001.md

@@ -0,0 +1,78 @@
+# M00-QAR-FOV-SWEEP-001 — Empirical camera FOV sweep
+
+<!-- OPENWC_CLAIM:M00-QAR-FOV-SWEEP-001:sindo-main-codex:2026-07-14 -->
+
+## Ownership
+
+- Target: M00
+- Program: QAR
+- Owner/Agent ID: sindo-main-codex
+- Branch: `work/sindo-main-codex/m00-fov-sweep`
+- Lease expires UTC: 2026-07-14
+- Integrator: milestone integrator
+
+## Outcome
+
+Support a bounded empirical FOV sweep when build 12340 does not expose a readable camera FOV CVar.
+
+## Non-goals
+
+- Claiming exact WoW FOV from a single manually aimed screenshot.
+- Changing runtime player camera defaults.
+- Automating yaw/pitch registration or image warping.
+
+## Paths
+
+- Exclusive: FOV/filter CLI additions in checkpoint tools
+- Shared/hotspots: renderer baseline documentation
+- Generated/ignored: local sweep PNGs and JSON reports
+
+## Contracts and data
+
+- Public API/events: additive `--camera-fov` and comparator `--only` CLI options
+- Schema/format version: unchanged
+- Migration/compatibility: existing commands retain defaults
+- Consumers: M00 fidelity workflow
+
+## Dependencies
+
+- Requires: build 12340 reference JPG and paired-image comparator
+- Blocks: empirical projection ranking
+- External state: original screenshots remain outside Git
+
+## Verification
+
+- Commands: synthetic comparator test, capture dry-run with override, bounded real sweep
+- Fixtures: `elwynn_adt_boundary` reference
+- Fidelity evidence: ranked perceptual metrics with manual-direction limitation
+- Performance budget: offline diagnostic
+
+## Documentation deliverables
+
+- Inline public API docs: CLI headers
+- Module specification: verification/source map if needed
+- Data-flow diagram: FOV sweep flow
+- Sequence/state/dependency diagrams: not applicable
+- Source map/status updates: baseline findings
+
+## Simplicity and naming
+
+- Important names introduced: `camera_fov_override`, `only_filter`
+- Simplest considered solution: reuse capture and comparator
+- Rejected complexity/abstractions: computer vision registration framework
+- Unavoidable complexity and justification: multiple captures are required because client FOV is inaccessible
+- Measured optimization evidence: not applicable
+
+## Status
+
+- State: ready
+- Done: client FOV access audit, capture override, comparator filter/regression, dedicated-camera ownership fix and bounded sweep
+- Next: integrator review; obtain reproducible yaw/pitch/zoom metadata before changing normative FOV
+- Blocked by:
+
+## Handoff
+
+- Commit: branch HEAD
+- Results: corrected ranking 26=0.079588, 38=0.079633, 50=0.084353, 62=0.088360, 86=0.097993; plateau is inconclusive
+- Remaining risks: build 12340 FOV is inaccessible through tested APIs; manual direction/framing dominates metrics
+- Documentation updated: `docs/RENDER_BASELINE.md`, `docs/modules/world-renderer.md`

