diff --git a/README.md b/README.md
index 5f00738..b33ebea 100644
--- a/README.md
+++ b/README.md
@@ -27,8 +27,8 @@ Live through COM on Windows, or headless through ezdxf anywhere — one typed co
-> **v1.5 release snapshot:** 241 tools · 8 resources · 5 prompt templates · 4516 collected tests.
-> 241 is the **registered** count; a default install advertises 236 over `tools/list`,
+> **v1.5 release snapshot:** 242 tools · 8 resources · 5 prompt templates · 4516 collected tests.
+> 242 is the **registered** count; a default install advertises 237 over `tools/list`,
> because `ENABLE_3D` is unset. `system_about` is the runtime authority.
## Why this exists
@@ -122,7 +122,7 @@ Claude Desktop, Cursor, or any stdio MCP host. For HTTP: `autocad-mcp --transpor
| Quality loop | `drawing_preflight` → `drawing_plan` → `drawing_critique` → `drawing_refine` → `drawing_finalize` (0–100 score) |
| Delivery | `drawing_deliver`: DXF/PDF/PNG + SHA-256 manifest + reopen-parity checks |
-241 tools in 26 groups; TOOL_PACKS=core hides the nine pid_* tools, the 22 environment tools, the 14 mechanical tools and the 12 architectural tools (TOOL_PACKS=core,mech keeps the mechanical ones, TOOL_PACKS=core,arch the architectural ones) from a client that needs none of them. Plus 8 resources that cost nothing in the tool budget (autocad://drawing/info, layers, blocks, entities/stats, entities/{layer_name}, system/status, pid/symbols, standards/isa51) and 5 prompt templates.
+242 tools in 26 groups; TOOL_PACKS=core hides the nine pid_* tools, the 22 environment tools, the 14 mechanical tools and the 12 architectural tools (TOOL_PACKS=core,mech keeps the mechanical ones, TOOL_PACKS=core,arch the architectural ones) from a client that needs none of them. Plus 8 resources that cost nothing in the tool budget (autocad://drawing/info, layers, blocks, entities/stats, entities/{layer_name}, system/status, pid/symbols, standards/isa51) and 5 prompt templates.
**Two rules worth knowing.** Every coordinate in and out of a tool is WCS on both engines — the one exception is TEXT `rotation`, which stays in the entity frame because a mirrored TEXT is mirror-imaged and no scalar angle expresses that. And never read vertices back and shoelace them: that loses **28.2%** of the area on a semicircular edge, silently. `analysis_measure_entity(handle)` reads the real geometry and states its own accuracy.
diff --git a/discovery/aliases.py b/discovery/aliases.py
index 7286ead..9f32034 100644
--- a/discovery/aliases.py
+++ b/discovery/aliases.py
@@ -2465,6 +2465,19 @@ TOOL_ALIASES: dict[str, ToolAliases] = {
"mimari plan",
),
),
+ # ── Understanding & QA (SECTION 26, track H group D) ────────────────────
+ "drawing_diff": ToolAliases(
+ acad=("COMPARE",),
+ synonyms=(
+ "compare two drawings",
+ "compare two revisions",
+ "what changed between revisions",
+ "drawing differences",
+ "cloud the changes",
+ "revizyonlar arası değişiklik",
+ "revizyonlar arasi degisiklik",
+ ),
+ ),
}
diff --git a/docs/tool-inventory.json b/docs/tool-inventory.json
index 6874746..bcad53a 100644
--- a/docs/tool-inventory.json
+++ b/docs/tool-inventory.json
@@ -3,13 +3,13 @@
"schema_version": 1,
"package_version": "1.5.1",
"totals": {
- "tools": 241,
+ "tools": 242,
"resources": 8,
"prompts": 5,
"groups": 26
},
"groups": {
- "analysis": 12,
+ "analysis": 13,
"architecture": 12,
"batch": 3,
"blocks": 9,
@@ -897,6 +897,20 @@
"destructiveHint": true
}
},
+ {
+ "name": "drawing_diff",
+ "group": "analysis",
+ "cost": "mutate",
+ "summary": "Compare two revisions of a drawing: what moved, changed, was added or removed.",
+ "tags": [
+ "analysis",
+ "query"
+ ],
+ "annotations": {
+ "title": "Drawing: Diff Two Revisions",
+ "destructiveHint": false
+ }
+ },
{
"name": "drawing_export_dwg",
"group": "sheet",
diff --git a/engineering/understand/diff.py b/engineering/understand/diff.py
new file mode 100644
index 0000000..67fca14
--- /dev/null
+++ b/engineering/understand/diff.py
@@ -0,0 +1,645 @@
+"""``drawing_diff``: two revisions of one drawing, matched by geometric signature.
+
+Spec §10. Each revision is read once into ``EntityRecord``s
+(``engineering.understand.snapshot``) and the two are compared in memory.
+Nothing in this module writes to a drawing; the revision clouds of
+``markup=True`` are drawn by the tool through
+``engineering.sheet.revision.add_revision``.
+
+Matching runs in three stages, in this order, and a record leaves the pool the
+moment it is matched:
+
+1. **Exact.** Every record has a signature: its type, layer and space, its
+ geometry rounded to ``tol`` (points, bulges, closed flag, radius, arc
+ angles; the bounding box when it has no points), its text and text height,
+ its block name, rotation, scale and attributes. Equal signatures are the
+ same entity whatever their handles; where several old records share a
+ signature the one with the same handle is taken first. A LINE's two ends
+ are sorted, so a line drawn backwards is the same line.
+2. **Handle.** A record left over on both sides with the same handle and the
+ same type is the same entity, edited. A save that renumbers handles would
+ make this stage pair strangers, so it runs only while handles look stable:
+ when at least half of the exact matches kept their handle. The report says
+ which way it went (``handles_stable``).
+3. **Position.** Within one (type, layer, space), a leftover new record is
+ paired with the nearest leftover old record whose anchor (the mean of its
+ points, else the centre of its box) lies within ``move_tol``; the closest
+ pairs are taken first. ``move_tol`` defaults to 2 % of the drawing's robust
+ diagonal - the diagonal of the 1st-99th percentile box of every model-space
+ anchor of both revisions - so one stray entity kilometres away cannot
+ inflate it.
+
+What is left is added (new side) or removed (old side). A matched pair that
+is not identical is *changed*, and the row says what changed: ``moved_by``
+(dx, dy) when the geometry is the old geometry translated, else ``geometry``
+from -> to; ``text``, ``attribs`` (by tag), ``radius``, ``angles``,
+``height``, ``rotation``, ``scale``, ``block``, ``layer``, ``space``,
+``bulges``, ``closed`` from -> to.
+
+The changes are grouped into clusters for the revision clouds: every changed,
+added and removed record contributes its box (a changed one its old box and
+its new box), boxes within ``move_tol`` of each other join one cluster, and a
+cluster's box is padded by a quarter of ``move_tol`` (never less than ten
+``tol``). ``count`` is the number of distinct changes in the cluster.
+
+Block definitions are compared apart (``block_signatures`` /
+``block_changes``): the entity records carry INSERTs, not the blocks they
+reference, so a redrawn definition changes every drawn copy while every INSERT
+record stays identical.
+
+The 2 % move tolerance, the 1st/99th percentiles, the half-of-the-matches
+handle rule and the cluster padding are this module's declared defaults, not
+values from a standard.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import math
+import os
+from collections import defaultdict
+from collections.abc import Iterable, Sequence
+
+import config
+from engineering.understand.snapshot import EntityRecord, Pt, Snapshot, records_from_doc
+
+__all__ = [
+ "CATEGORIES",
+ "DEFAULT_LIMIT",
+ "DEFAULT_TOL",
+ "MOVE_FRACTION",
+ "PAD_FRACTION",
+ "ROBUST_QUANTILE",
+ "anchor",
+ "block_changes",
+ "block_signatures",
+ "cap_report",
+ "default_move_tol",
+ "diff_snapshots",
+ "read_revision",
+ "robust_box",
+ "signature",
+]
+
+DEFAULT_TOL = 0.01
+#: The default move tolerance, as a share of the robust diagonal.
+MOVE_FRACTION = 0.02
+#: The robust box runs from this quantile of the anchors to its complement.
