Add Phase Z Layer A placement planner
- add dormant placement planner integrating B1 / B2 / B3 - region 1:1 sub_zone mapping with narrowest-accepts-first heuristic - frame selection by accepted_content_types coverage + declaration order
This commit is contained in:
387
src/phase_z2_placement_planner.py
Normal file
387
src/phase_z2_placement_planner.py
Normal file
@@ -0,0 +1,387 @@
|
||||
"""Phase Z-2 Placement planner (B4 v0 — dormant module).
|
||||
|
||||
SPEC v1 §4 의 2-stage placement (Layer A → Layer B) 통합 module.
|
||||
|
||||
v0 minimal :
|
||||
- 지원 case : text_block only / text_block + transform_table 2 가지 (B1/B2 정합)
|
||||
- dormant — runtime path 미연결 (pipeline / composition / mapper / partial / yaml 미터치)
|
||||
- Stage A = B2 plan_internal_regions() *호출만* (logic 중복 X)
|
||||
- frame 선택 = accepted_content_types cover + frame_contracts 입력 순서 first
|
||||
- Stage B 매핑 단위 = region 1:1 sub_zone (단순화)
|
||||
- sub_zone 선택 = narrowest accepts first + declaration order tie-break (deadlock 방지)
|
||||
- cardinality.strict 초과 → rejection / under-fill → 허용
|
||||
- display_strategy = 모두 inline_full / overflow_buffer = []
|
||||
- partial_target_path = sub_zone 으로부터 *읽어서 보존만* (실제 marker 미적용)
|
||||
- F13 multi-pillar distribution 미지원 (1 ContentObject per region 만)
|
||||
- V4 rank / multi-frame ranking / display_only path 미활성
|
||||
|
||||
책임 boundary :
|
||||
- B4 = Stage A wrapping + frame 선택 + Stage B (region 1:1 sub_zone) 매핑
|
||||
- 별 axis = display_only path / preview·details 활성 / partial template marker /
|
||||
telemetry 연동 / runtime pipeline 연결
|
||||
|
||||
검증 :
|
||||
- dormancy : MDX 03 final.html SHA = canonical 유지 (runtime path 미연결)
|
||||
- correctness : __main__ self-test (text-only 1 case + text+transform 1 case)
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any, Optional
|
||||
|
||||
# B4 v0 input contract = B1 (ContentObject) + B2 (InternalRegion / ZoneRegionPlan).
|
||||
# 세 module 모두 dormant — runtime path 와 무관한 layer-agnostic 의존.
|
||||
from phase_z2_content_extractor import ContentObject
|
||||
from phase_z2_internal_region_planner import (
|
||||
InternalRegion,
|
||||
ZoneRegionPlan,
|
||||
plan_internal_regions,
|
||||
)
|
||||
|
||||
|
||||
# ─── Output schema (SPEC v1 §4.1) ────────────────────────────────
|
||||
|
||||
|
||||
@dataclass
|
||||
class SlotAssignment:
|
||||
"""SPEC v1 §4.3 Stage B 의 slot_assignment.
|
||||
|
||||
region 의 content_unit 이 frame 의 어느 Frame Slot 으로 가는지 명시.
|
||||
partial_target_path = B3 catalog 의 sub_zone.partial_target_path 그대로 보존
|
||||
(실제 marker 적용은 별 axis B5 영역).
|
||||
"""
|
||||
|
||||
region_id: str
|
||||
content_unit_id: str
|
||||
frame_slot_id: str
|
||||
partial_target_path: str
|
||||
display_strategy: str # v0 = "inline_full" 만
|
||||
|
||||
|
||||
@dataclass
|
||||
class PlacementPlan:
|
||||
"""B4 v0 의 출력 — section 의 *전체 placement* 결과 (Stage A + Stage B 통합).
