fix(task-5): pid_from_spec shields its rollback so a real client cancellation rolls the sheet back on both engines

The MCP SDK cancels a request by cancelling an anyio scope, and anyio
cancellation is level-triggered: task.cancel() is re-issued on every loop
iteration while the task is inside the cancelled scope. run_spec's
`except BaseException: await backend.transaction_rollback()` was therefore
cancelled at its first await, before the rollback body ran. Measured through
the real transport (fastmcp.Client + notifications/cancelled): five symbols
left in an open transaction on ezdxf; on the COM shape EndUndoMark / _UNDO B
never sent while the backend's `finally` already cleared _transaction_active,
so the next transaction_begin nested a StartUndoMark. A native Task.cancel()
(what the plan's Step-4 note reasoned about) is delivered once and clears,
which is why raising CancelledError from a mock passed trivially.

The rollback now runs inside anyio.CancelScope(shield=True). The COM hard
timeout (asyncio.wait_for inside ComBackend._run) still fires inside the
shield, measured at 0.62 s with a 0.5 s deadline, so a hung rollback cannot
freeze the server.

Tests: the mock-raise test is replaced by real cancellations that fail on the
unshielded code — a cancelled anyio task-group scope (parametrized alongside
the native Task.cancel() case), the real transport via Client.cancel(request
id) polled mid-run, and the COM shape (real ComBackend._run, a real
single-thread executor and the real transaction_rollback over a recording
ActiveX document, with a call in flight when the scope is cancelled) which
asserts StartUndoMark, EndUndoMark, _UNDO B land and the next begin is fresh.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FKFStjEFzpGyLq5bMpWxTd

engineering/pid/spec.py

@@ -6,6 +6,8 @@ import copy
 import math
 from typing import TYPE_CHECKING
 
+import anyio
+
 from .critique import PID_FOCUSES, issues_for
 from .drawlines import draw_line
 from .graph import build_graph
@@ -443,7 +445,18 @@ async def run_spec(backend: AutoCADBackend, spec: dict, dry_run: bool = False) -
         # a ``BaseException``) leave the placed symbols in the drawing inside
         # an open transaction — on COM with the undo mark still open, so every
         # later ``pid_from_spec`` was refused until a manual rollback.
-        await backend.transaction_rollback()
+        #
+        # The rollback is shielded because the MCP SDK cancels a request by
+        # cancelling an anyio scope, and anyio cancellation is level-triggered:
+        # ``task.cancel()`` is re-issued on every loop iteration while the task
+        # is still inside the cancelled scope. A bare ``await`` here would be
+        # cancelled at its first suspension, before the rollback body ran —
+        # measured through the real transport: five symbols left in an open
+        # transaction, and on COM the undo mark left open while the backend's
+        # flag already said "no transaction". A native ``Task.cancel()`` is
+        # delivered once and clears, so it was never the case that mattered.
+        with anyio.CancelScope(shield=True):
+            await backend.transaction_rollback()
         raise
     await backend.transaction_commit()
     graph = await build_graph(backend)

tests/test_pid_from_spec.py

@@ -110,20 +110,180 @@ async def test_non_finite_numbers_are_refused_by_path_before_any_write(backend,
     assert (await backend.system_status())["transaction_depth"] == 0
 
 
-async def test_a_cancellation_mid_run_rolls_the_sheet_back(backend, monkeypatch):
+async def _wait_until(predicate, *, timeout: float = 5.0) -> None:
-    """``CancelledError`` is a ``BaseException``; ``except Exception`` let the
+    """Poll an async predicate; a rollback that never lands fails here, not by hanging."""
-    placed symbols stay inside an open transaction."""
     import asyncio
+    import time
+
+    deadline = time.monotonic() + timeout
+    while not await predicate():
+        if time.monotonic() > deadline:
+            raise AssertionError(f"condition not met within {timeout:.1f}s")
+        await asyncio.sleep(0.02)
+
+
+@pytest.mark.parametrize("how", ["scope", "native"])
+async def test_a_cancellation_mid_run_rolls_the_sheet_back(backend, monkeypatch, how):
+    """A *real* cancellation, not a ``CancelledError`` raised from a mock.
+
+    The MCP SDK cancels a request through an anyio cancel scope, which is
+    level-triggered: ``task.cancel()`` is re-issued on every loop iteration
+    while the task is inside the cancelled scope, so an unshielded rollback
+    was cancelled at its first ``await`` before its body ran (five symbols
+    left in an open transaction). A native ``Task.cancel()`` is delivered once
+    and clears — raising from the mock only ever exercised that easy case.
+    """
+    import asyncio
+
+    import anyio
 
     from engineering.pid import spec as spec_module
 
     before = await backend.entity_count()
+    in_flight = asyncio.Event()
 
-    async def cancelled(*args, **kwargs):
+    async def slow_draw_line(*args, **kwargs):
-        raise asyncio.CancelledError()
+        in_flight.set()
+        await asyncio.sleep(30)
+
+    monkeypatch.setattr(spec_module, "draw_line", slow_draw_line)
+
+    if how == "scope":
+        async with anyio.create_task_group() as tg:
+            tg.start_soon(run_spec, backend, EXAMPLE_SPEC)
+            await in_flight.wait()
+            assert await backend.entity_count() > before, "symbols placed before the cancel"
+            tg.cancel_scope.cancel()
+    else:
+        task = asyncio.ensure_future(run_spec(backend, EXAMPLE_SPEC))
+        await in_flight.wait()
+        assert await backend.entity_count() > before, "symbols placed before the cancel"
+        task.cancel()
+        with pytest.raises(asyncio.CancelledError):
+            await task
 
