feat(R06): workspace session snapshot, atomic persistence, history-dir race fix

EPIC R06 (Team Hoa) - workspace/filesystem isolation, no cross-project
mutable state.

R06-T01 domain/workspaces/workspace_session.py
  WorkspaceSession - project_id/workspace_root/sandbox_dir/allowed_paths
  frozen snapshot + is_allowed(path), same "capture once at submit time"
  shape as R04's ConversationExecutionRequest.

R06-T02 infrastructure/persistence/json/{atomic_write,workspace_repository_impl,conversation_repository_impl}.py
  Real bug fixed: core/projects.py::save_project and core/history.py's
  save_conversation/rename_conversation/set_pinned did a plain
  path.write_text(json.dumps(...)) - two syscalls, no atomicity. A crash
  between them leaves a half-written file that load_project/load_conversation
  then silently treat as "missing". All four now write through
  atomic_write.write_json (temp file + os.replace). WorkspaceRepository/
  ConversationRepository are thin object-shaped facades over the same
  (now-atomic) functions, for future application-layer callers.
  NOTE: atomic_write.py is deliberately NOT named atomic_json_file.py -
  R02-T01 (Team Nam) claims that filename for the same purpose app-wide;
  see the checklist for the consolidation TODO.

R06-T03 infrastructure/filesystem/execution_workspace.py
  ExecutionWorkspace names the output_dir/scratch_dir split that already
  exists (core/chat_agent.py's flat workspace_root/.scratch) - does not
  move anything.

R06-T04 ui/chat_panel.py
  The actual race: ChatPanel._persist_session (saves a BACKGROUND turn's
  conversation) resolved its save directory via a live
  self.ctx.config.history_dir() read at save time. ui/workspace_tab.py::
  _load_current mutates that same config field on every project switch, so
  a turn still running when the user switched projects got saved into the
  NEW project's history folder. Fixed by adding "home_history_dir" to the
  per-turn ctx dict (same "home_*" snapshot convention already used for
  session id/messages/title), captured at submit time. Verified with a real
  offscreen-Qt test, not just a unit double:
  tests/integration/test_history_dir_race.py.

R06-T05 application/workspaces/file_workspace_service.py
  FileWorkspaceService - the File Explorer / AI Editor entry point for the
  same safe read/write/edit operations the agent tool loop has, by calling
  core/tools.py::execute_tool directly (same dispatch, same ToolContext
  containment, same audit log) rather than reimplementing any of it.

New tests: tests/unit/test_workspace_session.py,
test_atomic_write_and_repositories.py, test_execution_workspace.py,
test_file_workspace_service.py, tests/integration/test_history_dir_race.py
(29 new tests, incl. 2 real offscreen-Qt integration tests).

Suite: 283 passed, 4 pre-existing failures unrelated to R05/R06 (see
checklist). check_imports: PASS. All new files < 400 LOC.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
2026-08-21 22:34:57 +09:00
co-authored by Claude Sonnet 5
parent ae4fe72b2e
commit cf542b7416
19 changed files with 853 additions and 27 deletions
+5
View File
@@ -0,0 +1,5 @@
"""Workspace file operations for non-agent-loop callers (EPIC R06)."""
from .file_workspace_service import FileWorkspaceService
__all__ = ["FileWorkspaceService"]
@@ -0,0 +1,81 @@
"""FileWorkspaceService - the safe file operations File Explorer and the AI
File Editor need, outside the agent tool loop (R06-T05).
``ui/folder_tab.py`` (File Explorer) and the AI File Editor dialog need the
exact same guarantees the agent's tools already have — path containment
inside the workspace, precise context-anchored edits, syntax warnings on a
bad Python write — but today that logic only exists wired to a model's tool
call (``core/tools.py::execute_tool``). A UI action that isn't a tool call
(browsing the tree, applying an AI-suggested diff from a review dialog) has
no equivalent entry point of its own.
This service IS that entry point. It reuses ``core/tools.py::execute_tool``
verbatim - same dispatch table, same ``ToolContext`` containment check, same
audit-log entry, same Python-syntax warning on write/edit - rather than
re-implementing any of it, so a fix to one path fixes both. It only adds the
:class:`~domain.workspaces.workspace_session.WorkspaceSession` seam: which
workspace root a call is scoped to is decided by the session, not by
whichever folder a widget happens to have open.
"""
from __future__ import annotations
from typing import Any, Dict
class FileWorkspaceService:
"""File operations scoped to one :class:`WorkspaceSession`.
Read-only by name (``list_tree``/``read_preview``) vs. writing
(``write_file``/``apply_edit``) mirrors the same READ/WRITE split
``domain/tools/tool_registry.py`` uses for the agent's own tools - a
caller that only wants to browse never accidentally has write access.
"""
def __init__(self, session) -> None: # WorkspaceSession - see module docstring
self._session = session
def list_tree(self, rel: str = ".") -> Dict[str, Any]:
"""Entries at ``rel`` (default: the workspace root)."""
return self._execute("list_dir", {"path": rel})
def read_preview(self, rel: str) -> Dict[str, Any]:
"""A text file's content (truncated by
``infrastructure/filesystem/file_tools.py::MAX_READ_BYTES``, same as
the agent's ``read_file`` tool)."""
return self._execute("read_file", {"path": rel})
def write_file(self, rel: str, content: str) -> Dict[str, Any]:
"""Create or fully overwrite ``rel``."""
return self._execute("write_file", {"path": rel, "content": content})
def apply_edit(self, rel: str, old_string: str, new_string: str,
replace_all: bool = False) -> Dict[str, Any]:
"""Replace an exact snippet in an existing file - the same
context-anchored algorithm the agent's ``edit_file`` tool uses, so an
AI-suggested diff applies with the same precision and the same
"old_string not found / ambiguous" failure messages either path
would give the caller."""
return self._execute("edit_file", {
"path": rel, "old_string": old_string, "new_string": new_string,
"replace_all": replace_all,
})
# -- internals --------------------------------------------------------- #
def _tool_context(self):
"""A ``ToolContext`` scoped to this session's workspace root.