|
||||
|
||||
Fields :
|
||||
section_id : section 식별자 (region_id prefix 와 일관)
|
||||
selected_frame_id : frame.frame_id (or None — frame cover 실패 시)
|
||||
selected_template_id : frame.template_id (or None)
|
||||
internal_regions : Stage A 결과 (B2 planner 출력 그대로)
|
||||
slot_assignments : Stage B 결과 (region.content_unit → Frame Slot)
|
||||
overflow_buffer : v0 = 빈 list (preview/details path 미활성)
|
||||
rejection : 매칭 안 된 / cardinality 초과 등 — 자동 렌더 X 신호
|
||||
"""
|
||||
|
||||
section_id: str
|
||||
selected_frame_id: Optional[str] = None
|
||||
selected_template_id: Optional[str] = None
|
||||
internal_regions: list[InternalRegion] = field(default_factory=list)
|
||||
slot_assignments: list[SlotAssignment] = field(default_factory=list)
|
||||
overflow_buffer: list[dict] = field(default_factory=list)
|
||||
rejection: list[dict] = field(default_factory=list)
|
||||
|
||||
|
||||
# ─── Frame selection ─────────────────────────────────────────────
|
||||
|
||||
|
||||
def _select_frame(
|
||||
content_objects: list[ContentObject],
|
||||
frame_contracts: list[dict[str, Any]],
|
||||
) -> Optional[dict[str, Any]]:
|
||||
"""frame_contracts 중 *content_type_set 을 모두 cover* 하는 첫 frame.
|
||||
|
||||
rule (B4 v0 lock) :
|
||||
1. content_type_set = {obj.type for obj in content_objects}
|
||||
2. frame_contract.accepted_content_types ⊇ content_type_set 인 후보 모음
|
||||
3. frame_contracts 입력 순서 (= YAML declaration order) 첫 entry 선택
|
||||
|
||||
Returns :
|
||||
frame_contract dict 또는 None (cover 가능 frame 없음)
|
||||
"""
|
||||
content_type_set = {obj.type for obj in content_objects}
|
||||
for fc in frame_contracts:
|
||||
accepted = set(fc.get("accepted_content_types") or [])
|
||||
if content_type_set <= accepted: # ⊇ check
|
||||
return fc
|
||||
return None
|
||||
|
||||
|
||||
# ─── Sub_zone assignment (Stage B) ───────────────────────────────
|
||||
|
||||
|
||||
def _assign_region_to_sub_zone(
|
||||
region: InternalRegion,
|
||||
frame_sub_zones: list[dict[str, Any]],
|
||||
assigned_sub_zone_ids: set[str],
|
||||
) -> Optional[dict[str, Any]]:
|
||||
"""region 에 매칭할 sub_zone 선택 (B4 v0 narrowest-first heuristic).
|
||||
|
||||
rule (B4 v0 lock — F29 deadlock 방지) :
|
||||
1. not-yet-assigned 중 region.content_type 을 accepts 하는 후보 수집
|
||||
2. 후보 중 accepts list 가장 *좁은* sub_zone 우선
|
||||
3. 동률이면 declaration order (Python sort 의 stability 활용)
|
||||
|
||||
예 (F29) :
|
||||
region.content_type = text_block
|
||||
candidates = [process_column(accepts=[text,transform], size 2),
|
||||
product_column(accepts=[text], size 1)]
|
||||
→ product_column 선택 (narrowest)
|
||||
|
||||
region.content_type = transform_table (이후 호출, product_column 이미 assigned)
|
||||
candidates = [process_column] 만
|
||||
→ process_column 선택
|
||||
|
||||
Returns :
|
||||
sub_zone dict 또는 None (compatible 후보 없음)
|
||||
"""
|
||||
candidates: list[dict[str, Any]] = []
|
||||
for sz in frame_sub_zones:
|
||||
if sz["id"] in assigned_sub_zone_ids:
|
||||
continue
|
||||
accepts = sz.get("accepts") or []
|
||||
if region.content_type in accepts:
|
||||
candidates.append(sz)
|
||||
|
||||
if not candidates:
|
||||
return None
|
||||
|
||||
# narrowest first — accepts size 작을수록 우선. Python sort stable → 동률은 declaration order 보존.
|
||||
candidates.sort(key=lambda sz: len(sz.get("accepts") or []))
|
||||
return candidates[0]
|
||||
|
||||
|
||||
# ─── Public entry ────────────────────────────────────────────────
|
||||
|
||||
|
||||
def plan_placement(
|
||||
content_objects: list[ContentObject],
|
||||
frame_contracts: list[dict[str, Any]],
|
||||
section_id: str = "",
|
||||
) -> PlacementPlan:
|
||||
"""ContentObject[] + frame_contracts → PlacementPlan (Stage A + Stage B 통합).
