Tool count 131 -> 154; 474 -> 1242 collected tests. The release exists for two measured reasons. A 149-tool catalog costs a client ~40,300 tokens before it has asked for anything, and a drafter searching the way drafters think -- FILLET, BPOLY, QSELECT, WBLOCK -- scored df=0 against the stock index, because those words were simply not in the data. Discovery and batching - DISCOVERY_MODE=search replaces the catalog with search_tools + call_tool, ranked over an AutoCAD command and synonym corpus (discovery/aliases.py) on top of fastmcp's BM25 index. Measured: 40,305 -> 356 tokens advertised. - cad_batch runs a step list in one round trip; fields= projects result-heavy responses. The `core` profile is gone -- discovery replaced it. Honesty debt (Tier 0) - OCS/ECS: `extrusion` appeared nowhere in this codebase, so mirrored and rotated geometry reported coordinates that were quietly wrong. backends/ocs.py translates at the backend boundary; every tool coordinate is WCS on both engines, and a tilted plane refuses rather than answering with a number that looks fine. - drawing_save_as(fmt="dwg") wrote DXF bytes to a .dwg path. drawing_audit never repaired. analysis_measure_area shoelaced points the caller typed in. - ezdxf calls gained a per-call timeout, so one hung call can no longer wedge a server whose document lock is a single asyncio.Lock. Architecture - The 1854-line backend ABC is 19 contract modules; @capability(key, reason=) makes optional methods declarative and a gate fails on undeclared holes. - 8 layout/viewport tools, CHSPACE, CAD_PROGID escape hatch. - Wave A: selection filters, boundary tracing, hatch depth, annotation objects, measurement -- 13 tools of a planned 39. The REGION and 2D-boolean family was cut on evidence: add_region() yields zero ACIS bytes and the greiner_hormann substitute loses 28.2% of the area on a square with one semicircular edge. Verified against a live AutoCAD 2026, which found four COM defects no amount of reading would have caught -- AcadPViewport has no ViewCenter, and the line that set it had been swallowed by a bare except since v1.4. Benchmark v3 (15 tasks) and what it caught - v2 scored this server 10/10, which carried no information: every task in it exercised something the server was built around. Five tasks were added because they can fail, and three did while being written -- a hatch could not be measured though the refusal listed HATCH as measurable, the boundary tools and the measure tool disagreed about the same shape (139.2177 vs 139.2699), and a circle stored as two bulged vertices measured 0.0. - Competitor reports are NOT re-scored. They are pinned v1.4 runs against v2 and were never asked these five questions; the chart shows those rows `not run`. - Correctness A/B: 21/24 -> 24/24, zero regressions. Performance, stated plainly The published v1.4 perf report was recorded on CPython 3.14 and the new one on 3.11.15, so the apparent 2-6x speedup is the interpreter, not this release. Re-measured on one interpreter, v1.5.0 is 28% slower creating entities and 22% slower on the 10k roundtrip. Fully attributed: EZDXF_CALL_TIMEOUT=0 restores v1.4.0's exact numbers. That is the deadlock fix being paid for, the knob is documented, and winning it back is on the 1.6 roadmap. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
135 lines
5.8 KiB
Python
135 lines
5.8 KiB
Python
"""Optional contract methods: declared once, refused loudly, gated by a test.
|
|
|
|
The backend contract is large (112 abstract methods before this file existed,
|
|
and the plan adds ~39 more). Every one of them is an `@abstractmethod`, which
|
|
means adding a single method to the contract makes **both** backends fail at
|
|
instantiation until both are written — and two test modules construct
|
|
`ComBackend` at import time, so the whole suite goes red on a half-finished
|
|
edit. That is a real cost paid on every contract change.
|
|
|
|
The obvious fix is to give new methods a default body. The obvious default is
|
|
wrong: returning ``{"ok": False, "capability": ...}`` from a method nobody
|
|
implemented produces exactly the quiet "this didn't work and nothing told you"
|
|
that the v1.5.0 honesty pass spent a milestone deleting. A default that lies
|
|
quietly is worse than a build break.
|
|
|
|
So `@capability` gives a default that **raises**
|
|
`UnsupportedCapabilityError` — the same typed refusal the deliberate
|
|
boundaries raise, carrying the same `capability` key, surfaced by the same
|
|
`CapabilityRefusalMiddleware`, and indistinguishable to a client from a
|
|
hand-written refusal. Three properties follow:
|
|
|
|
* An unimplemented method refuses instead of pretending. It cannot silently
|
|
succeed and it cannot silently do nothing.
|
|
* The refusal is *declared*: `test_capability.py` fails if a key used here is
|
|
missing from either backend's `CapabilityMap`, so `system_capabilities` can
|
|
never disagree with what a call actually does. This is the part that keeps
|
|
the escape hatch honest — without it, `@capability` would just be a way to
|
|
ship undeclared holes.