``flatten_writes=False`` (unlike Cowork's agent context) - File
Explorer must preserve whatever subfolder structure the user is
actually browsing, not collapse every write into the root."""
from cowork_local.infrastructure.filesystem.tool_context import ToolContext
return ToolContext(self._session.workspace_root, flatten_writes=False)
def _execute(self, name: str, args: Dict[str, Any]) -> Dict[str, Any]:
"""Dispatch through ``core/tools.py::execute_tool`` - see the module
docstring for why this delegates instead of reimplementing."""
from cowork_local.core.tools import execute_tool
return execute_tool(self._tool_context(), name, args)
__all__ = ["FileWorkspaceService"]
+9 -3
View File
@@ -66,7 +66,9 @@ def save_conversation(
"outputs": list(outputs or []),
"messages": messages,
}
path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
# R06-T02: atomic write - see infrastructure/persistence/json/atomic_write.py.
from ..infrastructure.persistence.json.atomic_write import write_json
write_json(path, payload)
return path
@@ -78,15 +80,19 @@ def delete_conversation(path) -> None:
def rename_conversation(path, new_title: str) -> None:
from ..infrastructure.persistence.json.atomic_write import write_json
data = load_conversation(path)
data["title"] = new_title
Path(path).write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
write_json(Path(path), data)
def set_pinned(path, pinned: bool) -> None:
from ..infrastructure.persistence.json.atomic_write import write_json
data = load_conversation(path)
data["pinned"] = bool(pinned)
Path(path).write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
write_json(Path(path), data)
def load_conversation(path: Path) -> Dict[str, Any]:
+5 -3
View File
@@ -116,10 +116,12 @@ def new_project(name: str, description: str = "", instructions: str = "",
def save_project(project: Project, directory: Path = None) -> Path:
directory = directory or PROJECTS_DIR
directory.mkdir(parents=True, exist_ok=True)
path = directory / f"{project.project_id}.json"
path.write_text(json.dumps(asdict(project), ensure_ascii=False, indent=2),
encoding="utf-8")
# R06-T02: atomic write — a crash/kill between truncate and write used to
# leave a half-written project.json that load_project() then silently
# treats as "missing" (see infrastructure/persistence/json/atomic_write.py).
from ..infrastructure.persistence.json.atomic_write import write_json
write_json(path, asdict(project))
return path
+26 -17
View File
@@ -57,22 +57,29 @@
---
## 📊 TIẾN ĐỘ THỰC TẾ — TEAM HOA (cập nhật `2026-08-21 22:19`)
## 📊 TIẾN ĐỘ THỰC TẾ — TEAM HOA (cập nhật `2026-08-21 22:57`)
> [!NOTE]
> ### ✅ ĐÃ HOÀN TẤT: 5/5 task của **R05** — branch `feature/teamhoa/r05-r06` (tạo từ `origin/feature/deltateam/refactor-plan`, có sẵn nền R01/R03/R04)
> ### ✅ ĐÃ HOÀN TẤT: 10/10 task của **R05 + R06** — branch `feature/teamhoa/r05-r06` (tạo từ `origin/feature/deltateam/refactor-plan`, có sẵn nền R01/R03/R04)
>
> | EPIC | Task | Trạng thái |
> | :--- | :--- | :--- |
> | **R05** Tool, MCP & Connector Policy | T01 → T05 | ✅ 5/5 |
> | **R06** Workspace, Filesystem & History Isolation | T01 → T05 | ⬜ chưa bắt đầu |
> | **R06** Workspace, Filesystem & History Isolation | T01 → T05 | ✅ 5/5 |
>
> **Kiểm chứng (chạy thật):**
> * `pytest tests/` ➔ **254 pass / 4 fail** (+12 test mới cho R05: `tests/unit/test_tool_registry_and_policy.py`, `test_code_agent_tool_policy.py`, `test_cowork_extra_tool_policy.py`, `test_mcp_source_manager.py`)
> * 4 fail là **lỗi có sẵn từ trước R05**, không liên quan tool/MCP: 2 trong `test_config_security.py` (EPIC R02, đã ghi nhận bởi Team Duy) + 2 trong `test_routing_wiring.py` (môi trường máy này có Ollama/llama3.1 thật + config routing cục bộ khác "fresh install" — không phải do R05).
> * `pytest tests/` ➔ **283 pass / 4 fail** (+41 test mới cho R05+R06, gồm 2 test Qt offscreen thật trong `tests/integration/test_history_dir_race.py`)
> * 4 fail là **lỗi có sẵn từ trước**, không liên quan R05/R06: 2 trong `test_config_security.py` (EPIC R02, đã ghi nhận bởi Team Duy) + 2 trong `test_routing_wiring.py` (môi trường máy này có Ollama/llama3.1 thật + config routing cục bộ khác "fresh install").
> * `python scripts/check_imports.py` ➔ **PASS** (0 Qt import trong `domain/`, `application/`)
> * Mọi file mới **< 400 dòng** (lớn nhất: `domain/tools/tool_registry.py` 125 dòng). `core/tools.py` giảm từ 566 ➔ 291 dòng.
>
> ### 🔧 TÓM TẮT R06
> * **R06-T01**: `domain/workspaces/workspace_session.py::WorkspaceSession` — snapshot bất biến (project_id, workspace_root, sandbox_dir, allowed_paths) + `is_allowed(path)`.
> * **R06-T02**: `infrastructure/persistence/json/{workspace_repository_impl,conversation_repository_impl}.py` bọc `core/projects.py`/`core/history.py`. **Đã sửa bug thật**: `save_project`/`save_conversation`/`rename_conversation`/`set_pinned` trước đây `path.write_text()` không atomic (crash giữa lúc ghi = file JSON hỏng, `load_project`/`load_conversation` coi file hỏng như "không tồn tại" — mất project/hội thoại âm thầm). Giờ cả 4 hàm ghi qua `infrastructure/persistence/json/atomic_write.py::write_json` (temp file + `os.replace`). Có test giả lập crash giữa lúc ghi xác nhận file cũ không bị hỏng.
> * **R06-T03**: `infrastructure/filesystem/execution_workspace.py::ExecutionWorkspace` — đặt tên cho quy ước `.scratch` đã có sẵn (không đổi vị trí file).