docs/RENDER_BASELINE.md

@@ -179,3 +179,18 @@ flowchart LR
 ```
 
 Ни одна из пяти камер не находится внутри опубликованной scene-tree geometry. Terrain-overview segment пересекает четыре Stormwind WMO groups, large-WMO segment — три Goldshire Inn groups, waterfall segment — liquid surface; ADT boundary и dense-M2 segments не пересекают placement AABB. Поэтому прежнее визуальное впечатление «камера внутри WMO/placements» не подтверждается. Основной paired gap сейчас — неточно воспроизведённые manual look direction/target/FOV reference-кадров: например, автоматический Goldshire target направляет луч через фасад внутрь WMO, тогда как reference был вручную скадрирован на весь фасад. Probe охватывает только scene-tree MeshInstance/MultiMesh; RID-only instances не имеют доступного semantic path и явно исключены из coverage.
+
+## Empirical FOV sweep
+
+В build 12340 `GetCVar("cameraFoV")` возвращает `nil`, а `/console cameraFoV` и `ConsoleExec("cameraFoV")` не выдают значения. Поэтому capture tool поддерживает additive `--camera-fov`, а comparator — `--only`, позволяющий ограничить sweep одним reference checkpoint.
+
+```mermaid
+flowchart LR
+    F[Candidate vertical FOV] --> C[Dedicated checkpoint camera]
+    C --> P[Filtered checkpoint PNGs]
+    R[One original-client reference] --> D[Perceptual comparator --only]
+    P --> D
+    D --> M[Ranked metrics]
+```
+
+Первый sweep обнаружил capture defect: разные FOV иногда создавали одинаковые hashes, потому что scene camera могла перехватить viewport после входа в tree. Capture теперь вызывает `camera.make_current()` после добавления world и перед каждым checkpoint. После исправления ADT-boundary ranking стал: `26° → 0.079588`, `38° → 0.079633`, `50° → 0.084353`, `62° → 0.088360`, `86° → 0.097993` mean error. Plateau `26–38°` и несовпавший changed-pixel optimum показывают доминирование manual look direction/framing; эти данные не обосновывают изменение нормативного manifest FOV `62°`. Для настоящей калибровки reference capture должен сохранять воспроизводимые yaw/pitch/zoom или независимый projection fixture.

docs/modules/world-renderer.md

@@ -141,6 +141,9 @@ sequenceDiagram
     Stream->>Budget: enqueue finalize operations
     Budget->>Render: attach bounded terrain/M2/WMO/liquid work
     Stream->>Render: evict outside retention range
+    Stream->>Worker: shutdown: wait for WorkerThreadPool tasks
+    Stream->>Stream: shutdown: finish registered ResourceLoader requests
+    Stream->>Render: clear queues, nodes, caches and RIDs
 ```
 
 ## Ownership, threading and resources
@@ -149,7 +152,7 @@ sequenceDiagram
 - Worker tasks не должны менять SceneTree и shared Resource concurrently.
 - Parsed/grouped results передаются обратно через guarded result queues.
 - Mesh/material/node/RID finalization выполняется main thread и ограничивается exported budgets.
-- Shutdown/map switch обязан отменить/дождаться jobs и освободить RIDs; M00 всё ещё фиксирует leaked-resource risk.
+- Shutdown/map switch обязан дождаться WorkerThreadPool jobs и зарегистрированных threaded ResourceLoader requests до очистки очередей, nodes, caches и RIDs.
 - Cache resources считаются immutable после публикации.
 
 ## Errors, cancellation and recovery
@@ -162,7 +165,7 @@ sequenceDiagram
 | Main-thread hitch | Named section timing | Frame spike, work remains queued | `HITCH` log | Lower budget/fix finalize path |
 | D3D12 descriptor exhaustion | Rendering backend error | Render failure/fallback backend | Godot error + baseline report | Dedup resources/fix settings |
 | Teleport/map change | Focus/session transition | Old jobs become stale | Target/session generation | Cancel/drop stale results |
-| Shutdown leak | Godot leak/RID diagnostics | Resource retained | Shutdown report | Ownership cleanup before DONE |
+| Shutdown leak | Godot leak/RID diagnostics | Resource retained | Verbose shutdown report | Drain owned worker and resource requests before clearing registries |
 
 ## Configuration and capabilities
 
@@ -225,7 +228,8 @@ Exact exported settings and cache versions remain documented in [`../../RENDER.m
 - Первый paired run выявил coordinate/placement mismatch: несколько совпадающих server-derived camera positions оказываются под terrain или внутри WMO/rocks OpenWC.
 - Terrain-height probe исключил under-terrain состояние для всех пяти точек; waterfall exact-XZ miss классифицирован как TriangleMesh seam/edge и подтверждён nearby sample в 2 units.
 - Camera-occluder probe не нашёл camera containment в пяти точках; paired mismatch локализован прежде всего в manual look direction/target/FOV calibration, с явным ограничением по RID-only geometry.
-- D3D12 descriptor and shutdown RID/resource issues remain.
+- Empirical FOV sweep выявил, что checkpoint camera должна явно вызывать `make_current()`; после исправления projection ranking остаётся inconclusive из-за неизвестного manual yaw/pitch/framing reference.
+- D3D12 descriptor issues remain; the capture-path anonymous `RefCounted` shutdown leak is regression-covered by a clean verbose dry-run, while other RID/resource diagnostics still require independent evidence.
 - M2/WMO/material/particle/ribbon/portal parity incomplete.
 - Public API is mostly exported configuration rather than stable contracts.
 

