Files
houseplan-card/custom_components/houseplan/junction_limits.py
T
Codex c90f5bf052 perf: junction limits scale — executor, rev cache, linear П3/П4, shared masonry pass (#330)
Six cuts, zero verdict changes (spec §3; equivalence pinned by units, the
parity suite and the smokes):

- §4.1 the CPU chain of ws_config_set and ws_plan_optimize runs in the
  executor; write_lock still serialises writes, only the HA event loop is
  freed (2.8 s of blocking per 576-atom write before).
- §4.2 the stored document's violation counts are cached on the runtime by
  rev (store.py junction_baseline); a repeated write never re-judges
  `previous`. validate_junction_limits takes baseline_counts and returns the
  candidate's counts to cache after a successful save.
- §4.3 П3 builds its node index once per check in both mirrors
  (289→11 ms TS, 285→~50 ms py).
- §4.5 П4 uses a bucket grid with the threshold as cell size in both
  mirrors (104→19 ms TS, 372→44 ms py); pair enumeration switches to
  lexicographic order — same verdict set, equivalence pinned against a
  brute-force oracle on cell borders.
- §4.6 a document already carrying the current catalogue is judged as-is:
  a no-op re-migration cost 815 ms py / 69 ms TS. Legacy documents migrate
  exactly as before (the #329 H1 test stays green).
- §4.7 П5 shares one junction-topology pass per check and pays the masonry
  union only when multi-wall nodes exist — and the resize path hands over
  the preflight's own artifact, so a pointermove never builds the union
  twice (4.2 s → 88 ms full candidate on the benchmark grid).

The frontend baseline is cached per (document identity, config epoch): ten
pointermoves make N+1 limit computations, not 2N — pinned by the smoke on a
real pointer gesture.

demo/benchmark_junction_limits.mjs (npm run benchmark:junction-limits) pins
the budgets for both mirrors: TS full candidate ≤100 ms (measured 88), py
warm validate ≤250 ms (measured 45), cold legacy ≤3.5 s — that path is
one-off and lives in the executor.

Issue: #330
User-Visible: yes
2026-08-28 03:07:11 +03:00

391 lines
16 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""Issue #329 — wall junction limits, Python mirror of src/junction-limits.ts.
The card refuses a WRITE that would ADD a junction violation; this module is
the backend half of that contract, so a stale or hostile client cannot post
what the editor refuses. It mirrors П1–П4 exactly, including the two legal
incidences of П4 (a shared node and a T-joint into the middle of a foreign
wall) and the П3 rule that measures the collinear same-thickness WALL RUN
rather than one catalogue atom.
П5 (the room keeps at least 25 cm² of interior) is deliberately NOT mirrored:
it is a statement about the rendered wall bodies, and reproducing the mitre /
inset pipeline in Python would be a second geometry implementation whose drift
is a larger risk than the rule it guards. A document that violates only П5 is
ugly, not corrupt — see docs/specs/329-junction-limits.md §5.
"""
from __future__ import annotations
import math
from .wall_segment_model import WALL_SEGMENT_MODEL_VERSION, commit_wall_segment_model
MIN_JUNCTION_ANGLE_DEG = 15.0
MAX_JUNCTION_VALENCE = 6
MIN_SEGMENT_LENGTH_CM = 20.0
MIN_NODE_DISTANCE_CM = 5.0
GRID_STEP_N = 1 / 240
_EPS = 1e-9
# Below this a node is ON the wall (T-joint), not near it.
_INCIDENT_EPS = 1e-9
class JunctionLimitError(ValueError):
"""A write introduced a wall-junction violation (#329)."""
def __init__(self, space_id: str, rule: str, subject: str,
actual: float, limit: float) -> None:
self.space_id = space_id
self.rule = rule
self.subject = subject
self.actual = actual
self.limit = limit
self.code = f"junction_limit_{rule}"
super().__init__(
f"space={space_id}; rule={rule}; subject={subject}; "
f"actual={actual:.12g}; limit={limit:.12g}"
)
def _finite_point(point: object) -> bool:
return (
isinstance(point, (list, tuple))
and len(point) >= 2
and all(isinstance(value, (int, float)) and math.isfinite(value)
for value in point[:2])
)
def _key(point) -> str:
return f"{point[0]:.6f},{point[1]:.6f}"
def _length(a, b) -> float:
return math.hypot(b[0] - a[0], b[1] - a[1])
def cm_to_units(cm: float, cell_cm: float, grid_pitch: float = GRID_STEP_N) -> float:
return (cm / (cell_cm or 1)) * grid_pitch
def limit_segments(space: dict) -> list[dict]:
"""Every wall the limits judge: contour atoms, partitions, draft segments."""