> * **R06-T04**: Sửa race trong `ui/chat_panel.py` (không phải trực tiếp `_load_current`, xem "còn nợ" #2). `ChatPanel._persist_session` (lưu hội thoại của turn CHẠY NGẦM, không phải conversation đang xem) trước đây gọi `self.ctx.config.history_dir()` SỐNG tại thời điểm turn xong — nếu user đổi project khi turn còn chạy (`_load_current` ghi `config._project_history_dir`), turn nền lưu nhầm vào thư mục lịch sử của project MỚI. Fix: thêm `"home_history_dir"` vào dict `ctx` per-turn đã có sẵn (cùng quy ước với `home_id`/`home_messages`/`home_title`), chụp tại lúc submit. Test thật bằng Qt offscreen: `tests/integration/test_history_dir_race.py`.
> * **R06-T05**: `application/workspaces/file_workspace_service.py::FileWorkspaceService` — cho File Explorer/AI Editor gọi `execute_tool` (list_dir/read_file/write_file/edit_file) giống agent, không tự viết lại logic.
>
> ### 🔧 TÓM TẮT R05
> * **R05-T01/T02**: `core/tools.py`'s if/elif dispatcher tách thành `infrastructure/filesystem/{file_tools,command_tools,fetch_tools,tool_context}.py` + `domain/tools/{tool_descriptor,tool_registry}.py`. `core/tools.py` còn lại là shim strangler-fig (re-export `ToolContext`/`ToolError`, dispatch qua dict).
> * **R05-T03**: `application/conversations/tool_policy_gateway.py::ToolPolicyGateway` — thay `if gate is not None and name in ("run_command","install_package")` (chat_agent.py) và `if name in (WRITE_TOOLS|MS365_WRITE_TOOLS)` (code_agent.py) bằng một lookup capability chung. Đã verify bằng test: đúng 2 tool cũ vẫn được gate, không tool nào khác bị ảnh hưởng.
@@ -80,8 +87,10 @@
> * **R05-T05**: `infrastructure/mcp/mcp_source_manager.py::McpToolSourceManager` — tách lifecycle connection (cache/lock/start-or-skip) ra khỏi `state.py::AppContext` (trước đây inline trong `_mcp_connections`/`_conn_lock`). `AppContext` giờ chỉ gọi `self._mcp_manager.ensure/stop/stop_all`. `_ext_connections` (Connectors CAD/CAE/MS365/Other) KHÔNG thuộc phạm vi T05, vẫn giữ `_conn_lock` riêng như cũ.
>
> ### 📌 CÒN NỢ / CẦN QUYẾT ĐỊNH
> 1. **Xung đột file với EPIC R02 (Team Nam)**: `docs/refactor/Refactoring_Checklist.md` dòng ~83 giao `infrastructure/persistence/json/atomic_json_file.py` cho Team Nam (R02-T01). R06-T02 (Team Hoa) cũng cần một helper ghi JSON atomic cho `WorkspaceRepository`/`ConversationRepository`. Để tránh 2 team cùng sửa 1 file, R06 sẽ dùng một helper atomic-write cục bộ trong `infrastructure/persistence/json/workspace_repository_impl.py`/`conversation_repository_impl.py` cho tới khi R02 xong, rồi hợp nhất vào `atomic_json_file.py` chung — **cần Team Nam xác nhận** khi họ bắt đầu R02-T01.
> 2. Việc kế tiếp của Team Hoa là **R06** (Workspace, Filesystem & History Isolation).
> 1. **Xung đột file với EPIC R02 (Team Nam)**: R02-T01 giao `infrastructure/persistence/json/atomic_json_file.py` cho Team Nam. R06-T02 cần atomic write NGAY (bug thật, không chờ được) nên đã tạo `infrastructure/persistence/json/atomic_write.py` — tên khác, cùng thư mục, không đụng file của Team Nam. `core/projects.py`/`core/history.py` đang dùng module này trực tiếp. **Cần Team Nam xác nhận khi bắt đầu R02-T01**: nên hợp nhất `atomic_write.py` vào `atomic_json_file.py` (Team Hoa đổi 4 import) hay giữ 2 module riêng (rủi ro trôi giữa 2 cách ghi atomic).
> 2. **`WorkspaceRepository`/`ConversationRepository`/`FileWorkspaceService` chưa có nơi gọi thật** — giống tình trạng `ProviderRegistry` của Team Duy ở R03. Mọi call site sản xuất (`ui/workspace_tab.py`, `ui/folder_tab.py`, `state.py`, task executors) vẫn dùng trực tiếp `core/projects.py`/`core/history.py`/`core/tools.py::execute_tool` — các class mới là seam cho tầng application ở EPIC sau (R07/R08), chưa nối dây.
> 3. **R06-T04 phạm vi thực tế khác một chút so với mô tả gốc**: bug không nằm ở `ui/workspace_tab.py::_load_current` (hàm đó chỉ *set* `config._project_history_dir`, không tự đọc lại nó) mà ở `ui/chat_panel.py::_persist_session` — nơi một turn chạy ngầm đọc SỐNG giá trị đó lúc turn xong. Đã sửa đúng điểm đọc, có test Qt offscreen thật (`tests/integration/test_history_dir_race.py`), nhưng chưa đổi kiến trúc `_load_current` như plan gốc gợi ý (dùng session id thay biến toàn cục) — việc đó cần tách `ChatPanel`/`WorkspaceTab` sâu hơn, thuộc phạm vi R08 (UI/Application Separation).
> 4. R05/R06 xong toàn bộ — Team Hoa chờ chỉ đạo cho **R07** (Scheduling & Workflow Runtime, phối hợp Team Nam) hoặc merge/review trước khi tiếp tục.
---
@@ -180,16 +189,16 @@
* **Team chịu trách nhiệm**: 🟢 **Team Hoa** (Chủ trì)
* **Mục tiêu**: Xóa bỏ biến toàn cục `state.py::active_project_id`, đóng gói workspace per-turn thành `WorkspaceSession` bất biến, bảo vệ an toàn đường dẫn sandbox.