+ROBUST_QUANTILE = 0.01
+#: A cluster's box is padded by this share of ``move_tol``.
+PAD_FRACTION = 0.25
+DEFAULT_LIMIT = 200
+CATEGORIES = ("changed", "added", "removed")
+
+Box = tuple[float, float, float, float]
+
+
+def _positive(name: str, value) -> float:
+ try:
+ number = float(value)
+ except (TypeError, ValueError):
+ raise ValueError(f"{name} must be a positive number; got {value!r}") from None
+ if not math.isfinite(number) or number <= 0:
+ raise ValueError(f"{name} must be a positive number; got {value!r}")
+ return number
+
+
+def _q(value, tol: float):
+ """``value`` on the ``tol`` grid; a non-finite value is kept as its repr."""
+ number = float(value)
+ if not math.isfinite(number):
+ return repr(number)
+ return int(round(number / tol))
+
+
+def _qp(point, tol: float) -> tuple:
+ return (_q(point[0], tol), _q(point[1], tol))
+
+
+def signature(rec: EntityRecord, tol: float = DEFAULT_TOL) -> tuple:
+ """Everything that makes two records the same entity, rounded to ``tol``."""
+ points = tuple(_qp(p, tol) for p in rec.points)
+ if rec.type == "LINE" and len(points) == 2:
+ points = tuple(sorted(points))
+ if points:
+ geometry = points
+ elif rec.bbox is not None:
+ geometry = tuple(_q(v, tol) for v in rec.bbox)
+ else:
+ geometry = ()
+ return (
+ rec.type,
+ rec.layer,
+ rec.space,
+ geometry,
+ tuple(_q(b, tol) for b in rec.bulges),
+ bool(rec.closed),
+ None if rec.radius is None else _q(rec.radius, tol),
+ None
+ if rec.angles is None
+ else (_q(rec.angles[0] % 360.0, tol), _q(rec.angles[1] % 360.0, tol)),
+ rec.text,
+ None if rec.height is None else _q(rec.height, tol),
+ rec.block,
+ tuple(sorted(rec.attribs)),
+ _q(rec.rotation % 360.0, tol),
+ (_q(rec.scale[0], tol), _q(rec.scale[1], tol)),
+ )
+
+
+def anchor(rec: EntityRecord) -> Pt | None:
+ """The mean of the record's points, else the centre of its box, else None."""
+ if rec.points:
+ n = len(rec.points)
+ point = (sum(p[0] for p in rec.points) / n, sum(p[1] for p in rec.points) / n)
+ elif rec.bbox is not None:
+ x0, y0, x1, y1 = rec.bbox
+ point = ((x0 + x1) / 2.0, (y0 + y1) / 2.0)
+ else:
+ return None
+ if not (math.isfinite(point[0]) and math.isfinite(point[1])):
+ return None
+ return point
+
+
+def _box(rec: EntityRecord) -> Box | None:
+ if rec.bbox is not None:
+ return tuple(float(v) for v in rec.bbox)
+ if rec.points:
+ xs = [p[0] for p in rec.points]
+ ys = [p[1] for p in rec.points]
+ return (min(xs), min(ys), max(xs), max(ys))
+ return None
+
+
+def _quantile_pair(values: list[float], q: float) -> tuple[float, float]:
+ values = sorted(values)
+ last = len(values) - 1
+ low = int(math.floor(q * last))
+ return values[low], values[last - low]
+
+
+def robust_box(*snaps: Snapshot) -> Box | None:
+ """The 1st-99th percentile box of the model-space anchors of ``snaps``
+ (every space when model space is empty); None when there is no anchor."""
+ records = [r for s in snaps for r in s.records]
+ anchors = [a for r in records if r.space == "Model" and (a := anchor(r)) is not None]
+ if not anchors:
+ anchors = [a for r in records if (a := anchor(r)) is not None]
+ if not anchors:
+ return None
+ x0, x1 = _quantile_pair([a[0] for a in anchors], ROBUST_QUANTILE)
+ y0, y1 = _quantile_pair([a[1] for a in anchors], ROBUST_QUANTILE)
+ return (x0, y0, x1, y1)
+
+
+def default_move_tol(old: Snapshot, new: Snapshot, tol: float = DEFAULT_TOL) -> float:
+ """2 % of the robust diagonal of both revisions, never less than ``tol``."""
+ box = robust_box(old, new)
+ diagonal = 0.0 if box is None else math.hypot(box[2] - box[0], box[3] - box[1])
+ return max(MOVE_FRACTION * diagonal, tol)
+
+
+def _nearest_pairs(
+ olds: Sequence[EntityRecord],
+ news: Sequence[EntityRecord],
+ left_old: Iterable[int],
+ left_new: Iterable[int],
+ move_tol: float,
+) -> list[tuple[int, int]]:
+ """Stage 3: closest pairs first, within one (type, layer, space)."""
+ groups: dict[tuple, tuple[list[int], list[int]]] = defaultdict(lambda: ([], []))
+ for i in left_old:
+ groups[(olds[i].type, olds[i].layer, olds[i].space)][0].append(i)
+ for j in left_new:
+ groups[(news[j].type, news[j].layer, news[j].space)][1].append(j)
+ candidates: list[tuple[float, str, str, int, int]] = []
+ for old_ids, new_ids in groups.values():
+ if not old_ids or not new_ids:
+ continue
+ grid: dict[tuple[int, int], list[tuple[int, Pt]]] = defaultdict(list)
+ for i in old_ids:
+ a = anchor(olds[i])
+ if a is not None:
+ cell = (math.floor(a[0] / move_tol), math.floor(a[1] / move_tol))
+ grid[cell].append((i, a))
+ for j in new_ids:
+ b = anchor(news[j])
+ if b is None:
+ continue
+ cx, cy = math.floor(b[0] / move_tol), math.floor(b[1] / move_tol)
+ for dx in (-1, 0, 1):
+ for dy in (-1, 0, 1):
+ for i, a in grid.get((cx + dx, cy + dy), ()):
+ distance = math.dist(a, b)
+ if distance <= move_tol:
+ candidates.append((distance, olds[i].handle, news[j].handle, i, j))
+ candidates.sort()
+ used_old: set[int] = set()
+ used_new: set[int] = set()
+ pairs: list[tuple[int, int]] = []
+ for _distance, _old_handle, _new_handle, i, j in candidates:
+ if i in used_old or j in used_new:
+ continue
+ used_old.add(i)
+ used_new.add(j)
+ pairs.append((i, j))
+ return pairs
+
+
+def _number_changed(a, b, tol: float, *, modulo: float | None = None) -> bool:
+ if a is None or b is None:
+ return (a is None) != (b is None)
+ if modulo is not None:
+ a, b = a % modulo, b % modulo
+ return _q(a, tol) != _q(b, tol)
+
+
+def _movement(o: EntityRecord, n: EntityRecord, tol: float) -> dict:
+ """``moved_by`` for a pure translation, ``geometry`` for anything else."""