|
|
* Overriding is ordinary subclassing. A backend that implements the method
|
|
simply defines it; nothing needs to be un-registered.
|
|
|
|
`@abstractmethod` remains the default for anything both backends must have.
|
|
Reach for `@capability` only when a method is genuinely optional — an engine
|
|
boundary (no ACIS kernel headlessly) rather than an unfinished chore.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import functools
|
|
from collections.abc import Callable
|
|
from typing import Any
|
|
|
|
__all__ = [
|
|
"UnsupportedCapabilityError",
|
|
"capability",
|
|
"capability_keys",
|
|
"is_capability_default",
|
|
]
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# The refusal vocabulary
|
|
#
|
|
# This lives here rather than in ``backends.base`` because the contracts import
|
|
# the decorator and ``base`` imports the contracts — putting the exception in
|
|
# ``base`` makes that a cycle. ``backends.base`` re-exports both names, so
|
|
# ``from backends.base import UnsupportedCapabilityError`` is unchanged for the
|
|
# rest of the repo.
|
|
#
|
|
# There is exactly one authoring rule: a backend that cannot do a thing RAISES
|
|
# ``UnsupportedCapabilityError(capability, message)``, where ``capability`` is a
|
|
# key of ``capabilities().features`` so a client can pre-check the same boundary
|
|
# through ``system_capabilities`` instead of parsing English. Tools never catch
|
|
# it and never hand-build refusal payloads — one middleware turns it into the
|
|
# wire payload, and ``cad_batch`` turns it into a step row. One class object,
|
|
# defined once: two look-alike types would silently break every ``isinstance``
|
|
# classifier.
|
|
#
|
|
# The ``{"ok": False, "error": ...}`` returned-dict shape stays legal only for
|
|
# soft failures that are not capability boundaries ("Layout not found", "No
|
|
# active transaction").
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
def _unsupported(error: str, capability_key: str) -> dict:
|
|
"""Build a backend's machine-readable "this backend can't" payload."""
|
|
return {"ok": False, "error": error, "capability": capability_key}
|
|
|
|
|
|
class UnsupportedCapabilityError(RuntimeError):
|
|
"""Raised when a call needs a capability this backend does not have.
|
|
|
|
Stays a plain ``RuntimeError``: rebasing it on ``FastMCPError`` was measured
|
|
and changes nothing a client can observe, because the refusal is carried by
|
|
``ToolResult(structured_content=..., is_error=True)`` rather than by the
|
|
exception class. ``to_dict()`` keeps the ``{ok, error, capability}`` payload
|
|
attached to the raise so the boundary survives the ``__cause__`` chain.
|
|
"""
|
|
|
|
def __init__(self, capability: str, error: str):
|
|
super().__init__(error)
|
|
self.capability = capability
|
|
|
|
def to_dict(self) -> dict:
|
|
return _unsupported(str(self), self.capability)
|
|
|
|
|
|
#: ``{capability_key: (method_name, ...)}`` for every method declared optional.
|
|
#: Populated at import by the decorator and read by the parity gate.
|
|
_REGISTRY: dict[str, set[str]] = {}
|
|
|
|
|
|
def capability(key: str, *, reason: str = "") -> Callable[[Callable], Callable]:
|
|
"""Mark a contract method optional, defaulting to a typed refusal.
|
|
|
|
Args:
|
|
key: capability name; must appear in **both** backends' capability maps.
|
|
reason: appended to the refusal message. Say what the caller can do
|
|
instead, not merely that they cannot do this.
|
|
"""
|
|
|
|
def decorate(func: Callable) -> Callable:
|
|
@functools.wraps(func)
|
|
async def default(self, *args: Any, **kwargs: Any):
|
|
message = f"{func.__name__}: not supported by the {self.name} backend"
|
|
raise UnsupportedCapabilityError(key, f"{message}. {reason}".rstrip())
|
|
|
|
default.__capability__ = key
|
|
default.__is_capability_default__ = True
|
|
_REGISTRY.setdefault(key, set()).add(func.__name__)
|
|
return default
|
|
|
|
return decorate
|
|
|
|
|
|
def capability_keys() -> dict[str, set[str]]:
|
|
"""Every optional-capability key and the methods declared under it."""
|
|
return {key: set(names) for key, names in _REGISTRY.items()}
|
|
|
|
|
|
def is_capability_default(owner: type, method_name: str) -> bool:
|
|
"""True when ``owner`` inherited the refusing default rather than implementing it."""
|
|
attribute = getattr(owner, method_name, None)
|
|
return bool(getattr(attribute, "__is_capability_default__", False))
|