- [ ] **R06-T01 (Team Hoa)**: Định nghĩa `WorkspaceSession` chứa snapshot project id, workspace root ➔ `domain/workspaces/workspace_session.py`
*Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
- [ ] **R06-T02 (Team Hoa)**: Xây dựng `WorkspaceRepository` từ `core/projects.py` & `ConversationRepository` từ `core/history.py` ➔ `infrastructure/persistence/json/workspace_repository_impl.py`
*Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
- [ ] **R06-T03 (Team Hoa)**: Xây dựng `ExecutionWorkspace` quản lý output/scratch files ➔ `infrastructure/filesystem/execution_workspace.py`
*Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
- [ ] **R06-T04 (Team Hoa)**: Khắc phục race condition trong `ui/workspace_tab.py::_load_current`
*Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
- [ ] **R06-T05 (Team Hoa)**: Xây dựng `FileWorkspaceService` xử lý thao tác file cho File Explorer và AI File Editor ➔ `application/workspaces/file_workspace_service.py`
*Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
- [x] **R06-T01 (Team Hoa)**: Định nghĩa `WorkspaceSession` chứa snapshot project id, workspace root ➔ `domain/workspaces/workspace_session.py`
*Start: `2026-08-21 22:19` | End: `2026-08-21 22:24`*
- [x] **R06-T02 (Team Hoa)**: Xây dựng `WorkspaceRepository` từ `core/projects.py` & `ConversationRepository` từ `core/history.py` ➔ `infrastructure/persistence/json/workspace_repository_impl.py`
*Start: `2026-08-21 22:24` | End: `2026-08-21 22:35`*
- [x] **R06-T03 (Team Hoa)**: Xây dựng `ExecutionWorkspace` quản lý output/scratch files ➔ `infrastructure/filesystem/execution_workspace.py`
*Start: `2026-08-21 22:35` | End: `2026-08-21 22:40`*
- [x] **R06-T04 (Team Hoa)**: Khắc phục race condition trong `ui/workspace_tab.py::_load_current`
*Start: `2026-08-21 22:40` | End: `2026-08-21 22:50`*
- [x] **R06-T05 (Team Hoa)**: Xây dựng `FileWorkspaceService` xử lý thao tác file cho File Explorer và AI File Editor ➔ `application/workspaces/file_workspace_service.py`
*Start: `2026-08-21 22:50` | End: `2026-08-21 22:57`*
---
+5
View File
@@ -0,0 +1,5 @@
"""Domain entities for workspace/project isolation (EPIC R06)."""
from .workspace_session import WorkspaceSession
__all__ = ["WorkspaceSession"]
+96
View File
@@ -0,0 +1,96 @@
"""WorkspaceSession - an immutable snapshot of which project a turn belongs
to and where it may touch the filesystem (R06-T01).
``state.py::AppContext.active_project_id`` is a single mutable field read by
every background worker thread. ``ui/workspace_tab.py::_load_current`` writes
it (and the related ``config._project_history_dir``) on the UI thread the
moment the user switches projects - while a turn already running on a
worker thread may read either field mid-switch and end up acting on the
OTHER project's workspace/history for the rest of its run (the race
R06-T04 fixes).
The fix, same shape as R04's ``ConversationExecutionRequest``: capture the
workspace facts a turn needs ONCE, on the thread that knows which project is
selected, into one frozen object. Whatever the user does to the UI afterwards,
the turn keeps using the workspace it was handed at submit time.
Pure domain code: stdlib only, no Qt, no network. It does touch ``Path`` (not
plain strings, unlike ``ConversationExecutionRequest``) because its whole job
is path-containment checking - a snapshot with no room to answer "is this
path mine" would not replace what ``ToolContext.resolve`` currently does
inline.
"""
from __future__ import annotations
from dataclasses import dataclass, field
from pathlib import Path
from typing import Tuple
@dataclass(frozen=True)
class WorkspaceSession:
"""Everything a turn needs to know about ITS workspace, fixed at the
moment it was submitted.
Attributes:
project_id: the project this turn belongs to (``""`` when no project
is selected - e.g. the Code tab, which has no project concept).
workspace_root: the project's sandbox root (``Project.workspace_dir()``).
sandbox_dir: the ``.scratch`` subtree inside ``workspace_root`` used for
generator/helper scripts, never a final deliverable (see
``infrastructure/filesystem/file_tools.py::_flatten_rel``).
allowed_paths: every root a tool call may read/write under. Almost
always just ``(workspace_root,)``; a project with a custom
``output_dir`` outside the managed workspace tree still resolves
to exactly one root - the tuple exists so a future caller (e.g. a
step scoped to a shared input folder) can widen it without a
shape change.
"""
project_id: str
workspace_root: Path
sandbox_dir: Path
allowed_paths: Tuple[Path, ...] = field(default_factory=tuple)
def __post_init__(self) -> None:
if not self.allowed_paths:
object.__setattr__(self, "allowed_paths", (self.workspace_root,))
@classmethod
def from_project(cls, project) -> "WorkspaceSession":
"""Build a session from a ``core.projects.Project``. ``project`` is
typed loosely (not imported) so this module has no dependency on
``core/`` - the caller (``core/projects.py`` itself, or
``application/conversations``) already has the Project in hand."""
root = Path(project.workspace_dir())
return cls(project_id=project.project_id, workspace_root=root,
sandbox_dir=root / ".scratch", allowed_paths=(root,))
@classmethod
def unscoped(cls, workspace_root: Path) -> "WorkspaceSession":
"""A session for callers with no project concept (e.g. the Code tab,
which sandboxes to a plain folder rather than a ``Project``)."""
root = Path(workspace_root)
return cls(project_id="", workspace_root=root, sandbox_dir=root / ".scratch")
def is_allowed(self, path: Path) -> bool:
"""True when ``path`` resolves inside one of :attr:`allowed_paths`.
Same containment rule as ``ToolContext.resolve`` (an exact root match
or a real descendant), but side-effect-free: it reports the answer
instead of raising, so a caller (``FileWorkspaceService``, R06-T05)
can decide what "not allowed" means for its own UI instead of
catching a ``ToolError``.
"""
try:
resolved = Path(path).expanduser().resolve()
except OSError:
return False
for allowed in self.allowed_paths:
root = Path(allowed).resolve()
if resolved == root or root in resolved.parents:
return True
return False
__all__ = ["WorkspaceSession"]
@@ -0,0 +1,80 @@
"""ExecutionWorkspace - the output folder vs. the scratch folder for one
turn, as two distinct properties instead of a name convention (R06-T03).