+ if o.points and n.points:
+ before, after = list(o.points), list(n.points)
+ elif not o.points and not n.points:
+ if o.bbox is None or n.bbox is None:
+ if (o.bbox is None) != (n.bbox is None):
+ return {"geometry": {"from": o.bbox, "to": n.bbox}}
+ return {}
+ before = [(o.bbox[0], o.bbox[1]), (o.bbox[2], o.bbox[3])]
+ after = [(n.bbox[0], n.bbox[1]), (n.bbox[2], n.bbox[3])]
+ else:
+ return {
+ "geometry": {"from": [list(p) for p in o.points], "to": [list(p) for p in n.points]}
+ }
+ a, b = anchor(o), anchor(n)
+ if a is None or b is None:
+ return {"geometry": {"from": [list(p) for p in before], "to": [list(p) for p in after]}}
+ dx, dy = b[0] - a[0], b[1] - a[1]
+ half = tol / 2.0
+ translated = len(before) == len(after) and all(
+ abs(p[0] + dx - q[0]) <= half and abs(p[1] + dy - q[1]) <= half
+ for p, q in zip(before, after, strict=True)
+ )
+ if not translated:
+ return {"geometry": {"from": [list(p) for p in before], "to": [list(p) for p in after]}}
+ if abs(dx) > half or abs(dy) > half:
+ return {"moved_by": [dx, dy]}
+ return {}
+
+
+def _attrib_changes(o: EntityRecord, n: EntityRecord) -> list[dict]:
+ before, after = dict(o.attribs), dict(n.attribs)
+ return [
+ {"tag": tag, "from": before.get(tag), "to": after.get(tag)}
+ for tag in sorted(set(before) | set(after))
+ if before.get(tag) != after.get(tag)
+ ]
+
+
+def _changes(o: EntityRecord, n: EntityRecord, tol: float) -> dict:
+ out: dict = {}
+ if o.layer != n.layer:
+ out["layer"] = {"from": o.layer, "to": n.layer}
+ if o.space != n.space:
+ out["space"] = {"from": o.space, "to": n.space}
+ out.update(_movement(o, n, tol))
+ if tuple(_q(b, tol) for b in o.bulges) != tuple(_q(b, tol) for b in n.bulges):
+ out["bulges"] = {"from": list(o.bulges), "to": list(n.bulges)}
+ if bool(o.closed) != bool(n.closed):
+ out["closed"] = {"from": bool(o.closed), "to": bool(n.closed)}
+ if _number_changed(o.radius, n.radius, tol):
+ out["radius"] = {"from": o.radius, "to": n.radius}
+ if (o.angles is None) != (n.angles is None) or (
+ o.angles is not None
+ and n.angles is not None
+ and any(
+ _number_changed(a, b, tol, modulo=360.0)
+ for a, b in zip(o.angles, n.angles, strict=True)
+ )
+ ):
+ out["angles"] = {
+ "from": None if o.angles is None else list(o.angles),
+ "to": None if n.angles is None else list(n.angles),
+ }
+ if o.text != n.text:
+ out["text"] = {"from": o.text, "to": n.text}
+ if _number_changed(o.height, n.height, tol):
+ out["height"] = {"from": o.height, "to": n.height}
+ if o.block != n.block:
+ out["block"] = {"from": o.block, "to": n.block}
+ attribs = _attrib_changes(o, n)
+ if attribs:
+ out["attribs"] = attribs
+ if _number_changed(o.rotation, n.rotation, tol, modulo=360.0):
+ out["rotation"] = {"from": o.rotation, "to": n.rotation}
+ if any(_number_changed(a, b, tol) for a, b in zip(o.scale, n.scale, strict=True)):
+ out["scale"] = {"from": list(o.scale), "to": list(n.scale)}
+ return out
+
+
+def _row(rec: EntityRecord) -> dict:
+ at = anchor(rec)
+ return {
+ "handle": rec.handle,
+ "type": rec.type,
+ "layer": rec.layer,
+ "space": rec.space,
+ "at": None if at is None else [at[0], at[1]],
+ }
+
+
+def _tally(changed: list[dict], added: list[dict], removed: list[dict]) -> tuple[dict, dict]:
+ by_layer: dict[str, dict[str, int]] = {}
+ by_type: dict[str, dict[str, int]] = {}
+ for category, rows in (("changed", changed), ("added", added), ("removed", removed)):
+ for row in rows:
+ for table, key in ((by_layer, row["layer"]), (by_type, row["type"])):
+ counts = table.setdefault(key, {"changed": 0, "added": 0, "removed": 0})
+ counts[category] += 1
+ return dict(sorted(by_layer.items())), dict(sorted(by_type.items()))
+
+
+def _box_gap(a: Box, b: Box) -> float:
+ dx = max(0.0, b[0] - a[2], a[0] - b[2])
+ dy = max(0.0, b[1] - a[3], a[1] - b[3])
+ return math.hypot(dx, dy)
+
+
+def _clusters(items: list[tuple[str, Box, int]], *, link: float, pad: float) -> list[dict]:
+ """Single-linkage clusters of change boxes, per space, swept along x."""
+ parent = list(range(len(items)))
+
+ def find(k: int) -> int:
+ while parent[k] != k:
+ parent[k] = parent[parent[k]]
+ k = parent[k]
+ return k
+
+ by_space: dict[str, list[int]] = defaultdict(list)
+ for k, (space, _box_value, _key) in enumerate(items):
+ by_space[space].append(k)
+ for members in by_space.values():
+ members.sort(key=lambda k: items[k][1][0])
+ active: list[int] = []
+ for k in members:
+ box = items[k][1]
+ active = [m for m in active if items[m][1][2] + link >= box[0]]
+ for m in active:
+ if _box_gap(items[m][1], box) <= link:
+ parent[find(k)] = find(m)
+ active.append(k)
+ grouped: dict[int, list[int]] = defaultdict(list)
+ for k in range(len(items)):
+ grouped[find(k)].append(k)
+ clusters = []
+ for members in grouped.values():
+ boxes = [items[k][1] for k in members]
+ clusters.append(
+ {
+ "space": items[members[0]][0],
+ "box": [
+ min(b[0] for b in boxes) - pad,
+ min(b[1] for b in boxes) - pad,
+ max(b[2] for b in boxes) + pad,
+ max(b[3] for b in boxes) + pad,
+ ],
+ "count": len({items[k][2] for k in members}),
+ }
+ )
+ clusters.sort(key=lambda c: (c["space"], c["box"]))
+ return clusters
+
+
+def diff_snapshots(
+ old: Snapshot, new: Snapshot, *, tol: float = DEFAULT_TOL, move_tol: float | None = None
+) -> dict:
+ """Match ``new`` against ``old`` and report every change (spec §10).
+
+ Refuses a non-positive ``tol`` or ``move_tol`` before any work.
+ """
+ tol = _positive("tol", tol)
+ move_tol = (
+ default_move_tol(old, new, tol) if move_tol is None else _positive("move_tol", move_tol)
+ )
+ olds = list(old.records)
+ news = list(new.records)
+
+ # 1. exact
+ pools: dict[tuple, list[int]] = defaultdict(list)
+ for i, rec in enumerate(olds):
+ pools[signature(rec, tol)].append(i)
+ used_old: set[int] = set()
+ left_new: list[int] = []
+ exact = kept_handle = 0
+ for j, rec in enumerate(news):
+ pool = pools.get(signature(rec, tol))
+ if not pool:
+ left_new.append(j)
+ continue
+ at = next((k for k, i in enumerate(pool) if olds[i].handle == rec.handle), 0)
+ i = pool.pop(at)
+ used_old.add(i)
+ exact += 1
+ kept_handle += int(olds[i].handle == rec.handle)
+ left_old = [i for i in range(len(olds)) if i not in used_old]
+ handles_stable = exact == 0 or 2 * kept_handle >= exact
+
+ pairs: list[tuple[int, int, str]] = []
+ # 2. handle
+ if handles_stable:
+ by_handle = {olds[i].handle: i for i in left_old}
+ still_new: list[int] = []
+ for j in left_new:
+ i = by_handle.get(news[j].handle)
+ if i is not None and olds[i].type == news[j].type:
+ pairs.append((i, j, "handle"))
+ del by_handle[news[j].handle]
+ else:
+ still_new.append(j)
+ left_new = still_new
+ taken = {i for i, _j, _how in pairs}
+ left_old = [i for i in left_old if i not in taken]
+
+ # 3. position
+ pairs.extend(
+ (i, j, "position") for i, j in _nearest_pairs(olds, news, left_old, left_new, move_tol)
+ )
+ taken_old = {i for i, _j, _how in pairs}
+ taken_new = {j for _i, j, _how in pairs}
+
+ changed: list[dict] = []
+ items: list[tuple[str, Box, int]] = []
+ for key, (i, j, how) in enumerate(
+ sorted(pairs, key=lambda p: (olds[p[0]].handle, news[p[1]].handle))
+ ):
+ o, n = olds[i], news[j]
+ at = anchor(n)
+ changed.append(
+ {
+ "old_handle": o.handle,
+ "new_handle": n.handle,
+ "type": n.type,
+ "layer": n.layer,
+ "space": n.space,
+ "matched_by": how,
+ "at": None if at is None else [at[0], at[1]],
+ "changes": _changes(o, n, tol),
+ }
+ )
+ for rec in (o, n):
+ box = _box(rec)
+ if box is not None:
+ items.append((rec.space, box, key))
+ base = len(changed)
+ removed_recs = sorted((olds[i] for i in left_old if i not in taken_old), key=lambda r: r.handle)
+ added_recs = sorted((news[j] for j in left_new if j not in taken_new), key=lambda r: r.handle)
+ for offset, rec in enumerate(removed_recs + added_recs):
+ box = _box(rec)
+ if box is not None:
+ items.append((rec.space, box, base + offset))
+ removed = [_row(rec) for rec in removed_recs]
+ added = [_row(rec) for rec in added_recs]
+ by_layer, by_type = _tally(changed, added, removed)
+ pad = max(PAD_FRACTION * move_tol, 10.0 * tol)
+ return {
+ "tol": tol,
+ "move_tol": move_tol,
+ "handles_stable": handles_stable,
+ "summary": {
+ "old_entities": len(olds),
+ "new_entities": len(news),
+ "unchanged": exact,
+ "changed": len(changed),
+ "added": len(added),
+ "removed": len(removed),
+ },
+ "by_layer": by_layer,
+ "by_type": by_type,
+ "changed": changed,
+ "added": added,
+ "removed": removed,
+ "clusters": _clusters(items, link=move_tol, pad=pad),
+ }
+
+
+def cap_report(result: dict, limit: int = DEFAULT_LIMIT) -> dict:
+ """``result`` with each list of ``CATEGORIES`` cut to ``limit`` rows and
+ ``truncated`` saying how many rows each cut dropped."""