src/scenes/streaming/streaming_world_loader.gd

@@ -2073,7 +2073,7 @@ func _wait_for_tile_tasks() -> void:
 			var path: String = String(pending.get("path", ""))
 			if not path.is_empty():
 				var status := ResourceLoader.load_threaded_get_status(path)
-				if status == ResourceLoader.THREAD_LOAD_LOADED:
+				if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 					ResourceLoader.load_threaded_get(path)
 	_tile_loading_tasks.clear()
 
@@ -2094,7 +2094,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_m2_mesh_load_requests.clear()
 	_m2_mesh_finalize_queue.clear()
@@ -2105,7 +2105,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_m2_animation_load_requests.clear()
 	_m2_animation_finalize_queue.clear()
@@ -2115,7 +2115,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_wmo_scene_load_requests.clear()
 
@@ -2124,7 +2124,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_wmo_render_load_requests.clear()
 	_wmo_render_missing_cache.clear()
@@ -2137,7 +2137,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_terrain_upgrade_tasks.clear()
 
@@ -2148,7 +2148,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_terrain_control_splat_cache_tasks.clear()
 
@@ -2162,7 +2162,7 @@ func _wait_for_tile_tasks() -> void:
 		if path.is_empty():
 			continue
 		var status := ResourceLoader.load_threaded_get_status(path)
-		if status == ResourceLoader.THREAD_LOAD_LOADED:
+		if status == ResourceLoader.THREAD_LOAD_IN_PROGRESS or status == ResourceLoader.THREAD_LOAD_LOADED:
 			ResourceLoader.load_threaded_get(path)
 	_terrain_splat_cache_tasks.clear()
 	_terrain_splat_result_mutex.lock()

src/tools/capture_render_checkpoints.gd

@@ -2,6 +2,7 @@ extends SceneTree
 
 const DEFAULT_MANIFEST_PATH := "res://src/tools/render_baseline_manifest.json"
 const M2_NATIVE_ANIMATED_BUILDER := preload("res://addons/mpq_extractor/loaders/m2_native_animated_builder.gd")
+const SHUTDOWN_DRAIN_FRAMES := 2
 
 
 func _initialize() -> void:
@@ -22,6 +23,7 @@ func _capture_async() -> void:
 	var measure_seconds := float(_arg(args, "--measure", str(manifest.get("default_measure_seconds", 3.0))))
 	var revision := _arg(args, "--revision", "worktree")
 	var cache_state := _arg(args, "--cache-state", "existing")
+	var camera_fov_override := float(_arg(args, "--camera-fov", str(manifest.get("camera_fov", 62.0))))
 	var headless := DisplayServer.get_name().to_lower() == "headless"
 	var dry_run := args.has("--dry-run") or headless
 	var viewport: Array = manifest.get("viewport", [1280, 900])
@@ -55,7 +57,7 @@ func _capture_async() -> void:
 	var camera := Camera3D.new()
 	camera.name = "CheckpointCamera"
 	camera.current = true
-	camera.fov = float(manifest.get("camera_fov", 62.0))
+	camera.fov = camera_fov_override
 	camera.far = 50000.0
 	camera.position = _vector3(first.get("camera", [0.0, 0.0, 0.0]))
 	(world as Node3D).add_child(camera)
@@ -65,6 +67,7 @@ func _capture_async() -> void:
 	get_root().add_child(world)
 	await process_frame
 	await process_frame
+	camera.make_current()
 
 	var player := world.get_node_or_null("ThirdPersonPlayer") as Node3D
 	if player != null:
@@ -82,6 +85,7 @@ func _capture_async() -> void:
 		"created_utc": Time.get_datetime_string_from_system(true, true),
 		"dry_run": dry_run,
 		"cache_state": cache_state,
+		"camera_fov": camera_fov_override,
 		"environment": _environment_metadata(),
 		"comparison_budgets": manifest.get("comparison_budgets", {}),
 		"cache_contract": manifest.get("cache_contract", {}),
@@ -110,6 +114,7 @@ func _capture_async() -> void:
 
