gjm
1e5bff268a
新增 MCP 工具 set_drawing_area(mode="max"|"restore")：最大化主窗口 + CLEANSCREENON （隐藏功能区/命令行/面板）。实测绘图区 2110x462 -> 3310x1084（面积 3.68x）， restore 精确回位（窗口矩形/最大化态/绘图区尺寸全部还原）。 坑：CLEANSCREENOFF 是异步命令，AutoCAD 会按「进入干净屏幕时的布局」重设窗口几何， 把同步的 SetWindowPlacement 覆盖掉（表现为窗口停在最大化）。解法是插件内 SetTimer 延迟 700ms 再压回；实测窗口在 0.5~1.0s 之间收敛，故调用方等待取 1.5s。 取证层： - 新增 maximized_drawing_area(ctx) 上下文管理器，__main__ 用它包住整个取证循环 （多图框只进出一次）；受 DRAWING_AREA_MAXIMIZE 控制，工具缺失/调用失败静默降级， restore 失败不掩盖业务异常。 - 内容过滤统一到规划与回退两条路径，共用同一谓词；过滤后为空则退回原集合。 - 判定改用「实体 bbox」而非「占用格」：格子量化会把跨界内容算进相邻格，反而更粗 （实测实体 bbox 正好压象限边界时 4 个象限全判有内容）。heatmap.occupied_boxes 随之删除。 实测（真实图纸 A0/A1/A2 三图框）：截图张数 12/8/4 -> 8/5/4；pytest 296 passed。 新增 verify_drawing_area.py 作为该工具的验收脚本。
306 lines
12 KiB
Python
306 lines
12 KiB
Python
"""阶段 2：逐审查单元取证（多源证据层 + 视觉多图）。
|
||
|
||
对每个审查单元：取结构化数据（实体/文字/标注/块/多段线）+ 视觉截图（整体 + 可选局部）。
|
||
结构化数据是权威依据，截图仅作视觉上下文。
|
||
|
||
多源设计：`collect()` 按审查单元来源分发；当前仅实现 CAD 源，PDF / 图片源预留 seam
|
||
（未来在对应分支产出**同形** evidence，下游无需改动）。
|
||
"""
|
||
|
||
import contextlib
|
||
import math
|
||
import time
|
||
|
||
from .. import config, evidence_model, heatmap, tiling
|
||
from ..llm_client import image_size_b64
|
||
from ..mcp_client import result_images, result_json
|
||
|
||
_LABEL_OVERALL = "overall"
|
||
|
||
# 绘图区最大化用的工具名（旧版插件没有此工具 → 静默降级）
|
||
_TOOL_DRAWING_AREA = "set_drawing_area"
|
||
|
||
|
||
@contextlib.contextmanager
|
||
def maximized_drawing_area(ctx):
|
||
"""截屏前最大化绘图区（窗口铺满屏幕 + 干净屏幕），退出时还原。
|
||
|
||
实测绘图区像素 2110x462 → 3310x1084（3.68x）；在同样的长边下采样预算下，
|
||
有效像素更多、细节更清楚，自适应瓦片张数也更少。
|
||
|
||
- 受 `config.DRAWING_AREA_MAXIMIZE` 控制；
|
||
- 工具不存在（旧版插件）/ 调用失败 → **静默降级**：照常取证，只是绘图区小一些；
|
||
- 退出时一定尝试还原；**还原失败不掩盖业务异常**（不 raise）。
|
||
"""
|
||
if not config.DRAWING_AREA_MAXIMIZE:
|
||
yield
|
||
return
|
||
|
||
entered = False
|
||
try:
|
||
if ctx.tools.has(_TOOL_DRAWING_AREA):
|
||
ctx.tools.call(_TOOL_DRAWING_AREA, {"mode": "max"})
|
||
entered = True
|
||
time.sleep(config.DRAWING_AREA_ENTER_WAIT)
|
||
except Exception: # noqa: BLE001 - 界面优化失败不应中断审查
|
||
entered = False
|
||
|
||
try:
|
||
yield
|
||
finally:
|
||
if entered:
|
||
try:
|
||
ctx.tools.call(_TOOL_DRAWING_AREA, {"mode": "restore"})
|
||
time.sleep(config.DRAWING_AREA_RESTORE_WAIT)
|
||
except Exception: # noqa: BLE001
|
||
pass
|
||
|
||
|
||
def bbox_to_window(bbox):
|
||
"""[minx,miny,maxx,maxy] -> window 对象。"""
|
||
return {
|
||
"min_x": float(bbox[0]),
|
||
"min_y": float(bbox[1]),
|
||
"max_x": float(bbox[2]),
|
||
"max_y": float(bbox[3]),
|
||
}
|
||
|
||
|
||
def sub_windows(window, tiles):
|
||
"""把 window 等分为 tiles 个子窗口（tiles=4 → 2x2 象限）。
|
||
|
||
tiles<=1 返回 []。返回 [{"min_x",..,"max_x",..}, ...]，行优先（左上→右上→左下→右下）。
|
||
"""
|
||
if not tiles or tiles <= 1:
|
||
return []
|
||
cols = int(math.ceil(math.sqrt(tiles)))
|
||
rows = int(math.ceil(tiles / cols))
|
||
min_x, min_y = window["min_x"], window["min_y"]
|
||
max_x, max_y = window["max_x"], window["max_y"]
|
||
w = (max_x - min_x) / cols
|
||
h = (max_y - min_y) / rows
|
||
out = []
|
||
for r in range(rows):
|
||
for c in range(cols):
|