+ if int(limit) < 1:
+ raise ValueError(f"limit must be at least 1; got {limit!r}")
+ capped = dict(result)
+ capped["truncated"] = {}
+ for category in CATEGORIES:
+ rows = list(result.get(category) or [])
+ capped[category] = rows[: int(limit)]
+ capped["truncated"][category] = max(0, len(rows) - int(limit))
+ return capped
+
+
+# -- block definitions -----------------------------------------------------------------
+
+_BLOCK_ATTRS = (
+ "start",
+ "end",
+ "center",
+ "insert",
+ "radius",
+ "start_angle",
+ "end_angle",
+ "rotation",
+ "height",
+ "text",
+ "tag",
+ "name",
+ "xscale",
+ "yscale",
+ "major_axis",
+ "ratio",
+)
+
+
+def _q_value(value, tol: float):
+ if isinstance(value, bool):
+ return value
+ if isinstance(value, (int, float)):
+ return _q(value, tol)
+ if isinstance(value, str):
+ return value
+ try:
+ return tuple(_q(v, tol) for v in value)
+ except TypeError:
+ return repr(value)
+
+
+def _block_entity_key(entity, tol: float) -> str:
+ key: list = [entity.dxftype(), entity.dxf.get("layer", "0")]
+ for attr in _BLOCK_ATTRS:
+ if entity.dxf.hasattr(attr):
+ key.append((attr, _q_value(entity.dxf.get(attr), tol)))
+ kind = entity.dxftype()
+ if kind == "LWPOLYLINE":
+ key.append(("points", tuple(_q_value(p, tol) for p in entity.get_points("xyb"))))
+ key.append(("closed", bool(entity.closed)))
+ elif kind == "POLYLINE":
+ key.append(("points", tuple(_q_value(v.dxf.location, tol) for v in entity.vertices)))
+ elif kind == "MTEXT":
+ key.append(("mtext", entity.text))
+ return repr(tuple(key))
+
+
+def block_signatures(doc, *, tol: float = DEFAULT_TOL) -> dict[str, str]:
+ """``{block name: digest}`` for every named block definition of an ezdxf
+ document: its base point and, per entity, its type, layer and main DXF
+ attributes rounded to ``tol``. Layout blocks and anonymous blocks
+ (``*U``, ``*D`` ...) are left out - their names are not stable across saves."""
+ tol = _positive("tol", tol)
+ out: dict[str, str] = {}
+ for block in doc.blocks:
+ name = block.name
+ if block.block_record.is_any_layout or name.startswith("*"):
+ continue
+ base = _q_value(block.block.dxf.base_point, tol)
+ rows = sorted(_block_entity_key(entity, tol) for entity in block)
+ out[name] = hashlib.sha256(repr((base, rows)).encode("utf-8")).hexdigest()
+ return dict(sorted(out.items()))
+
+
+def block_changes(old: dict[str, str] | None, new: dict[str, str] | None) -> dict:
+ """Block definitions added, removed and redrawn between two revisions."""
+ if old is None or new is None:
+ return {
+ "compared": False,
+ "reason": "block definitions are compared only when both revisions are DXF files "
+ "or the current document of the headless engine",
+ }
+ return {
+ "compared": True,
+ "added": sorted(set(new) - set(old)),
+ "removed": sorted(set(old) - set(new)),
+ "changed": sorted(name for name in set(old) & set(new) if old[name] != new[name]),
+ }
+
+
+def read_revision(path: str, *, tol: float = DEFAULT_TOL) -> tuple[Snapshot, dict[str, str]]:
+ """One read of a DXF revision: its records and its block signatures.
+
+ Refused before reading: a file that cannot be stat'ed and one over
+ ``MAX_DXF_BYTES`` (the size and the variable named); a file ezdxf cannot
+ parse is refused with ezdxf's reason.
+ """
+ import ezdxf
+
+ limit = int(config.settings.max_dxf_bytes)
+ try:
+ size = os.path.getsize(path)
+ except OSError as exc:
+ raise ValueError(f"{path}: cannot be read ({exc.strerror or exc})") from exc
+ if limit > 0 and size > limit:
+ raise ValueError(
+ f"{path} is {size} bytes, over the MAX_DXF_BYTES limit of {limit}; "
+ "raise MAX_DXF_BYTES to read it"
+ )
+ try:
+ doc = ezdxf.readfile(path)
+ except (OSError, ezdxf.DXFError) as exc:
+ raise ValueError(f"{path}: not a readable DXF ({exc})") from exc
+ return records_from_doc(doc, source=str(path)), block_signatures(doc, tol=tol)
diff --git a/server.py b/server.py
index 9e3704d..312ad8f 100644
--- a/server.py
+++ b/server.py
@@ -10554,6 +10554,220 @@ async def arch_plan_from_spec(
raise ToolError(str(exc)) from exc
+# ---------------------------------------------------------------------------
+# ── SECTION 26: Understanding & QA (1 tool) ─────────────────────────────────
+# ---------------------------------------------------------------------------
+
+
+async def _diff_revision(backend, path: str | None, *, tol: float, role: str):
+ """One revision for ``drawing_diff``: its records and, where they can be
+ read, its block signatures (None otherwise).
+
+ A DXF path is read once, here, for both. The current document goes through
+ ``take_snapshot``; its block table is read only on the headless engine,
+ under the backend's own lock. Any other path (a DWG) goes through the
+ snapshot too, which opens it read-only on the live engine.
+ """
+ import asyncio
+
+ from engineering.understand.diff import block_signatures, read_revision
+ from engineering.understand.snapshot import take_snapshot
+
+ if path is None:
+ snap = await take_snapshot(backend, None)
+ blocks = None
+ run = getattr(backend, "_async", None)
+ if backend.name == "ezdxf" and run is not None:
+ blocks = await run(lambda: block_signatures(backend._doc, tol=tol))
+ return snap, blocks
+ validated = validate_path(path)
+ if not validated.is_file():
+ raise ToolError(f"{role}: {validated} does not exist or is not a file")
+ if validated.suffix.lower() == ".dxf":
+ try:
+ return await asyncio.to_thread(read_revision, str(validated), tol=tol)
+ except ValueError as exc:
+ raise ToolError(f"{role}: {exc}") from exc
+ try:
+ return await take_snapshot(backend, str(validated)), None
+ except ValueError as exc: # an unreadable suffix; the dwg refusal is not a ValueError
+ raise ToolError(f"{role}: {exc}") from exc
+
+
+async def _diff_markup(backend, result: dict, new_snap, *, rev: str, description: str) -> dict:
+ """Revision clouds around every model-space cluster of change, filed under
+ ``rev`` through ``engineering.sheet.revision.add_revision`` - the only
+ write ``drawing_diff`` ever makes."""