 			camera.global_position = _vector3(checkpoint.get("camera", [0.0, 0.0, 0.0]))
 			_orient_camera_without_roll(camera, _vector3(checkpoint.get("target", [0.0, 0.0, 0.0])))
+			camera.make_current()
 			if player != null:
 				player.global_position = _vector3(checkpoint.get("player", checkpoint.get("target", [0.0, 0.0, 0.0])))
 			_set_sky_time(world, float(checkpoint.get("time_hours", 13.0)))
@@ -163,6 +168,10 @@ func _capture_async() -> void:
 	print("RENDER_BASELINE report=%s results=%d" % [report_path, captured])
 
 	world.queue_free()
+	# SceneTree deletion is deferred. Allow the world and its queued children
+	# to run their shutdown paths before engine teardown.
+	for unused_frame in SHUTDOWN_DRAIN_FRAMES:
+		await process_frame
 	if captured <= 0:
 		push_error("No checkpoints selected. --only filter was: %s" % only)
 		quit(1)

src/tools/compare_render_checkpoints.gd

@@ -56,7 +56,7 @@ func _parse_arguments(raw_arguments: PackedStringArray) -> Dictionary:
 			parsed.self_test = true
 			index += 1
 			continue
-		if argument not in ["--reference", "--candidate", "--output", "--pixel-threshold", "--mean-threshold", "--changed-ratio-threshold"]:
+		if argument not in ["--reference", "--candidate", "--output", "--only", "--pixel-threshold", "--mean-threshold", "--changed-ratio-threshold"]:
 			return {"error": "unknown argument: %s" % argument}
 		if index + 1 >= raw_arguments.size():
 			return {"error": "missing value for %s" % argument}
@@ -76,10 +76,13 @@ func _compare_directories(reference_directory: String, candidate_directory: Stri
 	reference_directory = ProjectSettings.globalize_path(reference_directory)
 	candidate_directory = ProjectSettings.globalize_path(candidate_directory)
 	var reference_files := _reference_image_file_names(reference_directory)
+	var only_filter := String(options.get("only", "")).to_lower()
 	var results: Array[Dictionary] = []
 	var failed_count := 0
 	var missing_count := 0
 	for file_name in reference_files:
+		if not only_filter.is_empty() and not file_name.get_basename().to_lower().contains(only_filter):
+			continue
 		var reference_path := reference_directory.path_join(file_name)
 		var checkpoint_name := file_name.get_basename()
 		var candidate_file_names := _candidate_file_names(candidate_directory, checkpoint_name)
@@ -203,6 +206,7 @@ func _run_self_test() -> int:
 	var changed_directory := root.path_join("changed")
 	for directory_path in [reference_directory, identical_directory, changed_directory]:
 		DirAccess.make_dir_recursive_absolute(ProjectSettings.globalize_path(directory_path))
+		_clear_directory_files(directory_path)
 	var reference_image := Image.create(2, 2, false, Image.FORMAT_RGBA8)
 	reference_image.fill(Color(0.25, 0.5, 0.75, 1.0))
 	var changed_image := reference_image.duplicate()
@@ -225,5 +229,21 @@ func _run_self_test() -> int:
 	if not identical_report.passed or changed_report.passed:
 		push_error("RENDER_CHECKPOINT_DIFF SELF_TEST: expected identical pass and changed failure")
 		return 1
+	if reference_image.save_jpg(reference_directory.path_join("ignored.jpg"), 1.0) != OK:
+		return 1
+	var filtered_options := options.duplicate()
+	filtered_options["only"] = "synthetic"
+	var filtered_report := _compare_directories(reference_directory, identical_directory, filtered_options)
+	if not filtered_report.passed or filtered_report.compared_count != 2:
+		push_error("RENDER_CHECKPOINT_DIFF SELF_TEST: --only must exclude unrelated references")
+		return 1
 	print("RENDER_CHECKPOINT_DIFF SELF_TEST PASS")
 	return 0
+
+
+func _clear_directory_files(directory_path: String) -> void:
+	var directory := DirAccess.open(directory_path)
+	if directory == null:
+		return
+	for file_name in directory.get_files():
+		directory.remove(file_name)