||
if len(out) >= tiles:
|
||
break
|
||
out.append(
|
||
{
|
||
"min_x": min_x + c * w,
|
||
"min_y": max_y - (r + 1) * h,
|
||
"max_x": min_x + (c + 1) * w,
|
||
"max_y": max_y - r * h,
|
||
}
|
||
)
|
||
return out
|
||
|
||
|
||
def _zoom_and_shot(ctx, window, margin=0.02):
|
||
"""缩放到窗口、等待渲染、截一张图，返回 (b64, mime) 或 None。"""
|
||
ctx.tools.call("zoom_to_window", dict(window, margin=margin))
|
||
time.sleep(config.RENDER_WAIT)
|
||
shot = ctx.tools.call("get_viewport_screenshot", {})
|
||
images = result_images(shot)
|
||
if not images:
|
||
return None
|
||
return images[0]
|
||
|
||
|
||
def _is_truncated(*envelopes):
|
||
"""任一信封 `total > len(items)` 即视为**被截断**（数据不全）。"""
|
||
for env in envelopes:
|
||
if not isinstance(env, dict):
|
||
continue
|
||
items = env.get("items")
|
||
if not isinstance(items, list):
|
||
continue
|
||
try:
|
||
total = int(env.get("total") or len(items))
|
||
except (TypeError, ValueError):
|
||
total = len(items)
|
||
if total > len(items):
|
||
return True
|
||
return False
|
||
|
||
|
||
def _text_boxes(entities, texts, dimensions):
|
||
"""收集「文字类」实体 bbox。
|
||
|
||
标准文字 / 标注信封（`get_text_content` / `get_dimensions`）全收；
|
||
再加**类名含 text/label/dim 等关键词**的自定义实体（`tiling.is_text_type`）。
|
||
"""
|
||
boxes = list(heatmap.entity_boxes(texts))
|
||
boxes.extend(heatmap.entity_boxes(dimensions))
|
||
items = entities.get("items") if isinstance(entities, dict) else None
|
||
for it in items or []:
|
||
if isinstance(it, dict) and tiling.is_text_type(it.get("type")):
|
||
b = heatmap.norm_box(it.get("bbox"))
|
||
if b is not None:
|
||
boxes.append(b)
|
||
return boxes
|
||
|
||
|
||
def _content_predicate(boxes):
|
||
"""构造「该窗口内是否有实体」谓词；无实体返回 None。
|
||
|
||
用于剔除空白瓦片：簇 / 象限的 bbox 只是**包围盒**，L 形或斜向分布会含大片空白。
|
||
这里直接按**实体 bbox** 判断——比先量化成占用格更精确（格子的跨界膨胀会把
|
||
相邻空白瓦片也判成有内容）。接触即算有内容（保守，宁可多拍不漏）。
|
||
"""
|
||
boxes = [b for b in (boxes or []) if b]
|
||
if not boxes:
|
||
return None
|
||
|
||
def _has_content(win):
|
||
box = (win["min_x"], win["min_y"], win["max_x"], win["max_y"])
|
||
return any(tiling.boxes_intersect(box, b) for b in boxes)
|
||
|
||
return _has_content
|
||
|
||
|
||
def _apply_keep(regions, keep):
|
||
"""按 keep 过滤窗口；过滤后为空则**退回原集合**（宁可多拍也不漏）。"""
|
||
if keep is None or not regions:
|
||
return regions
|
||
kept = [w for w in regions if keep(w)]
|
||
return kept or regions
|
||
|
||
|
||
def _plan_tiles(window, frame, boxes, text_boxes, truncated, view_px):
|
||
"""尝试按实体分布规划瓦片窗口；任何条件不满足都返回 []（调用方退回固定策略）。
|
||
|
||
降级条件：规划被关闭 / 实体数据被截断 / 缺图框 / 无实体 / 拿不到绘图区像素 /
|
||
比例未知（`frame_scale` 返 0）。
|
||
"""
|
||
if not config.TILING_ENABLED or truncated or not frame or not boxes or not view_px:
|
||
return []
|
||
scale = tiling.frame_scale(frame)
|
||
s_text = tiling.world_per_px(scale, config.TEXT_HEIGHT_MM, config.TEXT_READABLE_PX)
|
||
# 线条区只需看清走向：同以「最小关注特征尺寸」（纸面 TEXT_HEIGHT_MM）为基准，
|
||
# 但像素门槛更低（R_line < R_text）→ 覆盖范围更大、瓦片更少。
|
||
s_line = tiling.world_per_px(scale, config.TEXT_HEIGHT_MM, config.LINE_READABLE_PX)
|
||
if s_text <= 0 or s_line <= 0:
|
||
return []
|
||
