fix: derive folder/project hierarchy from code paths

Stop syncing asset_hierarchy folder/project rows from Chroma result ids and derive them deterministically from code document paths so tree-stage DB state and embedding flow stay consistent.

Made-with: Cursor

Author: Ara5429

src/service/github_embedding/service.py

@@ -17,7 +17,7 @@
     run_github_embedding_pipeline,
 )
 from src.service.user.asset_hierarchy_sync import (
-    sync_folder_project_rows_from_embedding_result,
+    sync_folder_project_rows_from_code_document_ids,
 )
 
 
@@ -135,12 +135,9 @@ async def run_github_repo_embedding_job(
         chroma=chroma,
         include_summaries=include_summaries,
     )
-    ids_raw = result.get("ids")
-    if isinstance(ids_raw, list):
-        await sync_folder_project_rows_from_embedding_result(
-            user_id=user_id,
-            repo_full_name=repo_full_name,
-            code_document_ids=code_document_ids,
-            result_ids=[str(x) for x in ids_raw],
-        )
+    await sync_folder_project_rows_from_code_document_ids(
+        user_id=user_id,
+        repo_full_name=repo_full_name,
+        code_document_ids=code_document_ids,
+    )
     return result

src/service/user/asset_hierarchy_sync.py

@@ -1,7 +1,7 @@
 """
 selected_repo_assets의 code 항목을 asset_hierarchy(code)에 반영 (데모·임베딩 SSoT).
 
-임베딩 잡 완료 후 folder/project 행을 Chroma ``ids``와 맞추는 동기화도 제공한다.
+folder/project 행은 (임베딩 결과가 아니라) code 문서 id 경로로부터 결정적으로 파생해 동기화한다.
 """
 
 from __future__ import annotations
@@ -10,7 +10,10 @@
 from typing import Any
 
 from src.db.sqlite import connect
-from src.service.github_embedding.paths import split_chroma_document_id
+from src.service.github_embedding.paths import (
+    collect_parent_directories,
+    split_chroma_document_id,
+)
 from src.service.user.selected_assets import get_selected_repo_assets
 
 
@@ -30,6 +33,8 @@ async def sync_code_rows_from_selected_assets(
     해당 ``selected_repo_id``의 ``type=code`` ``asset_hierarchy`` 행을 지우고,
     ``selected_repo_assets`` 중 ``asset_type=code``만 다시 넣는다.
 
+    또한 code 경로로부터 결정되는 ``type IN ('folder','project')`` 행도 함께 갱신한다.
+
     Returns:
         ``{"inserted": int, "ids": list[str]}``
     """
@@ -51,6 +56,7 @@ async def sync_code_rows_from_selected_assets(
         full_name = str(row["repo_full_name"])
         assets = await get_selected_repo_assets(selected_repo_id, db_path=db_path)
 
+        inserted_rel_paths: set[str] = set()
         await conn.execute(
             """
             DELETE FROM asset_hierarchy
@@ -67,6 +73,10 @@ async def sync_code_rows_from_selected_assets(
             if not isinstance(rp, str) or not rp.strip():
                 continue
             doc_id = _chroma_doc_id(full_name, rp)
+            # doc_id := f"{repo_full_name}/{rel}"
+            prefix = f"{full_name}/"
+            rel = doc_id[len(prefix) :]
+            inserted_rel_paths.add(rel)
             await conn.execute(
                 """
                 INSERT INTO asset_hierarchy (id, selected_repo_id, type)
@@ -76,13 +86,71 @@ async def sync_code_rows_from_selected_assets(
             )
             inserted_ids.append(doc_id)
 
+        # folder/project는 code 경로로부터 결정되므로 함께 재생성한다.
+        await conn.execute(
+            """
+            DELETE FROM asset_hierarchy
+            WHERE selected_repo_id = ? AND type IN ('folder', 'project')
+            """,
+            (selected_repo_id,),
+        )
+
+        if inserted_rel_paths:
+            folder_paths = collect_parent_directories(sorted(inserted_rel_paths))
+            for folder in sorted(folder_paths):
+                folder_id = f"{full_name}/{folder}"
+                await conn.execute(
+                    """
+                    INSERT INTO asset_hierarchy (id, selected_repo_id, type)
+                    VALUES (?, ?, 'folder')
+                    """,
+                    (folder_id, selected_repo_id),
+                )
+
+            project_id = f"{full_name}/"
+            await conn.execute(
+                """
+                INSERT INTO asset_hierarchy (id, selected_repo_id, type)
+                VALUES (?, ?, 'project')
+                """,
+                (project_id, selected_repo_id),
+            )
+
         await conn.commit()
     finally:
         await conn.close()
 