+ from engineering.sheet.bom import LIST_WIDTH, ROW_HEIGHT
+ from engineering.sheet.revision import add_revision
+ from engineering.understand.diff import robust_box
+
+ clusters = [c for c in result["clusters"] if c["space"] == "Model"]
+ paper = len(result["clusters"]) - len(clusters)
+ if not clusters:
+ return {
+ "created": False,
+ "clouds": [],
+ "reason": "no change in model space to cloud",
+ "paper_space_clusters": paper,
+ }
+ boxes = [c["box"] for c in clusters]
+ shortest = min(min(b[2] - b[0], b[3] - b[1]) for b in boxes)
+ box = robust_box(new_snap) or (
+ min(b[0] for b in boxes),
+ min(b[1] for b in boxes),
+ max(b[2] for b in boxes),
+ max(b[3] for b in boxes),
+ )
+ summary = result["summary"]
+ text = description or (
+ f"drawing_diff: {summary['changed']} changed, {summary['added']} added, "
+ f"{summary['removed']} removed"
+ )
+ drawn = await add_revision(
+ backend,
+ rev=rev,
+ description=text,
+ clouds=boxes,
+ at=(box[2] - LIST_WIDTH, box[3] + 3.0 * ROW_HEIGHT),
+ cloud_segment=shortest / 4.0,
+ )
+ drawn["paper_space_clusters"] = paper
+ return drawn
+
+
+@cad_tool(
+ summary="Compare two revisions of a drawing: what moved, changed, was added or removed.",
+ cost="mutate",
+)
+@mcp.tool(
+ annotations={"title": "Drawing: Diff Two Revisions", "destructiveHint": False},
+ tags={"analysis", "query"},
+)
+async def drawing_diff(
+ old_path: Annotated[
+ str,
+ Field(
+ description="The OLDER revision: a DXF path (a DWG only on the live engine, which "
+ "exports it read-only)."
+ ),
+ ],
+ new_path: Annotated[
+ str | None,
+ Field(
+ default=None,
+ description="The NEWER revision: a DXF path, or omitted for the current document.",
+ ),
+ ] = None,
+ tol: Annotated[
+ float,
+ Field(
+ default=0.01,
+ gt=0,
+ description="Geometry is compared rounded to this many drawing units.",
+ ),
+ ] = 0.01,
+ move_tol: Annotated[
+ float | None,
+ Field(
+ default=None,
+ gt=0,
+ description="An unmatched entity is paired with the nearest one of its type and layer "
+ "within this distance and reported as changed; default 2 % of the drawing's robust "
+ "diagonal.",
+ ),
+ ] = None,
+ limit: Annotated[
+ int,
+ Field(
+ default=200, ge=1, le=5000, description="Rows kept per list (changed, added, removed)."
+ ),
+ ] = 200,
+ markup: Annotated[
+ bool,
+ Field(
+ default=False,
+ description="Draw revision clouds around every cluster of change on the current "
+ "document (the newer revision) and file them under `rev`.",
+ ),
+ ] = False,
+ rev: Annotated[
+ str,
+ Field(
+ default="",
+ description="Revision code for markup=True: one to three capitals or digits.",
+ ),
+ ] = "",
+ description: Annotated[
+ str,
+ Field(default="", description="Revision-row text for markup=True; default: the counts."),
+ ] = "",
+ ctx: Context = None,
+) -> dict:
+ """Compare two revisions of one drawing by geometric signature (spec §10).
+
+ Entities are matched exactly first (type, layer, geometry rounded to `tol`,
+ text, block, attributes - handles do not matter), then by handle while
+ handles look stable, then by type + layer + nearest position within
+ `move_tol`; what is left is added or removed. The report counts changes by
+ layer and type, lists each changed entity with what changed (moved by, text
+ from -> to, attribute from -> to, radius from -> to, ...), compares the
+ block definitions when both revisions are DXF files or the headless current
+ document, and groups the changes into clusters.
+
+ Read-only unless `markup=True`: then revision clouds are drawn around the
+ model-space clusters on the current document, which must be the newer
+ revision, and filed as one revision row - the only write this tool makes.
+
+ Refused before anything is read: `markup=True` with `new_path` (the clouds go
+ on the current document), `markup=True` without a valid `rev`, `markup=True`
+ on an engine without the revcloud capability (the live engine -
+ `capability: "revcloud"`), a path that does not exist, a DXF over
+ MAX_DXF_BYTES (the size and the variable named), a file that is not a
+ readable DXF, and a non-positive `tol` or `move_tol`.
+ """
+ from backends.base import UnsupportedCapabilityError
+ from engineering.sheet.revision import REV_RE
+ from engineering.understand.diff import block_changes, cap_report, diff_snapshots
+
+ backend = _backend(ctx)
+ if markup:
+ if new_path:
+ raise ToolError(
+ "markup draws revision clouds on the current document, which must be the newer "
+ "revision: open it with drawing_open and call again without new_path"
+ )
+ if not REV_RE.fullmatch(rev.strip().upper()):
+ raise ToolError(
+ f"markup needs `rev`, the revision code the clouds are filed under - one to "
+ f"three capitals or digits (A, B, 01); got {rev!r}"
+ )
+ feature = backend.capabilities().features.get("revcloud")
+ if feature is None or not feature.supported:
+ raise UnsupportedCapabilityError(
+ "revcloud",
+ f"drawing_diff(markup=True): the {backend.name} backend cannot draw revision "
+ "clouds (no ActiveX member; REVCLOUD is command-line only). Run the diff "
+ "without markup, or on the ezdxf engine.",
+ )
+ await ctx.info(f"drawing_diff: {old_path} -> {new_path or 'the current document'}")
+ old_snap, old_blocks = await _diff_revision(backend, old_path, tol=tol, role="old_path")
+ new_snap, new_blocks = await _diff_revision(backend, new_path, tol=tol, role="new_path")
+ try:
+ result = diff_snapshots(old_snap, new_snap, tol=tol, move_tol=move_tol)
+ except ValueError as exc:
+ raise ToolError(str(exc)) from exc
+ result["old"] = old_snap.source
+ result["new"] = new_snap.source
+ result["blocks"] = block_changes(old_blocks, new_blocks)
+ report = cap_report(result, limit)
+ if markup:
+ report["markup"] = await _diff_markup(
+ backend, result, new_snap, rev=rev.strip().upper(), description=description
+ )
+ return report
+
+
# ---------------------------------------------------------------------------
# ── RESOURCES ───────────────────────────────────────────────────────────────
# ---------------------------------------------------------------------------
diff --git a/tests/fixtures/revisions.py b/tests/fixtures/revisions.py
new file mode 100644
index 0000000..6e6d056
--- /dev/null
+++ b/tests/fixtures/revisions.py
@@ -0,0 +1,96 @@
+"""Two revisions of the synthetic plant P&ID with known edits, for ``drawing_diff``.
+
+Built from ``build_plant_pair`` (``tests/fixtures/plant_pair.py``) with
+ezdxf. The older revision is the P&ID plus one tagged block reference; the
+newer one moves a line by half the move tolerance, rewrites a text, changes
+the block's attribute, deletes one line, adds a circle - and then gives every
+model-space entity a new handle, the way a save through another program
+renumbers them. What was edited is
+picked from the older revision's own records (entities whose signature is
+unique, so a double-drawn line can never be the one edited) and returned with
+the paths: the expected answer is read off the fixture, never typed in.
+"""
+
+from __future__ import annotations
+
+from collections import Counter
+from pathlib import Path
+
+import ezdxf
+
+from engineering.understand.diff import diff_snapshots, read_revision, robust_box, signature
+from engineering.understand.snapshot import records_from_doc
+from tests.fixtures.plant_pair import build_plant_pair
+
+TAG_BLOCK = "DIFF_TAG"
+TAG_LAYER = "DIFF-TAGS"
+ADDED_LAYER = "DIFF-ADDED"
+NEW_TEXT = "REV-B NOTE"
+TAG_BEFORE = "X-1"
+TAG_AFTER = "X-2"
+
+
+def renumber_modelspace(doc) -> None:
+ """Every model-space entity replaced by a copy under a fresh handle."""