Today the ``.scratch`` subtree is a special case buried inside
``_flatten_rel`` (``infrastructure/filesystem/file_tools.py``): a generator
script writes there, the deliverable lands in the output root, and
``core/chat_agent.py`` cleans ``.scratch`` up after the turn — but nothing
NAMES "the scratch folder" as a thing; every call site re-derives
``workdir / ".scratch"`` (or checks ``Path(rel).parts[0] == ".scratch"``) by
hand. This class gives that convention one home.
It does not change WHERE files land - ``workspace_root/.scratch`` stays
exactly what it always was. It exists so a caller (an application service,
R06-T05's ``FileWorkspaceService``, or a future turn-cleanup step) can ask
for "the output dir" / "the scratch dir" instead of hand-building the path
and hoping the convention hasn't drifted.
"""
from __future__ import annotations
import shutil
from dataclasses import dataclass
from pathlib import Path
from cowork_local.domain.workspaces import WorkspaceSession
SCRATCH_DIRNAME = ".scratch"
@dataclass(frozen=True)
class ExecutionWorkspace:
"""The two folders a turn actually writes to, derived from a
:class:`WorkspaceSession`.
``output_dir`` is always the session's ``workspace_root`` itself, not a
per-turn subfolder - Cowork's whole design is that every deliverable lands
directly in the one configured Output folder (see
``infrastructure/filesystem/file_tools.py::_flatten_rel``'s docstring).
``scratch_dir`` is the SAME flat ``workspace_root/.scratch`` every turn on
that workspace already shares today (``core/chat_agent.py``'s
``_cleanup_cowork_intermediates`` operates on that exact path) - this
class does not introduce per-turn namespacing that doesn't exist in the
engine yet, only names the existing convention.
``turn_id`` is kept as metadata for callers that want to attribute a
workspace to the turn that used it (logging, future per-turn scratch
namespacing); it does not affect either path today.
"""
session: WorkspaceSession
turn_id: str
@property
def output_dir(self) -> Path:
return self.session.workspace_root
@property
def scratch_dir(self) -> Path:
return self.session.workspace_root / SCRATCH_DIRNAME
def ensure_dirs(self) -> None:
"""Create both folders if they don't exist yet. Callers that only
need one (most do) can skip this and let ``write_file`` create parents
on demand, same as today."""
self.output_dir.mkdir(parents=True, exist_ok=True)
self.scratch_dir.mkdir(parents=True, exist_ok=True)
def cleanup_scratch(self) -> None:
"""Unconditionally remove the scratch subtree.
Coarser than ``core/chat_agent.py::_cleanup_cowork_intermediates``,
which rescues any real deliverable a generator script wrote INSIDE
``.scratch`` before wiping it - that rescue logic stays there. This
is for callers that only need "make the scratch folder go away"
(e.g. before starting a fresh run) and know it holds nothing worth
saving.
"""
shutil.rmtree(self.scratch_dir, ignore_errors=True)
__all__ = ["ExecutionWorkspace", "SCRATCH_DIRNAME"]
+1
View File
@@ -0,0 +1 @@
"""Persistence adapters (EPIC R02/R06)."""
@@ -0,0 +1,8 @@
"""JSON-file persistence adapters: crash-safe writes and the workspace/
conversation repositories built on them (EPIC R06)."""
from .atomic_write import write_json
from .conversation_repository_impl import ConversationRepository
from .workspace_repository_impl import WorkspaceRepository
__all__ = ["write_json", "WorkspaceRepository", "ConversationRepository"]
@@ -0,0 +1,56 @@
"""write_json - crash-safe JSON writes (R06-T02).
``core/projects.py::save_project`` and ``core/history.py``'s
``save_conversation``/``rename_conversation``/``set_pinned`` all do a plain
``path.write_text(json.dumps(...))`` today. That is two syscalls with a gap in
between: a crash, a killed process, or a full disk between the truncate and
the write leaves a half-written, unparseable JSON file - the NEXT read of
that project/conversation then fails outright (``load_project`` /
``load_conversation`` already treat a parse error as "missing", so this isn't
even a loud failure - a project can silently vanish).
``write_json`` fixes this the standard way: write the full content to a
temporary file in the SAME directory (so the following replace is on one
filesystem, not crossing a mount point), then atomically rename it over the
target. Either the old file is still there, or the new one is fully there -
never a partial one.
Transitional note: EPIC R02 (Team Nam, ``docs/refactor/Refactoring_Checklist.md``
R02-T01) plans a shared ``infrastructure/persistence/json/atomic_json_file.py``
for the SAME purpose across the whole app (config, secrets, ...). This module
is deliberately named differently and scoped to R06's two repositories only,
so the two EPICs don't edit the same file in parallel; once R02-T01 lands,
``WorkspaceRepository``/``ConversationRepository`` should switch to it and
this module can go away.
"""
from __future__ import annotations
import json
import os
import tempfile
from pathlib import Path
from typing import Any
def write_json(path: Path, data: Any) -> None:
"""Serialize ``data`` as indented UTF-8 JSON and write it to ``path``
atomically. Creates parent directories if needed."""
path = Path(path)
path.parent.mkdir(parents=True, exist_ok=True)
text = json.dumps(data, ensure_ascii=False, indent=2)
fd, tmp_name = tempfile.mkstemp(dir=str(path.parent), prefix=f".{path.name}.", suffix=".tmp")
try:
with os.fdopen(fd, "w", encoding="utf-8") as handle:
handle.write(text)
handle.flush()
os.fsync(handle.fileno())
os.replace(tmp_name, path)
except BaseException:
try:
os.unlink(tmp_name)
except OSError:
pass
raise
__all__ = ["write_json"]
@@ -0,0 +1,54 @@
"""ConversationRepository - an object-shaped, atomic-write-backed facade over
``core/history.py`` (R06-T02). Same rationale as
``workspace_repository_impl.py``: the module-level functions in
``core/history.py`` are still what production code calls (they now write
atomically themselves), this class is the seam for application-layer code
that wants an object instead of a directory-parameterised function.