     return {"inserted": len(inserted_ids), "ids": inserted_ids}
 
 
+def _folder_project_rows_from_code_document_ids(
+    repo_full_name: str,
+    code_document_ids: list[str],
+) -> list[tuple[str, str]]:
+    """
+    code 문서 id(`owner/repo/<rel_path>`)들로부터 folder/project id를 파생한다.
+    """
+    rel_paths: set[str] = set()
+    for doc_id in code_document_ids:
+        r, rel = split_chroma_document_id(doc_id)
+        if r != repo_full_name:
+            continue
+        if rel == "/":
+            # code는 project root가 될 수 없다.
+            continue
+        rel_paths.add(rel)
+
+    if not rel_paths:
+        return []
+
+    folder_paths = collect_parent_directories(sorted(rel_paths))
+    rows: list[tuple[str, str]] = []
+    for folder in sorted(folder_paths):
+        rows.append((f"{repo_full_name}/{folder}", "folder"))
+    rows.append((f"{repo_full_name}/", "project"))
+    return rows
+
+
 def _folder_project_rows_from_embedding_ids(
     repo_full_name: str,
     code_document_ids: list[str],
@@ -122,8 +190,28 @@ async def sync_folder_project_rows_from_embedding_result(
     db_path: str | Path | None = None,
 ) -> None:
     """
-    ``result_ids``에서 folder/project id만 추출한 뒤,
-    해당 ``selected_repo``의 ``type IN ('folder','project')`` 행을 삭제하고 다시 INSERT 한다.
+    (호환용) folder/project를 sync할 때 Chroma `result_ids`에 의존하지 않고,
+    `code_document_ids` 경로로부터 결정적으로 재생성한다.
+    """
+    _ = result_ids
+    await sync_folder_project_rows_from_code_document_ids(
+        user_id=user_id,
+        repo_full_name=repo_full_name,
+        code_document_ids=code_document_ids,
+        db_path=db_path,
+    )
+
+
+async def sync_folder_project_rows_from_code_document_ids(
+    *,
+    user_id: str,
+    repo_full_name: str,
+    code_document_ids: list[str],
+    db_path: str | Path | None = None,
+) -> None:
+    """
+    `code_document_ids`로부터 folder/project id를 파생해,
+    해당 ``selected_repo``의 ``type IN ('folder','project')`` 행을 DELETE 후 INSERT 한다.
     """
     conn = await connect(db_path)
     try:
@@ -149,10 +237,9 @@ async def sync_folder_project_rows_from_embedding_result(
             (selected_repo_id,),
         )
 
-        for doc_id, htype in _folder_project_rows_from_embedding_ids(
-            repo_full_name,
-            code_document_ids,
-            result_ids,
+        for doc_id, htype in _folder_project_rows_from_code_document_ids(
+            repo_full_name=repo_full_name,
+            code_document_ids=code_document_ids,
         ):
             await conn.execute(
                 """

tests/api/test_selected_repo_assets_api.py

@@ -243,22 +243,26 @@ def test_asset_hierarchy_sync_inserts_code_rows(
     assert body["inserted"] == 1
     assert body["ids"] == ["owner/repo-a/src/a.py"]
 
-    async def check_db() -> list[str]:
+    async def check_db() -> list[tuple[str, str]]:
         conn = await connect(db_with_selected_repos["db_path"])
         try:
             cur = await conn.execute(
                 """
-                SELECT id FROM asset_hierarchy
-                WHERE selected_repo_id = ? AND type = 'code'
-                ORDER BY id
+                SELECT id, type FROM asset_hierarchy
+                WHERE selected_repo_id = ?
+                ORDER BY type ASC, id ASC
                 """,
                 (sid,),
             )
             rows = await cur.fetchall()
             await cur.close()
-            return [r["id"] for r in rows]
+            return [(r["id"], r["type"]) for r in rows]
         finally:
             await conn.close()
 
-    assert _run(check_db()) == ["owner/repo-a/src/a.py"]
+    assert _run(check_db()) == [
+        ("owner/repo-a/src/a.py", "code"),
+        ("owner/repo-a/src", "folder"),
+        ("owner/repo-a/", "project"),
+    ]
 

tests/service/test_asset_hierarchy_sync_folder_project.py

@@ -1,11 +1,11 @@
-"""asset_hierarchy folder/project 동기화 (임베딩 result ids)."""
+"""asset_hierarchy folder/project 동기화 (code document ids)."""
 
 from __future__ import annotations
 
 import asyncio
 
 from src.db.sqlite import connect, create_all_tables_async
-from src.service.user.asset_hierarchy_sync import sync_folder_project_rows_from_embedding_result
+from src.service.user.asset_hierarchy_sync import sync_folder_project_rows_from_code_document_ids
 
 
 def _run(coro):
@@ -43,7 +43,7 @@ async def setup() -> int:
         await conn.execute(
             """
             INSERT INTO asset_hierarchy (id, selected_repo_id, type)
-            VALUES ('o/r/a.py', ?, 'code')
+                VALUES ('o/r/src/a.py', ?, 'code')
             """,
             (sr_id,),
         )
@@ -60,14 +60,11 @@ async def setup() -> int:
 
     sr_id = _run(setup())
 