|
||||
|
||||
v0 algorithm :
|
||||
1. Stage A = B2 plan_internal_regions() 호출 → internal_regions 획득
|
||||
2. frame 선택 : accepted_content_types cover + 입력 순서 first
|
||||
- cover 실패 시 → rejection + early return
|
||||
3. selected_frame 의 sub_zones 읽음 (B3 catalog)
|
||||
4. Stage B (region 1:1 sub_zone 매핑) :
|
||||
- 각 region 마다 narrowest-accepts first + declaration order sub_zone 선택
|
||||
- region.content_unit_ids 를 sub_zone.cardinality 와 비교
|
||||
- count > strict → rejection 추가 / SlotAssignment 미생성
|
||||
- count ≤ strict → SlotAssignment 생성 (under-fill 허용)
|
||||
- 매칭 sub_zone 없는 region → rejection 추가
|
||||
5. display_strategy = inline_full 모두 / overflow_buffer = [] (v0)
|
||||
|
||||
Args :
|
||||
content_objects : list[ContentObject] — B1 v0 extractor 출력
|
||||
frame_contracts : list[dict] — frame_contracts.yaml 의 contract dict list
|
||||
(YAML declaration order = list 순서로 입력 권고)
|
||||
section_id : region_id / 결과 식별자 prefix
|
||||
|
||||
Returns :
|
||||
PlacementPlan
|
||||
"""
|
||||
plan = PlacementPlan(section_id=section_id)
|
||||
|
||||
if not content_objects:
|
||||
return plan
|
||||
|
||||
# 1. Stage A — B2 호출 (logic 중복 X)
|
||||
zone_plan: ZoneRegionPlan = plan_internal_regions(
|
||||
content_objects=content_objects,
|
||||
frame_contracts=frame_contracts,
|
||||
section_id=section_id,
|
||||
)
|
||||
plan.internal_regions = list(zone_plan.internal_regions)
|
||||
|
||||
# 2. frame 선택
|
||||
selected_frame = _select_frame(content_objects, frame_contracts)
|
||||
if selected_frame is None:
|
||||
plan.rejection.append({
|
||||
"reason": "no_frame_covers_content_types",
|
||||
"content_types": sorted({o.type for o in content_objects}),
|
||||
})
|
||||
return plan
|
||||
|
||||
plan.selected_template_id = selected_frame.get("template_id")
|
||||
fid = selected_frame.get("frame_id")
|
||||
plan.selected_frame_id = str(fid) if fid is not None else None
|
||||
|
||||
# 3. selected_frame 의 sub_zones (B3 catalog 의 Frame Slot 선언)
|
||||
sub_zones = list(selected_frame.get("sub_zones") or [])
|
||||
|
||||
# 4. Stage B — region 1:1 sub_zone 매핑
|
||||
assigned_sub_zone_ids: set[str] = set()
|
||||
|
||||
for region in plan.internal_regions:
|
||||
sub_zone = _assign_region_to_sub_zone(region, sub_zones, assigned_sub_zone_ids)
|
||||
if sub_zone is None:
|
||||
plan.rejection.append({
|
||||
"reason": "no_compatible_sub_zone",
|
||||
"region_id": region.region_id,
|
||||
"region_content_type": region.content_type,
|
||||
})
|
||||
continue
|
||||
assigned_sub_zone_ids.add(sub_zone["id"])
|
||||
|
||||
# cardinality 검사 (v0 = strict only)
|
||||
cardinality = sub_zone.get("cardinality") or {}
|
||||
strict = cardinality.get("strict")
|
||||
unit_count = len(region.content_unit_ids)
|
||||
if strict is not None and unit_count > strict:
|
||||
plan.rejection.append({
|
||||
"reason": "cardinality_strict_exceeded",
|
||||
"region_id": region.region_id,
|
||||
"frame_slot_id": sub_zone["id"],
|
||||
"cardinality_strict": strict,
|
||||
"unit_count": unit_count,
|
||||
})
|
||||
continue
|
||||
|
||||
# SlotAssignment 생성 (under-fill 허용 — strict 보다 적어도 OK)
|
||||
partial_path = sub_zone.get("partial_target_path") or ""
|
||||
for content_unit_id in region.content_unit_ids:
|
||||
plan.slot_assignments.append(
|
||||
SlotAssignment(
|
||||
region_id=region.region_id,
|
||||
content_unit_id=content_unit_id,
|
||||
frame_slot_id=sub_zone["id"],
|
||||
partial_target_path=partial_path,
|
||||
display_strategy="inline_full", # v0 default
|
||||
)
|
||||
)
|
||||
|
||||
# 5. v0 = overflow_buffer 미활성 (빈 list 그대로)
|
||||
return plan
|
||||
|
||||
|
||||
# ─── Self-test (B4 v0 correctness 검증) ─────────────────────────
|
||||
|
||||
|
||||
def _run_self_test():
|
||||
"""v0 unit test : Test 1 (text-only → F13) + Test 2 (text+transform → F29).