+ msp = doc.modelspace()
+ for entity in list(msp):
+ clone = entity.copy()
+ msp.add_entity(clone)
+ msp.delete_entity(entity)
+
+
+def build_revision_pair(directory, *, edits: bool = True) -> dict:
+ directory = Path(directory)
+ truth = build_plant_pair(directory / "plant")
+ base = ezdxf.readfile(truth["pid"])
+ box = robust_box(records_from_doc(base, source="base"))
+ centre = ((box[0] + box[2]) / 2.0, (box[1] + box[3]) / 2.0)
+ block = base.blocks.new(TAG_BLOCK)
+ block.add_circle((0.0, 0.0), 5.0)
+ block.add_attdef("TAG", (0.0, -8.0), dxfattribs={"height": 2.5})
+ base.layers.add(TAG_LAYER)
+ tag = base.modelspace().add_blockref(TAG_BLOCK, centre, dxfattribs={"layer": TAG_LAYER})
+ tag.add_auto_attribs({"TAG": TAG_BEFORE})
+ old_path = directory / "rev_a.dxf"
+ base.saveas(old_path)
+
+ old_snap, _blocks = read_revision(str(old_path))
+ move_tol = diff_snapshots(old_snap, old_snap)["move_tol"]
+ counts = Counter(signature(r) for r in old_snap.records)
+ unique = [r for r in old_snap.records if r.space == "Model" and counts[signature(r)] == 1]
+ moved = next(r for r in unique if r.type == "LINE")
+ text = next(r for r in unique if r.type in ("TEXT", "MTEXT"))
+ insert = next(r for r in old_snap.records if r.block == TAG_BLOCK)
+ deleted = next(
+ r
+ for r in unique
+ if r.type in ("LINE", "LWPOLYLINE") and (r.type, r.layer) != (moved.type, moved.layer)
+ )
+
+ new = ezdxf.readfile(old_path)
+ msp = new.modelspace()
+ if edits:
+ new.entitydb[moved.handle].translate(move_tol / 2.0, 0.0, 0.0)
+ target = new.entitydb[text.handle]
+ if target.dxftype() == "MTEXT":
+ target.text = NEW_TEXT
+ else:
+ target.dxf.text = NEW_TEXT
+ new.entitydb[insert.handle].get_attrib("TAG").dxf.text = TAG_AFTER
+ msp.delete_entity(new.entitydb[deleted.handle])
+ new.layers.add(ADDED_LAYER)
+ msp.add_circle(centre, move_tol / 10.0, dxfattribs={"layer": ADDED_LAYER})
+ renumber_modelspace(new)
+ new_path = directory / "rev_b.dxf"
+ new.saveas(new_path)
+ return {
+ "old": str(old_path),
+ "new": str(new_path),
+ "move_tol": move_tol,
+ "old_count": len(old_snap.records),
+ "moved": moved.handle,
+ "text": text.handle,
+ "text_before": text.text,
+ "insert": insert.handle,
+ "deleted": deleted.handle,
+ }
diff --git a/tests/test_tool_registry.py b/tests/test_tool_registry.py
index e8a1ecb..f09640e 100644
--- a/tests/test_tool_registry.py
+++ b/tests/test_tool_registry.py
@@ -495,7 +495,7 @@ async def test_tool_groups_is_byte_identical_to_the_source_declared_grouping():
async def test_tool_group_sizes_are_unchanged():
- """Frozen snapshot of the surface (241 tools, 26 groups).
+ """Frozen snapshot of the surface (242 tools, 26 groups).
Taken before @cad_tool landed at 131 tools; `batch` moved 2 -> 3 when
v1.5.0's `cad_batch` joined `entity_batch_create`/`entity_batch_modify`, and
@@ -591,10 +591,16 @@ async def test_tool_group_sizes_are_unchanged():
exactly `PACK_TOOL_NAMES["arch"]`, and every group the track-F branches
had grown while the tag was unranked on them returned to its
pre-track-F value.
+
+ Track H group D opened its copy of SECTION 26 (Understanding & QA) with
+ `drawing_diff`, tagged `analysis` and `query`; `analysis` outranks `query`
+ in `_GROUP_TAG_PRIORITY`, so it files under `analysis` (12 -> 13). This
+ snapshot is branch-local: the track H merge recomputes it once the
+ understand, network and diff branches have landed.
"""
sizes = {label: len(names) for label, names in (await server._tool_groups()).items()}
assert sizes == {
- "analysis": 12,
+ "analysis": 13,
"architecture": 12,
"batch": 3,
"blocks": 9,
@@ -621,4 +627,4 @@ async def test_tool_group_sizes_are_unchanged():
"validation": 1,
"view": 4,
}
- assert sum(sizes.values()) == 241
+ assert sum(sizes.values()) == 242
diff --git a/tests/test_understand_diff.py b/tests/test_understand_diff.py
new file mode 100644
index 0000000..26d185c
--- /dev/null
+++ b/tests/test_understand_diff.py
@@ -0,0 +1,309 @@
+"""drawing_diff's engine: signatures, the three matching stages, the report.
+
+Pure tests build ``EntityRecord``s by hand, so every coordinate is the
+answer. The known-edits tests build two revisions of the synthetic plant P&ID
+(``tests/fixtures/revisions.py``) and assert exactly the edits made - a move,
+a text change, an attribute change, an add, a delete, under renumbered
+handles - and nothing else.
+"""
+
+from __future__ import annotations
+
+import ezdxf
+import pytest
+
+import config
+from engineering.understand.diff import (
+ block_changes,
+ block_signatures,
+ cap_report,
+ diff_snapshots,
+ read_revision,
+ robust_box,
+ signature,
+)
+from engineering.understand.snapshot import EntityRecord, Snapshot
+from tests.fixtures.revisions import (
+ ADDED_LAYER,
+ NEW_TEXT,
+ TAG_AFTER,
+ TAG_BEFORE,
+ build_revision_pair,
+)
+
+
+def rec(handle, type_, layer="PIPE", points=(), **kw):
+ space = kw.pop("space", "Model")
+ return EntityRecord(
+ handle=handle, type=type_, layer=layer, space=space, points=tuple(points), **kw
+ )
+
+
+def snap(*records):
+ return Snapshot(
+ source="test",
+ insunits=4,
+ extmin=None,
+ extmax=None,
+ layers={},
+ layouts=(),
+ records=tuple(records),
+ )
+
+
+def line(handle, x1, y1, x2, y2, layer="PIPE"):
+ return rec(handle, "LINE", layer, ((x1, y1), (x2, y2)))
+
+
+STILL = [line("A", 0, 0, 100, 0), line("B", 0, 50, 100, 50), line("C", 0, 100, 100, 100)]
+RENUMBERED = [line("1A", 0, 0, 100, 0), line("1B", 0, 50, 100, 50), line("1C", 0, 100, 100, 100)]
+
+
+# -- signatures ------------------------------------------------------------------------
+
+
+def test_a_signature_is_rounded_to_tol_and_ignores_a_lines_direction():
+ a = line("A", 0.0, 0.0, 10.0, 0.0)
+ b = line("B", 10.001, 0.004, 0.0, 0.0)
+ assert signature(a, 0.01) == signature(b, 0.01)
+ assert signature(a, 0.01) != signature(line("C", 0.0, 0.0, 10.02, 0.0), 0.01)
+
+
+def test_identical_revisions_have_no_change_and_no_cluster():
+ result = diff_snapshots(snap(*STILL), snap(*STILL))
+ assert result["summary"] == {
+ "old_entities": 3,
+ "new_entities": 3,
+ "unchanged": 3,
+ "changed": 0,
+ "added": 0,
+ "removed": 0,
+ }
+ assert result["clusters"] == [] and result["handles_stable"] is True
+
+
+# -- the three stages ------------------------------------------------------------------
+
+
+def test_a_move_under_renumbered_handles_is_matched_by_position():
+ old = snap(*STILL, line("D", 0, 200, 100, 200))
+ new = snap(*RENUMBERED, line("1D", 3, 200, 103, 200))
+ result = diff_snapshots(old, new, move_tol=10.0)
+ assert result["handles_stable"] is False
+ (row,) = result["changed"]
+ assert (row["old_handle"], row["new_handle"], row["matched_by"]) == ("D", "1D", "position")
+ assert row["changes"] == {"moved_by": [pytest.approx(3.0), pytest.approx(0.0)]}
+ assert result["summary"]["unchanged"] == 3