"""
from __future__ import annotations
from pathlib import Path
from typing import Any, Dict, List, Optional
from cowork_local.config import HISTORY_DIR
from cowork_local.core.history import (
delete_conversation,
list_conversations,
load_conversation,
new_session_id,
rename_conversation,
save_conversation,
set_pinned,
)
class ConversationRepository:
"""CRUD + search over conversation JSON files, scoped to one
``directory`` (defaults to the app's real ``HISTORY_DIR``)."""
def __init__(self, directory: Optional[Path] = None) -> None:
self._directory = Path(directory) if directory is not None else HISTORY_DIR
def new_session_id(self) -> str:
return new_session_id()
def save(self, kind: str, session_id: str, messages: List[Dict[str, Any]], **kwargs) -> Path:
return save_conversation(self._directory, kind, session_id, messages, **kwargs)
def load(self, path: Path) -> Dict[str, Any]:
return load_conversation(path)
def list(self, query: str = "") -> List[Dict[str, Any]]:
return list_conversations(self._directory, query)
def delete(self, path: Path) -> None:
delete_conversation(path)
def rename(self, path: Path, new_title: str) -> None:
rename_conversation(path, new_title)
def set_pinned(self, path: Path, pinned: bool) -> None:
set_pinned(path, pinned)
__all__ = ["ConversationRepository"]
@@ -0,0 +1,59 @@
"""WorkspaceRepository - an object-shaped, atomic-write-backed facade over
``core/projects.py`` (R06-T02).
``core/projects.py``'s module-level functions (``list_projects``,
``load_project``, ``save_project``, ``new_project``, ``delete_project``) are
still what every existing call site (``ui/workspace_tab.py``, ``state.py``,
task executors) uses, and stay that way - they now write through
:func:`atomic_write.write_json` themselves, so the durability fix applies
whether or not a caller ever touches this class.
This repository exists for the application layer (``application/workspaces``,
R06-T05) to depend on an interface instead of reaching into ``core/`` -
useful once code above ``core/`` starts being written against
``domain``/``application`` seams instead of the legacy module functions. It
is a thin pass-through today, not a re-implementation: same on-disk format,
same directory, same functions underneath.
"""
from __future__ import annotations
from pathlib import Path
from typing import List, Optional
from cowork_local.core.projects import (
PROJECTS_DIR,
Project,
delete_project,
list_projects,
load_project,
new_project,
save_project,
)
class WorkspaceRepository:
"""CRUD over :class:`~cowork_local.core.projects.Project`, scoped to one
``directory`` (defaults to the app's real ``PROJECTS_DIR``; tests pass a
``tmp_path`` so nothing touches the user's real config folder)."""
def __init__(self, directory: Optional[Path] = None) -> None:
self._directory = directory or PROJECTS_DIR
def list(self) -> List[Project]:
return list_projects(self._directory)
def get(self, project_id: str) -> Optional[Project]:
return load_project(project_id, self._directory)
def save(self, project: Project) -> Path:
return save_project(project, self._directory)
def create(self, name: str, description: str = "", instructions: str = "",
output_dir: str = "") -> Project:
return new_project(name, description, instructions, output_dir, self._directory)
def delete(self, project_id: str) -> bool:
return delete_project(project_id, self._directory)
__all__ = ["WorkspaceRepository"]
@@ -0,0 +1,85 @@
"""EPIC R06-T04: the race in ``ui/workspace_tab.py::_load_current``.
``_load_current`` sets ``ctx.config._project_history_dir`` on the SHARED
``AppConfig`` every time the user switches projects in the Workspace screen.
A background turn (one that isn't the conversation currently displayed) used
to resolve its save directory by calling ``ctx.config.history_dir()`` at
``_persist_session`` time - i.e. whenever the turn actually finished, not
when it started. If the user switched projects while it was still running,
the turn's conversation got written into the NEW project's history folder
instead of the one it actually belongs to.
The fix threads a ``home_history_dir`` captured at submit time (same "home_*"
snapshot convention ``ui/chat_panel.py`` already uses for session id/title/
messages) through to the save call. This test drives the real
``ChatPanel._persist_session`` - the actual save path - offscreen.
"""
from __future__ import annotations
import os
from pathlib import Path
import pytest
os.environ.setdefault("QT_QPA_PLATFORM", "offscreen")
from cowork_local.config import AppConfig # noqa: E402
from cowork_local.core.history import list_conversations # noqa: E402
from cowork_local.state import AppContext # noqa: E402
pytest.importorskip("PySide6", reason="Qt is required for the integration suite")
@pytest.fixture(scope="module")
def qt_app():
from PySide6.QtWidgets import QApplication
return QApplication.instance() or QApplication([])
@pytest.fixture
def chat_panel(qt_app, tmp_path: Path):
from cowork_local.ui.chat_panel import ChatPanel
ctx = AppContext(AppConfig.load(tmp_path / "config.json"))
return ChatPanel(ctx, "cowork", "Test")
def test_background_turn_saves_into_the_project_it_started_in(chat_panel, tmp_path):
project_a_dir = tmp_path / "project-a-history"
project_b_dir = tmp_path / "project-b-history"
chat_panel.ctx.config._project_history_dir = project_a_dir
# What ChatPanel._start_turn captures into the per-turn ctx dict at
# submit time (see the "home_history_dir" entry added there for R06-T04).
turn_ctx = {
"home_id": chat_panel.session_id,
"home_messages": [{"role": "user", "content": "hi"}],
"home_title": "Background turn",
"home_history_dir": chat_panel.ctx.config.history_dir(),
"record": {},
}
assert turn_ctx["home_history_dir"] == project_a_dir
# The user switches projects in the Workspace screen WHILE this turn is
# still running - exactly what ui/workspace_tab.py::_load_current does.
chat_panel.ctx.config._project_history_dir = project_b_dir
chat_panel._persist_session(turn_ctx)
assert len(list_conversations(project_a_dir)) == 1
assert list_conversations(project_b_dir) == []
def test_the_currently_viewed_conversation_still_follows_live_selection(chat_panel, tmp_path):
"""_save_snapshot's OTHER caller (the initial "register it in History right
away" call, and _autosave) has no captured history_dir and must keep
resolving it live - that path is for the conversation ACTUALLY on screen,
which should follow whatever project the user has selected right now."""
project_dir = tmp_path / "currently-viewed"
chat_panel.ctx.config._project_history_dir = project_dir
chat_panel._save_snapshot(chat_panel.session_id,
[{"role": "user", "content": "hi"}], "Live view")
assert len(list_conversations(project_dir)) == 1
@@ -0,0 +1,83 @@
"""EPIC R06-T02: atomic JSON writes + the WorkspaceRepository/
ConversationRepository facades over core/projects.py and core/history.py.