segments: list[dict] = []
for segment in space.get("wall_segments") or []:
if _finite_point(segment.get("a")) and _finite_point(segment.get("b")):
segments.append({
"id": str(segment.get("id") or ""),
"a": segment["a"], "b": segment["b"],
"cm": float(segment.get("cm") or 0),
})
for partition in space.get("partitions") or []:
if _finite_point(partition.get("a")) and _finite_point(partition.get("b")):
segments.append({
"id": str(partition.get("id") or ""),
"a": partition["a"], "b": partition["b"],
"cm": float(partition.get("cm") or 0),
})
for draft in space.get("room_drafts") or []:
points = draft.get("points") or []
drafted = draft.get("segments") or []
for index in range(max(len(points) - 1, 0)):
if not (_finite_point(points[index]) and _finite_point(points[index + 1])):
continue
piece = drafted[index] if index < len(drafted) else {}
segments.append({
"id": str((piece or {}).get("id")
or f"{draft.get('id') or 'draft'}-{index}"),
"a": points[index], "b": points[index + 1],
"cm": float((piece or {}).get("cm") or 0),
})
return [segment for segment in segments
if _length(segment["a"], segment["b"]) > _EPS]
def check_nodes(segments: list[dict]) -> list[tuple[str, str, float, float]]:
"""П1 + П2: valence of a node and the narrowest wedge in it."""
rays: dict[str, list[float]] = {}
for segment in segments:
for start, end in ((segment["a"], segment["b"]), (segment["b"], segment["a"])):
rays.setdefault(_key(start), []).append(
math.atan2(end[1] - start[1], end[0] - start[0])
)
violations: list[tuple[str, str, float, float]] = []
for node, angles in rays.items():
if len(angles) > MAX_JUNCTION_VALENCE:
violations.append(
("valence", node, float(len(angles)), float(MAX_JUNCTION_VALENCE))
)
if len(angles) < 2:
continue
ordered = sorted(angles)
smallest = math.inf
for index, value in enumerate(ordered):
delta = ordered[(index + 1) % len(ordered)] - value
if index == len(ordered) - 1:
delta += math.pi * 2
degrees = (delta * 180) / math.pi
# Collinear rays of one straight wall are a 180° pair, not a wedge.
if _EPS < degrees < smallest:
smallest = degrees
if smallest < MIN_JUNCTION_ANGLE_DEG - 1e-9:
violations.append(("angle", node, smallest, MIN_JUNCTION_ANGLE_DEG))
return violations
def _axis_degrees(segment: dict) -> float:
degrees = math.degrees(math.atan2(
segment["b"][1] - segment["a"][1], segment["b"][0] - segment["a"][0]
))
return (degrees % 180 + 180) % 180
def _collinear(left: dict, right: dict, tolerance_deg: float = 1.0) -> bool:
delta = abs(_axis_degrees(left) - _axis_degrees(right))
return min(delta, 180 - delta) <= tolerance_deg
def _build_node_index(segments: list[dict]) -> dict[str, list[dict]]:
by_node: dict[str, list[dict]] = {}
for item in segments:
for point in (item["a"], item["b"]):
by_node.setdefault(_key(point), []).append(item)
return by_node
def collinear_run_length_units(
segment: dict, segments: list[dict],
by_node_index: dict[str, list[dict]] | None = None,
) -> float:
"""Length of the WALL a segment belongs to, not of the atom.
Atomisation at a thickness step leaves pieces nobody drew — where a 30 cm
wall meets a 20 cm one, a (30−20)/2 = 5 cm piece continues the same wall.
They are collinear continuations at the same thickness, so П3 measures the
maximal collinear chain through the segment's nodes.
#330 §4.3: building the index per SEGMENT made П3 quadratic (285 ms on
576 atoms); the per-check caller builds it once and passes it in.