-    monkeypatch.setattr(spec_module, "draw_line", cancelled)
-    with pytest.raises(asyncio.CancelledError):
-        await run_spec(backend, EXAMPLE_SPEC)
     assert await backend.entity_count() == before, "the placed symbols were rolled back"
     assert (await backend.system_status())["transaction_depth"] == 0, "no open transaction"
+
+
+async def test_a_client_cancellation_through_the_transport_rolls_the_sheet_back(
+    client, monkeypatch
+):
+    """The docstring's promise, measured end to end: ``notifications/cancelled``
+    from a real ``fastmcp.Client`` mid-run leaves no half-drawn sheet and no
+    open transaction."""
+    import asyncio
+
+    from fastmcp.exceptions import McpError
+
+    from engineering.pid import spec as spec_module
+
+    async def slow_draw_line(*args, **kwargs):
+        await asyncio.sleep(30)
+
+    monkeypatch.setattr(spec_module, "draw_line", slow_draw_line)
+
+    async def depth() -> int:
+        return (await client.call_tool("system_status", {})).structured_content["transaction_depth"]
+
+    async def entities() -> int:
+        return (await client.call_tool("analysis_entity_stats", {})).structured_content[
+            "total_entities"
+        ]
+
+    before = await entities()
+    # The session hands out ids sequentially, so the next one is the tool
+    # call's — provided nothing else is sent before the call goes out. Yield
+    # (no requests) until the counter has moved, then start polling.
+    request_id = client.session._request_id
+    call = asyncio.ensure_future(client.call_tool("pid_from_spec", {"spec": EXAMPLE_SPEC}))
+
+    async def call_sent() -> bool:
+        return client.session._request_id > request_id
+
+    await _wait_until(call_sent)
+    await _wait_until(lambda: _is_mid_run(depth, entities, before))
+    await client.cancel(request_id)
+    with pytest.raises(McpError, match="cancelled"):
+        await call
+    # The SDK answers the cancel before the handler's rollback has landed.
+    await _wait_until(lambda: _is_clean(depth, entities, before))
+
+
+async def _is_mid_run(depth, entities, before) -> bool:
+    return await depth() == 1 and await entities() > before
+
+
+async def _is_clean(depth, entities, before) -> bool:
+    return await depth() == 0 and await entities() == before
+
+
+async def test_a_cancellation_with_a_com_call_in_flight_still_closes_the_undo_mark(
+    monkeypatch,
+):
+    """COM shape: real ``ComBackend._run``, a real single-thread executor and the
+    real ``transaction_rollback``; only the ActiveX document is a recorder.
+
+    With a COM call blocking the worker when the scope is cancelled, the
+    unshielded rollback's ``_run`` was cancelled while queued behind it — so
+    ``EndUndoMark`` / ``_UNDO B`` never reached AutoCAD — yet its ``finally``
+    cleared ``_transaction_active``: ``system_status`` said "no transaction"
+    while the undo mark was open, and the next ``transaction_begin`` nested a
+    second ``StartUndoMark``.
+    """
+    import asyncio
+    import time
+    import types
+    from concurrent.futures import ThreadPoolExecutor
+
+    import anyio
+
+    from backends import com_backend as module
+    from engineering.pid import spec as spec_module
+
+    sent: list[str] = []
+    doc = types.SimpleNamespace(
+        Name="Drawing1.dwg",
+        FullName="",
+        StartUndoMark=lambda: sent.append("StartUndoMark"),
+        EndUndoMark=lambda: sent.append("EndUndoMark"),
+    )
+    app = types.SimpleNamespace(Documents=types.SimpleNamespace(Count=1), ActiveDocument=doc)
+    monkeypatch.setattr(module, "_acad_app", lambda: app)  # _ensure_document_state
+    monkeypatch.setattr(module, "_acad_doc", lambda: doc)  # begin / rollback
+    monkeypatch.setattr(
+        module.ComBackend,
+        "_safe_send_command",
+        staticmethod(lambda doc, cmd, deadline_s=8.0: sent.append(cmd) or []),
+    )
+    backend = module.ComBackend()
+    backend._executor = ThreadPoolExecutor(max_workers=1)  # no CoInitialize: nothing is COM here
+    try:
+        placed = 0
+
+        async def fake_place(backend, section, item):
+            nonlocal placed
+            placed += 1
+            return {"handle": f"H{placed}"}
+
+        in_flight = asyncio.Event()
+
+        async def slow_draw_line(backend, *args, **kwargs):
+            in_flight.set()
+            await backend._run(time.sleep, 0.5)  # the worker is inside a COM call
+
+        monkeypatch.setattr(spec_module, "_place", fake_place)
+        monkeypatch.setattr(spec_module, "draw_line", slow_draw_line)
+
+        async with anyio.create_task_group() as tg:
+            tg.start_soon(run_spec, backend, EXAMPLE_SPEC)
+            await in_flight.wait()
+            tg.cancel_scope.cancel()
+
+        assert sent == ["StartUndoMark", "EndUndoMark", "_UNDO B"], sent
+        assert backend._transaction_active is False
+        # The mark really is closed, so the next transaction is a fresh one.
+        assert (await backend.transaction_begin())["ok"] is True
+        assert sent[-1] == "StartUndoMark"
+    finally:
+        backend._executor.shutdown(wait=True)