+
+
+def test_a_text_change_is_reported_from_and_to():
+ old = snap(*STILL, rec("T", "TEXT", "NOTES", ((10, 10),), text="OLD", height=2.5))
+ new = snap(*STILL, rec("T", "TEXT", "NOTES", ((10, 10),), text="NEW", height=2.5))
+ (row,) = diff_snapshots(old, new)["changed"]
+ assert row["matched_by"] == "handle"
+ assert row["changes"] == {"text": {"from": "OLD", "to": "NEW"}}
+
+
+def test_an_attribute_change_is_reported_by_tag():
+ before = (("TAG", "A-1"), ("SERVICE", "CIP"))
+ after = (("TAG", "A-2"), ("SERVICE", "CIP"))
+ old = snap(*STILL, rec("I", "INSERT", "TAGS", ((500, 500),), block="TB", attribs=before))
+ new = snap(*STILL, rec("I", "INSERT", "TAGS", ((500, 500),), block="TB", attribs=after))
+ (row,) = diff_snapshots(old, new)["changed"]
+ assert row["changes"] == {"attribs": [{"tag": "TAG", "from": "A-1", "to": "A-2"}]}
+
+
+def test_a_radius_change_is_reported_from_and_to():
+ old = snap(*STILL, rec("R", "CIRCLE", "PIPE", ((0, 0),), radius=10.0))
+ new = snap(*STILL, rec("R", "CIRCLE", "PIPE", ((0, 0),), radius=12.0))
+ (row,) = diff_snapshots(old, new)["changed"]
+ assert row["changes"] == {"radius": {"from": 10.0, "to": 12.0}}
+
+
+def test_the_handle_stage_catches_a_layer_change_the_position_stage_cannot():
+ old = snap(*STILL, line("L", 0, 300, 100, 300, layer="P1"))
+ new = snap(*STILL, line("L", 0, 300, 100, 300, layer="P2"))
+ (row,) = diff_snapshots(old, new)["changed"]
+ assert row["matched_by"] == "handle"
+ assert row["changes"] == {"layer": {"from": "P1", "to": "P2"}}
+
+
+def test_stable_handles_pair_a_far_move_by_handle():
+ old = snap(*STILL, rec("20", "CIRCLE", "PIPE", ((0, 0),), radius=5.0))
+ new = snap(*STILL, rec("20", "CIRCLE", "PIPE", ((5000, 5000),), radius=5.0))
+ (row,) = diff_snapshots(old, new, move_tol=10.0)["changed"]
+ assert row["matched_by"] == "handle"
+ assert row["changes"] == {"moved_by": [pytest.approx(5000.0), pytest.approx(5000.0)]}
+
+
+def test_renumbered_handles_never_pair_strangers_that_share_a_handle():
+ old = snap(*STILL, rec("20", "CIRCLE", "PIPE", ((0, 0),), radius=5.0))
+ new = snap(*RENUMBERED, rec("20", "CIRCLE", "PIPE", ((5000, 5000),), radius=5.0))
+ result = diff_snapshots(old, new, move_tol=10.0)
+ assert result["handles_stable"] is False
+ assert result["changed"] == []
+ assert [r["handle"] for r in result["removed"]] == ["20"]
+ assert [r["at"] for r in result["added"]] == [[5000.0, 5000.0]]
+
+
+def test_beyond_move_tol_a_line_is_removed_and_another_added():
+ old = snap(*STILL, line("D", 0, 200, 100, 200))
+ new = snap(*STILL, line("E", 0, 700, 100, 700))
+ result = diff_snapshots(old, new, move_tol=10.0)
+ assert (result["summary"]["added"], result["summary"]["removed"]) == (1, 1)
+ assert result["changed"] == []
+
+
+# -- the move tolerance ----------------------------------------------------------------
+
+
+def test_the_default_move_tol_is_2_percent_of_the_robust_diagonal():
+ # 100 points on the diagonal of a 2970 x 3960 box, and one stray 10 km away:
+ # the 1st-99th percentile box is (30, 40)-(2970, 3960), diagonal 4900.
+ points = [rec(f"P{i}", "POINT", "PTS", ((i * 30.0, i * 40.0),)) for i in range(100)]
+ stray = rec("FAR", "POINT", "PTS", ((1.0e7, 1.0e7),))
+ drawing = snap(*points, stray)
+ assert robust_box(drawing) == (30.0, 40.0, 2970.0, 3960.0)
+ assert diff_snapshots(drawing, drawing)["move_tol"] == pytest.approx(98.0)
+
+
+def test_a_non_positive_tolerance_is_refused_by_name():
+ with pytest.raises(ValueError, match=r"^tol must be a positive number; got 0"):
+ diff_snapshots(snap(), snap(), tol=0)
+ with pytest.raises(ValueError, match=r"^move_tol must be a positive number; got -1"):
+ diff_snapshots(snap(), snap(), move_tol=-1)
+
+
+# -- the report ------------------------------------------------------------------------
+
+
+def test_changes_are_counted_by_layer_and_type_and_clustered():
+ old = snap(
+ line("U", 0, -500, 100, -500),
+ line("D", 0, 200, 100, 200),
+ rec("T", "TEXT", "NOTES", ((10, 10),), text="OLD", height=2.5),
+ )
+ new = snap(
+ line("U", 0, -500, 100, -500),
+ line("D", 3, 200, 103, 200),
+ rec("T", "TEXT", "NOTES", ((10, 10),), text="NEW", height=2.5),
+ rec("N1", "CIRCLE", "PIPE", ((5000, 5000),), radius=5.0, bbox=(4995, 4995, 5005, 5005)),
+ )
+ result = diff_snapshots(old, new, move_tol=10.0)
+ assert result["by_layer"] == {
+ "NOTES": {"changed": 1, "added": 0, "removed": 0},
+ "PIPE": {"changed": 1, "added": 1, "removed": 0},
+ }
+ assert result["by_type"] == {
+ "CIRCLE": {"changed": 0, "added": 1, "removed": 0},
+ "LINE": {"changed": 1, "added": 0, "removed": 0},
+ "TEXT": {"changed": 1, "added": 0, "removed": 0},
+ }
+ # pad = max(0.25 x 10, 10 x 0.01) = 2.5 around each cluster
+ assert result["clusters"] == [
+ {"space": "Model", "box": [-2.5, 197.5, 105.5, 202.5], "count": 1},
+ {"space": "Model", "box": [7.5, 7.5, 12.5, 12.5], "count": 1},
+ {"space": "Model", "box": [4992.5, 4992.5, 5007.5, 5007.5], "count": 1},
+ ]
+
+
+def test_cap_report_cuts_every_list_and_says_how_much():
+ old = snap(*STILL)
+ new = snap(*STILL, *(line(f"N{i}", 0, 1000 * (i + 1), 100, 1000 * (i + 1)) for i in range(5)))
+ capped = cap_report(diff_snapshots(old, new, move_tol=10.0), limit=2)
+ assert len(capped["added"]) == 2
+ assert capped["truncated"] == {"changed": 0, "added": 3, "removed": 0}
+ assert capped["summary"]["added"] == 5
+
+
+# -- block definitions -----------------------------------------------------------------
+
+
+def _blocks(radius: float, extra: bool):
+ doc = ezdxf.new()
+ doc.blocks.new("B1").add_circle((0, 0), radius)
+ doc.blocks.new("B2").add_line((0, 0), (10, 0))
+ if extra:
+ doc.blocks.new("B3").add_line((0, 0), (0, 10))
+ doc.blocks.new_anonymous_block().add_line((0, 0), (1, 1))
+ return doc
+
+
+def test_block_definitions_are_compared_by_content():
+ old = block_signatures(_blocks(5.0, extra=False))
+ new = block_signatures(_blocks(6.0, extra=True))
+ assert sorted(old) == ["B1", "B2"] # the anonymous *U block is left out
+ assert block_changes(old, new) == {
+ "compared": True,
+ "added": ["B3"],
+ "removed": [],
+ "changed": ["B1"],
+ }
+ assert block_changes(None, new)["compared"] is False
+
+
+# -- reading a revision ----------------------------------------------------------------
+
+
+def test_read_revision_refuses_a_file_over_the_size_limit(tmp_path, monkeypatch):
+ path = tmp_path / "big.dxf"
+ ezdxf.new().saveas(path)
+ monkeypatch.setattr(config.settings, "max_dxf_bytes", 10)
+ with pytest.raises(ValueError, match=r"over the MAX_DXF_BYTES limit of 10"):
+ read_revision(str(path))
+
+
+def test_read_revision_refuses_a_file_that_is_not_a_dxf(tmp_path):
+ path = tmp_path / "notes.dxf"
+ path.write_text("not a drawing\n", encoding="utf-8")
+ with pytest.raises(ValueError, match=r"not a readable DXF"):