"""
by_node = by_node_index if by_node_index is not None else _build_node_index(segments)
visited = [segment]
total = _length(segment["a"], segment["b"])
def walk(current: dict, node: str) -> None:
nonlocal total
for candidate in by_node.get(node, []):
if any(candidate is seen for seen in visited):
continue
if not _collinear(candidate, current):
continue
if float(candidate.get("cm") or 0) != float(current.get("cm") or 0):
continue
visited.append(candidate)
total += _length(candidate["a"], candidate["b"])
walk(candidate, _key(candidate["b"])
if _key(candidate["a"]) == node else _key(candidate["a"]))
return
walk(segment, _key(segment["a"]))
walk(segment, _key(segment["b"]))
return total
def check_segment_lengths(
segments: list[dict], cell_cm: float, grid_pitch: float = GRID_STEP_N,
) -> list[tuple[str, str, float, float]]:
"""П3: a wall is at least 20 cm and never shorter than its own thickness."""
violations: list[tuple[str, str, float, float]] = []
by_node = _build_node_index(segments)
for segment in segments:
units = collinear_run_length_units(segment, segments, by_node)
cm = (units / grid_pitch) * (cell_cm or 1)
thickness = float(segment.get("cm") or 0)
limit = max(MIN_SEGMENT_LENGTH_CM, thickness if thickness > 0 else 0.0)
if cm < limit - 1e-9:
violations.append((
"length", str(segment.get("id") or _key(segment["a"])), cm, limit,
))
return violations
def _distance_to_segment(point, a, b) -> float:
dx, dy = b[0] - a[0], b[1] - a[1]
length_sq = dx * dx + dy * dy
t = 0.0 if length_sq <= _EPS else max(0.0, min(1.0, (
(point[0] - a[0]) * dx + (point[1] - a[1]) * dy
) / length_sq))
return math.hypot(point[0] - (a[0] + dx * t), point[1] - (a[1] + dy * t))
def check_node_distances(
segments: list[dict], cell_cm: float, grid_pitch: float = GRID_STEP_N,
) -> list[tuple[str, str, float, float]]:
"""П4: non-incident nodes and node-to-foreign-wall clearance.
#330 §4.5: the all-pairs form cost 372 ms on 576 atoms. Nodes and padded
segment bounding boxes are hashed into a grid with the threshold as cell
size; each node is compared only against its 9-cell neighbourhood.
Verdicts are identical — pinned by the equivalence unit and the parity
suite. Pair order inside the subject follows the lexicographic key order,
which replaces the i<j of the all-pairs loop.
"""
nodes: dict[str, list] = {}
for segment in segments:
nodes[_key(segment["a"])] = segment["a"]
nodes[_key(segment["b"])] = segment["b"]
min_units = cm_to_units(MIN_NODE_DISTANCE_CM, cell_cm, grid_pitch)
size = min_units if min_units > _EPS else 1.0
node_grid: dict[tuple[int, int], list] = {}
for node_key, point in nodes.items():
node_grid.setdefault(
(int(point[0] // size), int(point[1] // size)), []
).append((node_key, point))
segment_grid: dict[tuple[int, int], list] = {}
for segment in segments:
x0 = min(segment["a"][0], segment["b"][0]) - min_units
x1 = max(segment["a"][0], segment["b"][0]) + min_units
y0 = min(segment["a"][1], segment["b"][1]) - min_units
y1 = max(segment["a"][1], segment["b"][1]) + min_units
for cx in range(int(x0 // size), int(x1 // size) + 1):
for cy in range(int(y0 // size), int(y1 // size) + 1):
segment_grid.setdefault((cx, cy), []).append(segment)
violations: list[tuple[str, str, float, float]] = []
for node_key, point in nodes.items():
cx, cy = int(point[0] // size), int(point[1] // size)
for dx in (-1, 0, 1):
for dy in (-1, 0, 1):
for other_key, other in node_grid.get((cx + dx, cy + dy), ()):
if node_key >= other_key:
continue
distance = _length(point, other)
if distance < min_units - 1e-9:
violations.append((
"distance", f"{node_key} ↔ {other_key}",
(distance / grid_pitch) * (cell_cm or 1),
MIN_NODE_DISTANCE_CM,
))
for segment in segment_grid.get((cx, cy), ()):
# A node that belongs to the wall is a legal corner or T-joint.
if _key(segment["a"]) == node_key or _key(segment["b"]) == node_key:
continue
distance = _distance_to_segment(point, segment["a"], segment["b"])
# Sitting exactly ON the wall is the other legal incidence.
if distance <= _INCIDENT_EPS:
continue
if distance < min_units - 1e-9:
violations.append((
"distance",
f"{node_key} → {segment.get('id') or _key(segment['a'])}",
(distance / grid_pitch) * (cell_cm or 1),
MIN_NODE_DISTANCE_CM,
))
return violations
def space_violations(space: dict) -> list[tuple[str, str, float, float]]:
"""П1–П4 over one space, in the frontend's order."""