The motivating bug: ``core/projects.py::save_project`` used to
``path.write_text(json.dumps(...))`` — two syscalls, no atomicity. A failure
between them must never leave a half-written file on disk; that is the one
property these tests exist to pin.
"""
from __future__ import annotations
import json
import pytest
from cowork_local.infrastructure.persistence.json import (
ConversationRepository,
WorkspaceRepository,
write_json,
)
from cowork_local.infrastructure.persistence.json.atomic_write import write_json as _write_json
def test_write_json_round_trips(tmp_path):
path = tmp_path / "a.json"
write_json(path, {"hello": "world", "n": 3})
assert json.loads(path.read_text(encoding="utf-8")) == {"hello": "world", "n": 3}
def test_write_json_leaves_no_temp_file_behind(tmp_path):
write_json(tmp_path / "a.json", {"x": 1})
assert list(tmp_path.iterdir()) == [tmp_path / "a.json"]
def test_a_failed_write_never_corrupts_the_existing_file(tmp_path, monkeypatch):
"""The whole point of write-temp-then-replace: if the replace step blows
up, the ORIGINAL file must still be there and still be readable — not
truncated, not half-written."""
path = tmp_path / "a.json"
write_json(path, {"version": 1})
import cowork_local.infrastructure.persistence.json.atomic_write as mod
def boom(*_a, **_k):
raise OSError("simulated crash between write and replace")
monkeypatch.setattr(mod.os, "replace", boom)
with pytest.raises(OSError):
write_json(path, {"version": 2})
assert json.loads(path.read_text(encoding="utf-8")) == {"version": 1}
# the abandoned temp file was cleaned up, not left orphaned
assert list(tmp_path.iterdir()) == [path]
def test_workspace_repository_crud_round_trip(tmp_path):
repo = WorkspaceRepository(tmp_path)
project = repo.create("My Project", description="d")
assert [p.project_id for p in repo.list()] == [project.project_id]
project.description = "updated"
repo.save(project)
assert repo.get(project.project_id).description == "updated"
assert repo.delete(project.project_id) is True
assert repo.get(project.project_id) is None
def test_conversation_repository_crud_round_trip(tmp_path):
repo = ConversationRepository(tmp_path)
session_id = repo.new_session_id()
path = repo.save("cowork", session_id, [{"role": "user", "content": "hi"}])
assert [c["session_id"] for c in repo.list()] == [session_id]
repo.rename(path, "Renamed")
repo.set_pinned(path, True)
data = repo.load(path)
assert data["title"] == "Renamed"
assert data["pinned"] is True
repo.delete(path)
assert repo.list() == []
+51
View File
@@ -0,0 +1,51 @@
"""EPIC R06-T03: ExecutionWorkspace names the output-dir/scratch-dir split
that already exists in ``core/chat_agent.py`` (``.scratch`` under the
workspace root) without changing where anything lands."""
from __future__ import annotations
from cowork_local.domain.workspaces import WorkspaceSession
from cowork_local.infrastructure.filesystem.execution_workspace import ExecutionWorkspace
def test_output_dir_is_the_workspace_root_itself(tmp_path):
session = WorkspaceSession.unscoped(tmp_path)
workspace = ExecutionWorkspace(session, turn_id="turn-1")
assert workspace.output_dir == tmp_path
def test_scratch_dir_matches_the_existing_flat_convention(tmp_path):
"""core/chat_agent.py::_cleanup_cowork_intermediates operates on
``output_dir / ".scratch"`` with no per-turn subfolder — this must agree,
or cleanup_scratch() would target a directory nothing ever wrote to."""
session = WorkspaceSession.unscoped(tmp_path)
workspace = ExecutionWorkspace(session, turn_id="turn-1")
assert workspace.scratch_dir == tmp_path / ".scratch"
def test_ensure_dirs_creates_both_folders(tmp_path):
session = WorkspaceSession.unscoped(tmp_path / "root")
workspace = ExecutionWorkspace(session, turn_id="t")
workspace.ensure_dirs()
assert workspace.output_dir.is_dir()
assert workspace.scratch_dir.is_dir()
def test_cleanup_scratch_removes_it_and_leaves_output_dir_alone(tmp_path):
session = WorkspaceSession.unscoped(tmp_path)
workspace = ExecutionWorkspace(session, turn_id="t")
workspace.ensure_dirs()
(workspace.scratch_dir / "helper.py").write_text("print(1)", encoding="utf-8")
(workspace.output_dir / "deliverable.txt").write_text("done", encoding="utf-8")
workspace.cleanup_scratch()
assert not workspace.scratch_dir.exists()
assert (workspace.output_dir / "deliverable.txt").exists()
def test_cleanup_scratch_is_a_no_op_when_never_created(tmp_path):
workspace = ExecutionWorkspace(WorkspaceSession.unscoped(tmp_path), turn_id="t")
workspace.cleanup_scratch() # must not raise
+62
View File
@@ -0,0 +1,62 @@
"""EPIC R06-T05: FileWorkspaceService gives File Explorer / AI Editor the
same safe file operations the agent tool loop already has, via the SAME
``core/tools.py::execute_tool`` dispatch (not a reimplementation)."""
from __future__ import annotations
from cowork_local.application.workspaces import FileWorkspaceService
from cowork_local.domain.workspaces import WorkspaceSession
def _service(tmp_path) -> FileWorkspaceService:
return FileWorkspaceService(WorkspaceSession.unscoped(tmp_path))
def test_write_then_read_round_trips(tmp_path):
service = _service(tmp_path)
written = service.write_file("notes.md", "# Hello")
assert written["ok"] is True
read = service.read_preview("notes.md")
assert read == {"ok": True, "output": "# Hello"}
def test_list_tree_reflects_written_files(tmp_path):
service = _service(tmp_path)
service.write_file("a.txt", "x")
listing = service.list_tree()
assert listing["ok"] is True and "a.txt" in listing["output"]
def test_apply_edit_uses_the_context_anchored_replace(tmp_path):
service = _service(tmp_path)
service.write_file("code.py", "value = 1\n")
edited = service.apply_edit("code.py", "value = 1", "value = 2")
assert edited["ok"] is True
assert service.read_preview("code.py")["output"].strip() == "value = 2"
def test_apply_edit_reports_ambiguous_match_like_the_agent_tool_does(tmp_path):
service = _service(tmp_path)
service.write_file("code.py", "x = 1\nx = 1\n")
edited = service.apply_edit("code.py", "x = 1", "x = 2")
assert edited["ok"] is False
assert "appears" in edited["output"]
def test_path_escape_is_refused_not_a_crash(tmp_path):
workspace = tmp_path / "workspace"
workspace.mkdir()
(tmp_path / "outside.txt").write_text("secret", encoding="utf-8")
service = FileWorkspaceService(WorkspaceSession.unscoped(workspace))
result = service.read_preview("../outside.txt")
assert result["ok"] is False
assert "outside the working folder" in result["output"]
def test_write_preserves_subfolders_unlike_cowork_flatten_writes(tmp_path):
"""File Explorer must not collapse a write into the workspace root the
way Cowork's agent context does (flatten_writes=True there, False here)."""