|
||||
|
||||
fixed input + 실제 frame_contracts.yaml 로드해서 검증.
|
||||
YAML declaration order = F13 / F29 / F16.
|
||||
"""
|
||||
import yaml
|
||||
from pathlib import Path
|
||||
|
||||
PROJECT_ROOT = Path(__file__).parent.parent
|
||||
catalog_path = PROJECT_ROOT / "templates" / "phase_z2" / "catalog" / "frame_contracts.yaml"
|
||||
catalog = yaml.safe_load(catalog_path.read_text(encoding="utf-8"))
|
||||
# YAML 의 top-level dict — Python 3.7+ insertion-order 보존. declaration order = F13/F29/F16.
|
||||
frame_contracts = list(catalog.values())
|
||||
|
||||
# ─── Test 1 : 1 text_block → F13 → pillar_1 ─────────────────
|
||||
text_obj = ContentObject(
|
||||
id="t1.text-1",
|
||||
type="text_block",
|
||||
role="summary",
|
||||
raw_payload="* 본문",
|
||||
size_estimate={"line_count": 6},
|
||||
type_specific={
|
||||
"format": "bullet_list", "bullet_count": 1,
|
||||
"max_indent_level": 0, "has_emphasis": False,
|
||||
},
|
||||
)
|
||||
plan1 = plan_placement([text_obj], frame_contracts, section_id="t1")
|
||||
|
||||
# frame 선택 = F13 (declaration order first, content_type_set={text_block} cover)
|
||||
assert plan1.selected_template_id == "three_parallel_requirements", \
|
||||
f"Test 1 frame=F13 기대, got {plan1.selected_template_id}"
|
||||
assert plan1.selected_frame_id == "1171281190", \
|
||||
f"Test 1 frame_id=1171281190 기대, got {plan1.selected_frame_id}"
|
||||
|
||||
# 1 region (region-single)
|
||||
assert len(plan1.internal_regions) == 1, \
|
||||
f"Test 1 1 region 기대, got {len(plan1.internal_regions)}"
|
||||
|
||||
# 1 SlotAssignment — region 1 → pillar_1 (declaration order tie-break, 모두 size=1)
|
||||
assert len(plan1.slot_assignments) == 1, \
|
||||
f"Test 1 slot_assignments=1 기대, got {len(plan1.slot_assignments)}"
|
||||
sa = plan1.slot_assignments[0]
|
||||
assert sa.frame_slot_id == "pillar_1", \
|
||||
f"Test 1 sub_zone=pillar_1 기대, got {sa.frame_slot_id}"
|
||||
assert sa.region_id == "t1.region-1"
|
||||
assert sa.content_unit_id == "t1.text-1"
|
||||
assert sa.display_strategy == "inline_full"
|
||||
assert "f13b__col" in sa.partial_target_path, \
|
||||
f"Test 1 partial_target_path 보존 기대, got {sa.partial_target_path}"
|
||||
|
||||
# under-fill 허용 — pillar_2 / pillar_3 미할당, rejection 0
|
||||
assert len(plan1.rejection) == 0, f"Test 1 rejection=0 기대, got {plan1.rejection}"
|
||||
assert plan1.overflow_buffer == []
|
||||
print("[OK] Test 1 (text-only → F13 → pillar_1) passed.")