segments = limit_segments(space)
cell_cm = float(space.get("cell_cm") or 1)
return [
*check_nodes(segments),
*check_segment_lengths(segments, cell_cm),
*check_node_distances(segments, cell_cm),
]
def _migrated_spaces(config: dict | None) -> dict[str, dict]:
"""Spaces of one document AFTER the wall-segment migration, by id.
The limits read `wall_segments`, so a document that predates the catalogue
reports NO walls at all — a legacy space would answer "no violations"
regardless of its geometry. Comparing such a baseline against a candidate
the client already migrated counts every inherited violation as new and
refuses an unrelated edit (spec §3 forbids exactly that). Both sides are
therefore judged after the SAME migration, mirroring the frontend barrier.
A document that cannot be migrated is not a reason to refuse the write:
the wall-model barrier owns that verdict and reports it with its own code.
Here it simply means there is no baseline to inherit from.
"""
if not isinstance(config, dict):
return {}
# #330 §4.6: a document that already carries the current catalogue is
# used as-is. The migration's purpose here is to never compare a
# catalogue-less legacy baseline against a migrated candidate; a
# current-version document has the catalogue, so the purpose is met with
# zero work — while a no-op re-migration costs 0.8 s of _atomize on a
# 576-atom plan. Equivalence "as-is == migrated" for current-version
# documents is pinned by a parity case.
if int(config.get("model_version") or 0) >= WALL_SEGMENT_MODEL_VERSION:
migrated = config
else:
try:
migrated, _ = commit_wall_segment_model(config)
except Exception: # noqa: BLE001 — see the docstring: not our verdict
migrated = config
return {
str(space.get("id")): space
for space in (migrated or {}).get("spaces") or []
if isinstance(space, dict)
}
def space_violation_counts(spaces: dict[str, dict]) -> dict[str, dict[str, int]]:
"""Violation counts per space per rule — the shape the barrier consumes
and the rev cache (#330 §4.2) stores. Documents are NOT retained."""
result: dict[str, dict[str, int]] = {}
for space_id, space in spaces.items():
counts: dict[str, int] = {}
for rule, _subject, _actual, _limit in space_violations(space):
counts[rule] = counts.get(rule, 0) + 1
result[space_id] = counts
return result
def validate_junction_limits(
config: dict, previous: dict | None = None, *,
baseline_counts: dict[str, dict[str, int]] | None = None,
) -> dict[str, dict[str, int]]:
"""Refuse a write that ADDS a junction violation; inherit the rest.
Counted per RULE, not per subject: a structural write re-atomises and
re-keys contour segments, so subject identity is not stable across the
barrier and matching by it would report an inherited violation as new
(the mistake that refused legitimate resizes on the frontend). Spec §3:
a write may keep existing violations, it may never add one.
A LEGACY document still goes through `commit_wall_segment_model` first,
so a catalogue-less baseline is compared in the same terms as the
candidate; a current-version document is used as-is (#330 §4.6).
`baseline_counts` (#330 §4.2) replaces the previous document entirely:
the caller caches the counts of the stored document by config_rev, so a
repeated write does not even touch `previous`. Returns the candidate's
counts — the value to cache after a successful write.
"""
if baseline_counts is not None:
old_counts = baseline_counts
else:
old_counts = space_violation_counts(_migrated_spaces(previous))
new_spaces = _migrated_spaces(config)
candidate_counts: dict[str, dict[str, int]] = {}
for space_id, space in new_spaces.items():
# A brand-new space has nothing to inherit from — but neither is a
# first write allowed to arrive already broken.
before = old_counts.get(space_id) or {}
after: dict[str, list[tuple[str, str, float, float]]] = {}
for violation in space_violations(space):
after.setdefault(violation[0], []).append(violation)
candidate_counts[space_id] = {
rule: len(items) for rule, items in after.items()
}
for rule, items in after.items():
if len(items) > before.get(rule, 0):
_, subject, actual, limit = items[0]
raise JunctionLimitError(space_id, rule, subject, actual, limit)
return candidate_counts