service = _service(tmp_path)
service.write_file("sub/dir/file.txt", "content")
assert (tmp_path / "sub" / "dir" / "file.txt").read_text(encoding="utf-8") == "content"
+64
View File
@@ -0,0 +1,64 @@
"""EPIC R06-T01: WorkspaceSession is a frozen snapshot, captured once, that a
turn keeps using regardless of what the UI does to the live project
selection afterwards - see the module docstring for the race this replaces.
"""
from __future__ import annotations
from pathlib import Path
import pytest
from cowork_local.domain.workspaces import WorkspaceSession
class _FakeProject:
def __init__(self, project_id: str, root: Path):
self.project_id = project_id
self._root = root
def workspace_dir(self) -> Path:
return self._root
def test_from_project_derives_sandbox_dir_under_the_workspace_root(tmp_path):
project = _FakeProject("proj-a", tmp_path)
session = WorkspaceSession.from_project(project)
assert session.project_id == "proj-a"
assert session.workspace_root == tmp_path
assert session.sandbox_dir == tmp_path / ".scratch"
assert session.allowed_paths == (tmp_path,)
def test_is_allowed_true_for_the_root_and_descendants(tmp_path):
session = WorkspaceSession.from_project(_FakeProject("p", tmp_path))
nested = tmp_path / "sub" / "file.txt"
nested.parent.mkdir(parents=True)
nested.write_text("x", encoding="utf-8")
assert session.is_allowed(tmp_path) is True
assert session.is_allowed(nested) is True
def test_is_allowed_false_outside_the_workspace(tmp_path):
session = WorkspaceSession.from_project(_FakeProject("p", tmp_path / "a"))
outside = tmp_path / "b" / "secret.txt"
assert session.is_allowed(outside) is False
def test_two_sessions_from_different_projects_stay_independent(tmp_path):
"""The exact race this snapshot exists to prevent: a turn holding session
A must never start accepting paths that belong to session B, no matter
what the (mutable, shared) AppContext does after the snapshot was taken."""
session_a = WorkspaceSession.from_project(_FakeProject("a", tmp_path / "a"))
session_b = WorkspaceSession.from_project(_FakeProject("b", tmp_path / "b"))
assert session_a.is_allowed(tmp_path / "b" / "file.txt") is False
assert session_b.is_allowed(tmp_path / "a" / "file.txt") is False
def test_unscoped_session_has_no_project_id(tmp_path):
session = WorkspaceSession.unscoped(tmp_path)
assert session.project_id == ""
assert session.is_allowed(tmp_path / "code.py") is True
+23 -4
View File
@@ -1126,6 +1126,15 @@ class ChatPanel(QWidget):
"snapshot_len": len(snapshot), "out_dir": out_dir,
"home_id": self.session_id, "home_messages": self.messages,
"home_title": self.title, "home_out_root": self.workspace_dir(),
# R06-T04: captured NOW, at submit time — see _persist_session's
# use of this. Without it, a background turn (this session isn't
# the one currently displayed) saves into whatever
# ctx.config.history_dir() resolves to AT THE TIME IT FINISHES,
# which is the *currently viewed* project's history folder if the
# user switched projects (ui/workspace_tab.py::_load_current)
# while this turn was still running — silently saving one
# project's conversation into another project's history folder.
"home_history_dir": self.ctx.config.history_dir(),
"detached": False,
# For re-rendering the in-progress turn if the user reopens this chat:
"display_text": text, "partial": "", "plan_steps": [],
@@ -1356,10 +1365,18 @@ class ChatPanel(QWidget):
return ctx.get("home_id") == self.session_id and not ctx.get("detached")
def _save_snapshot(self, session_id: str, messages: List[Dict[str, Any]],
title: str, inputs: Optional[List[str]] = None) -> None:
title: str, inputs: Optional[List[str]] = None,
history_dir: Optional[Path] = None) -> None:
"""Persist a conversation by id (used both to register it in History the
moment it starts and to save a finished background turn). No-op until it has
a user message. Never raises into the UI."""
a user message. Never raises into the UI.
``history_dir``, when given, is used INSTEAD of
``self.ctx.config.history_dir()`` — see ``_persist_session`` (R06-T04):
a background turn must save into the project it started in, not
whichever project happens to be selected in the Workspace screen by
the time the turn finishes.
"""
if not self.ctx.config.history.get("autosave", True):
return
if not any(m.get("role") == "user" for m in messages):
@@ -1367,7 +1384,8 @@ class ChatPanel(QWidget):
try:
from ..core.history import save_conversation
save_conversation(
self.ctx.config.history_dir(), self.kind, session_id,
history_dir if history_dir is not None else self.ctx.config.history_dir(),
self.kind, session_id,
messages, title, inputs=list(inputs or []), outputs=[],
# Only the CURRENT view knows its project for sure; a background
# turn's save must not overwrite another conversation's project
@@ -1383,7 +1401,8 @@ class ChatPanel(QWidget):
view-based _autosave can't). Outputs are rebuilt from disk on reopen."""
self._save_snapshot(ctx["home_id"], ctx["home_messages"],
ctx.get("home_title", ""),
inputs=ctx.get("record", {}).get("inputs", []))
inputs=ctx.get("record", {}).get("inputs", []),
history_dir=ctx.get("home_history_dir"))
self.history_changed.emit()
def running_session_ids(self):