-    result_ids = ["o/r/a.py", "o/r/src", "o/r/"]
-
     async def sync() -> None:
-        await sync_folder_project_rows_from_embedding_result(
+        await sync_folder_project_rows_from_code_document_ids(
             user_id="u1",
             repo_full_name="o/r",
-            code_document_ids=["o/r/a.py"],
-            result_ids=result_ids,
+            code_document_ids=["o/r/src/a.py"],
             db_path=db,
         )
 
@@ -91,7 +88,7 @@ async def verify() -> list[tuple[str, str]]:
         return [(str(r["id"]), str(r["type"])) for r in rows]
 
     assert _run(verify()) == [
-        ("o/r/a.py", "code"),
+        ("o/r/src/a.py", "code"),
         ("o/r/src", "folder"),
         ("o/r/", "project"),
     ]
@@ -142,11 +139,10 @@ async def setup() -> int:
     sr_id = _run(setup())
 
     async def sync() -> None:
-        await sync_folder_project_rows_from_embedding_result(
+        await sync_folder_project_rows_from_code_document_ids(
             user_id="u1",
             repo_full_name="o/r",
             code_document_ids=[],
-            result_ids=[],
             db_path=db,
         )
 

week-issues/week4-asset-hierarchy-embedding.md

@@ -7,10 +7,10 @@
 | `type` | 채우는 경로 |
 |--------|-------------|
 | **code** | `sync_code_rows_from_selected_assets(selected_repo_id)` — `selected_repo_assets`에 저장된 **파일(code)** 경로를 Chroma 문서 id 형식 `owner/repo/상대경로`로 INSERT |
-| **folder** / **project** | 임베딩 잡 완료 후 `sync_folder_project_rows_from_embedding_result` — 파이프라인이 Chroma에 넣은 `result["ids"]`에서 code가 아닌 id만 골라 DB 반영. `rel == '/'` → **project**, 그 외 상위 폴더 경로 → **folder** |
+| **folder** / **project** | 임베딩 잡 완료 후 `sync_folder_project_rows_from_code_document_ids` — `code_document_ids`(코드 경로)로부터 부모 디렉터리들(**folder**)과 프로젝트 루트(**project**)를 파생해 DB 반영. |
 
 - **code**는 “에셋 저장·동기화” 흐름에서 먼저 채워진다.
-- **folder** / **project**는 임베딩이 끝난 뒤, Chroma에 생성된 id 목록과 동기화된다.
+- **folder** / **project**는 임베딩이 끝난 뒤, `code_document_ids` 경로로부터 재생성되어 동기화된다.
 
 ## 임베딩 파이프라인
 

week-issues/week4-pr-asset-hierarchy.md

@@ -11,7 +11,7 @@ https://github.com/kocory1/AutoPolio/compare/main...week4/asset-hierarchy-embedd
 ## 제목
 
 ```
-feat: asset_hierarchy에 folder/project 동기화 (임베딩 result ids)
+feat: asset_hierarchy에 folder/project 동기화 (code path 기반)
 ```
 
 ---
@@ -20,12 +20,13 @@ feat: asset_hierarchy에 folder/project 동기화 (임베딩 result ids)
 
 ```markdown
 ## Summary
-- 임베딩 잡(`run_github_repo_embedding_job`) 완료 후 Chroma `result["ids"]`에서 **folder·project** id만 추출해 `asset_hierarchy`를 갱신한다 (기존 `folder`/`project` 행 DELETE 후 INSERT).
-- **`code`** 행은 기존과 같이 `sync_code_rows_from_selected_assets`로만 채운다. **folder/project**는 파이프라인이 만든 Chroma id와 DB를 맞춘다.
+- 임베딩 잡(`run_github_repo_embedding_job`) 완료 후 **`code_document_ids`(코드 경로)**로부터 결정적으로 **folder·project** id를 재생성해 `asset_hierarchy`를 갱신한다 (기존 `folder`/`project` 행 DELETE 후 INSERT).
+- **`code`** 행은 기존과 같이 `sync_code_rows_from_selected_assets`로만 채운다.
+- **`folder/project`**는 Chroma `result["ids"]`가 아니라 **code 경로(부모 디렉터리 + project root)**로 파생한다.
 - 설계·운영 전제는 `week-issues/week4-asset-hierarchy-embedding.md` 참고.
 
 ## Changes
-- `src/service/user/asset_hierarchy_sync.py`: `sync_folder_project_rows_from_embedding_result`, `_folder_project_rows_from_embedding_ids`
+- `src/service/user/asset_hierarchy_sync.py`: `sync_folder_project_rows_from_code_document_ids`
 - `src/service/github_embedding/service.py`: 파이프라인 직후 위 동기화 호출
 - `tests/service/test_asset_hierarchy_sync_folder_project.py`: 동기화 단위 테스트