|
||||
|
||||
# ─── Test 2 : 1 text + 1 transform → F29 → product_column / process_column ─
|
||||
text_obj2 = ContentObject(
|
||||
id="t2.text-1",
|
||||
type="text_block",
|
||||
role="summary",
|
||||
raw_payload="* 본문",
|
||||
size_estimate={"line_count": 6},
|
||||
type_specific={
|
||||
"format": "bullet_list", "bullet_count": 1,
|
||||
"max_indent_level": 0, "has_emphasis": False,
|
||||
},
|
||||
)
|
||||
transform_obj = ContentObject(
|
||||
id="t2.transform-1",
|
||||
type="transform_table",
|
||||
role="summary",
|
||||
raw_payload="| AS-IS | ➜ | TO-BE |\n|---|---|---|\n| a | ➜ | b |",
|
||||
size_estimate={"rows": 1},
|
||||
type_specific={
|
||||
"pair_count": 1, "arrow_glyph": "➜",
|
||||
"rows": [{"from": "a", "arrow": "➜", "to": "b"}],
|
||||
},
|
||||
)
|
||||
plan2 = plan_placement([text_obj2, transform_obj], frame_contracts, section_id="t2")
|
||||
|
||||
# frame 선택 = F29 (transform_table 수용 유일)
|
||||
assert plan2.selected_template_id == "process_product_two_way", \
|
||||
f"Test 2 frame=F29 기대, got {plan2.selected_template_id}"
|
||||
|
||||
# 2 regions (text=primary / transform=supporting)
|
||||
assert len(plan2.internal_regions) == 2, \
|
||||
f"Test 2 2 regions 기대, got {len(plan2.internal_regions)}"
|
||||
|
||||
# 2 SlotAssignments
|
||||
assert len(plan2.slot_assignments) == 2, \
|
||||
f"Test 2 slot_assignments=2 기대, got {len(plan2.slot_assignments)}"
|
||||
|
||||
# narrowest-first 검증 (F29 deadlock 방지 핵심) :
|
||||
# text region → product_column (accepts=[text], size 1, narrowest)
|
||||
# transform region → process_column (accepts=[text+transform], size 2, 남은 candidate)
|
||||
text_sa = next(
|
||||
(sa for sa in plan2.slot_assignments if sa.content_unit_id == "t2.text-1"),
|
||||
None,
|
||||
)
|
||||
transform_sa = next(
|
||||
(sa for sa in plan2.slot_assignments if sa.content_unit_id == "t2.transform-1"),
|
||||
None,
|
||||
)
|
||||
|
||||
assert text_sa is not None, "text content_unit SlotAssignment 존재 기대"
|
||||
assert text_sa.frame_slot_id == "product_column", \
|
||||
f"Test 2 text → product_column (narrowest) 기대, got {text_sa.frame_slot_id}"
|
||||
assert text_sa.region_id == "t2.region-1"
|
||||
|
||||
assert transform_sa is not None, "transform content_unit SlotAssignment 존재 기대"
|
||||
assert transform_sa.frame_slot_id == "process_column", \
|
||||
f"Test 2 transform → process_column 기대, got {transform_sa.frame_slot_id}"
|
||||
assert transform_sa.region_id == "t2.region-2"
|
||||
|
||||
# rejection 없음 / overflow_buffer 빈 list (under-fill 허용)
|
||||
assert len(plan2.rejection) == 0, f"Test 2 rejection=0 기대, got {plan2.rejection}"
|
||||
assert plan2.overflow_buffer == []
|
||||
print("[OK] Test 2 (text+transform → F29 → product_column / process_column) passed.")
|
||||
|
||||
print("\n=== B4 v0 self-test PASS ===")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
_run_self_test()
|
||||
Reference in New Issue
Block a user