+ read_revision(str(path))
+
+
+# -- the known edits on the synthetic plant --------------------------------------------
+
+
+def test_the_known_edits_and_nothing_else(tmp_path):
+ pair = build_revision_pair(tmp_path)
+ old, old_blocks = read_revision(pair["old"])
+ new, new_blocks = read_revision(pair["new"])
+ result = diff_snapshots(old, new, move_tol=pair["move_tol"])
+ total = pair["old_count"]
+ assert result["summary"] == {
+ "old_entities": total,
+ "new_entities": total,
+ "unchanged": total - 4,
+ "changed": 3,
+ "added": 1,
+ "removed": 1,
+ }
+ changed = {row["old_handle"]: row for row in result["changed"]}
+ assert set(changed) == {pair["moved"], pair["text"], pair["insert"]}
+ assert {row["matched_by"] for row in result["changed"]} == {"position"}
+ moved = changed[pair["moved"]]["changes"]
+ assert list(moved) == ["moved_by"]
+ assert moved["moved_by"][0] == pytest.approx(pair["move_tol"] / 2.0, abs=1e-9)
+ assert moved["moved_by"][1] == pytest.approx(0.0, abs=1e-9)
+ assert changed[pair["text"]]["changes"] == {
+ "text": {"from": pair["text_before"], "to": NEW_TEXT}
+ }
+ assert changed[pair["insert"]]["changes"] == {
+ "attribs": [{"tag": "TAG", "from": TAG_BEFORE, "to": TAG_AFTER}]
+ }
+ assert [row["handle"] for row in result["removed"]] == [pair["deleted"]]
+ assert [(row["type"], row["layer"]) for row in result["added"]] == [("CIRCLE", ADDED_LAYER)]
+ assert block_changes(old_blocks, new_blocks) == {
+ "compared": True,
+ "added": [],
+ "removed": [],
+ "changed": [],
+ }
+
+
+def test_renumbered_handles_alone_are_no_change(tmp_path):
+ pair = build_revision_pair(tmp_path, edits=False)
+ old, _ = read_revision(pair["old"])
+ new, _ = read_revision(pair["new"])
+ summary = diff_snapshots(old, new)["summary"]
+ assert (summary["changed"], summary["added"], summary["removed"]) == (0, 0, 0)
+ assert summary["unchanged"] == pair["old_count"]
diff --git a/tests/test_understand_diff_tool.py b/tests/test_understand_diff_tool.py
new file mode 100644
index 0000000..2f718c9
--- /dev/null
+++ b/tests/test_understand_diff_tool.py
@@ -0,0 +1,130 @@
+"""drawing_diff over the MCP wire, on the headless engine, against the synthetic revisions."""
+
+from __future__ import annotations
+
+import pytest
+import pytest_asyncio
+from fastmcp import Client
+
+import server
+from engineering.sheet.frames import SHEET_LAYER
+from tests.fixtures.revisions import build_revision_pair
+
+pytestmark = pytest.mark.asyncio
+
+
+@pytest_asyncio.fixture
+async def client(monkeypatch):
+ """In-memory client on a fresh headless document - never the operator's AutoCAD."""
+ monkeypatch.setenv("AUTOCAD_MCP_BACKEND", "ezdxf")
+ async with Client(server.mcp) as connected:
+ status = (await connected.call_tool("system_status", {})).structured_content or {}
+ assert status.get("backend") == "ezdxf", "these tests must never touch live AutoCAD"
+ await connected.call_tool("drawing_new", {})
+ yield connected
+
+
+async def _total(client) -> int:
+ stats = (await client.call_tool("analysis_entity_stats", {})).structured_content
+ return stats["total_entities"]
+
+
+async def test_the_current_document_against_its_older_revision(client, tmp_path):
+ pair = build_revision_pair(tmp_path)
+ await client.call_tool("drawing_open", {"path": pair["new"]})
+ before = await _total(client)
+ result = (
+ await client.call_tool(
+ "drawing_diff", {"old_path": pair["old"], "move_tol": pair["move_tol"]}
+ )
+ ).structured_content
+ assert (result["summary"]["changed"], result["summary"]["added"]) == (3, 1)
+ assert result["summary"]["removed"] == 1
+ assert {row["old_handle"] for row in result["changed"]} == {
+ pair["moved"],
+ pair["text"],
+ pair["insert"],
+ }
+ assert result["blocks"]["compared"] is True and result["blocks"]["changed"] == []
+ assert result["truncated"] == {"changed": 0, "added": 0, "removed": 0}
+ assert await _total(client) == before, "a diff without markup writes nothing"
+
+
+async def test_two_paths_and_the_list_cap(client, tmp_path):
+ pair = build_revision_pair(tmp_path)
+ result = (
+ await client.call_tool(
+ "drawing_diff",
+ {
+ "old_path": pair["old"],
+ "new_path": pair["new"],
+ "move_tol": pair["move_tol"],
+ "limit": 1,
+ },
+ )
+ ).structured_content
+ assert len(result["changed"]) == 1
+ assert result["truncated"]["changed"] == 2
+ assert result["summary"]["changed"] == 3
+
+
+async def test_markup_clouds_every_model_space_cluster(client, tmp_path):
+ pair = build_revision_pair(tmp_path)
+ await client.call_tool("drawing_open", {"path": pair["new"]})
+ result = (
+ await client.call_tool(
+ "drawing_diff",
+ {"old_path": pair["old"], "move_tol": pair["move_tol"], "markup": True, "rev": "b"},
+ )
+ ).structured_content
+ model = [c for c in result["clusters"] if c["space"] == "Model"]
+ markup = result["markup"]
+ assert markup["created"] is True and markup["rev"] == "B"
+ assert len(markup["clouds"]) == len(model) >= 1
+ for handle in markup["clouds"]:
+ cloud = (await client.call_tool("entity_get", {"handle": handle})).structured_content
+ assert (cloud["type"], cloud["layer"]) == ("LWPOLYLINE", SHEET_LAYER)
+
+
+@pytest.mark.parametrize(
+ ("arguments", "fragment"),
+ [
+ (
+ {"markup": True, "rev": "B", "new_path": "NEW"},
+ "markup draws revision clouds on the current",
+ ),
+ ({"markup": True}, "markup needs `rev`"),
+ ({"markup": True, "rev": "TOO-LONG"}, "markup needs `rev`"),
+ ],
+)
+async def test_markup_refusals_come_before_any_read(client, tmp_path, arguments, fragment):
+ pair = build_revision_pair(tmp_path)
+ if arguments.get("new_path") == "NEW":
+ arguments = {**arguments, "new_path": pair["new"]}
+ result = await client.call_tool(
+ "drawing_diff", {"old_path": pair["old"], **arguments}, raise_on_error=False
+ )
+ assert result.is_error is True
+ assert fragment in result.content[0].text
+
+
+async def test_a_missing_revision_is_refused_by_parameter(client, tmp_path):
+ result = await client.call_tool(
+ "drawing_diff", {"old_path": str(tmp_path / "nowhere.dxf")}, raise_on_error=False
+ )
+ assert result.is_error is True
+ assert "old_path:" in result.content[0].text and "does not exist" in result.content[0].text
+
+
+async def test_a_revision_that_is_neither_dxf_nor_dwg_is_refused_by_parameter(client, tmp_path):
+ notes = tmp_path / "notes.txt"
+ notes.write_text("not a drawing\n", encoding="utf-8")
+ result = await client.call_tool("drawing_diff", {"old_path": str(notes)}, raise_on_error=False)
+ assert result.is_error is True
+ text = result.content[0].text
+ assert "old_path:" in text and "reads a .dxf or a .dwg" in text
+
+
+async def test_the_diff_is_advertised_as_non_destructive(client):
+ tools = {tool.name: tool for tool in await client.list_tools()}
+ assert tools["drawing_diff"].annotations.destructiveHint is False