eff = tiling.effective_px(view_px, config.IMAGE_MAX_EDGE)
|
||
if eff[0] <= 0 or eff[1] <= 0:
|
||
return []
|
||
|
||
cell = heatmap.div_to_cell(window, config.HEATMAP_CELL_DIV)
|
||
grid = heatmap.build_grid(window, boxes, cell)
|
||
clusters = heatmap.cluster_cells(grid, dilate=config.HEATMAP_CLUSTER_DILATE)
|
||
if not clusters:
|
||
return []
|
||
flags = tiling.cluster_text_flags(clusters, text_boxes)
|
||
text_idx = {i for i, is_text in enumerate(flags) if is_text}
|
||
|
||
return tiling.plan_regions(
|
||
frame, clusters,
|
||
text_clusters=text_idx, s_text=s_text, s_line=s_line, view_px=eff,
|
||
max_tiles=config.TILE_MAX_COUNT, overlap=config.TILE_OVERLAP,
|
||
keep=_content_predicate(boxes),
|
||
)
|
||
|
||
|
||
def capture_images(ctx, window, *, tiles=0, max_images=None, frame=None,
|
||
boxes=None, text_boxes=None, truncated=False):
|
||
"""采集视觉证据：整体一张 + 局部若干张（自适应瓦片优先，否则等分象限回退）。
|
||
|
||
- 整体 label="overall"，局部 label="tile-1"..；行优先。
|
||
- 局部窗口优先由 `_plan_tiles` 按**实体分布**给出（空白区不出图）；
|
||
不可用时退回 `sub_windows(window, tiles)`，**同样按内容过滤**空白象限
|
||
（过滤后为空则退回未过滤集合，保证不漏拍）。
|
||
- 结束时**复位到整体视口**，保证后续流程从整幅视图起手。
|
||
- 受 max_images（默认 config.EVIDENCE_MAX_IMAGES）截断。
|
||
"""
|
||
if max_images is None:
|
||
max_images = config.EVIDENCE_MAX_IMAGES
|
||
images = []
|
||
|
||
def _add(shot, label):
|
||
if shot and len(images) < max_images:
|
||
images.append(evidence_model.make_image(shot[0], shot[1], label))
|
||
|
||
# 整体图：既是证据，也用于测量绘图区**实际像素**（决定单张截图的世界覆盖范围）
|
||
overall = _zoom_and_shot(ctx, window)
|
||
view_px = image_size_b64(overall[0]) if overall else None
|
||
_add(overall, _LABEL_OVERALL)
|
||
|
||
# 内容过滤：空白区不出图。按**实体 bbox** 判定，规划与回退两条路径共用同一谓词；
|
||
# 实体数据被截断时不可信，放弃过滤（宁可多拍也不漏）。
|
||
keep = None if truncated else _content_predicate(boxes)
|
||
|
||
# 局部：优先自适应瓦片，失败退回等分象限（回退路径同样按内容过滤）
|
||
regions = _plan_tiles(window, frame, boxes, text_boxes, truncated, view_px)
|
||
if not regions:
|
||
regions = _apply_keep(sub_windows(window, tiles), keep)
|
||
|
||
for i, sub in enumerate(regions, 1):
|
||
if len(images) >= max_images:
|
||
break
|
||
_add(_zoom_and_shot(ctx, sub), f"tile-{i}")
|
||
|
||
# 复位整体视口（不截图）
|
||
ctx.tools.call("zoom_to_window", dict(window, margin=0.02))
|
||
return images
|
||
|
||
|
||
def collect_frame(ctx, index, frame):
|
||
"""采集单个图框（CAD 源）的证据；bbox 无效时返回 None。"""
|
||
bbox = frame.get("bbox")
|
||
if not bbox or len(bbox) != 4:
|
||
return None
|
||
|
||
win = bbox_to_window(bbox)
|
||
limit = ctx.frame_limit
|
||
|
||
# a. 结构化数据（权威；按 window 查询，不依赖缩放）
|
||
entities = result_json(ctx.tools.call("query_entities", {"window": win, "limit": limit}), {})
|
||
texts = result_json(ctx.tools.call("get_text_content", {"window": win, "limit": limit}), {})
|
||
dimensions = result_json(ctx.tools.call("get_dimensions", {"window": win, "limit": limit}), {})
|
||
blocks = result_json(
|
||
ctx.tools.call("get_block_references", {"window": win, "limit": limit}), {}
|
||
)
|
||
polylines = result_json(
|
||
ctx.tools.call("get_polylines_info", {"window": win, "limit": limit}), {}
|
||
)
|
||
|
||
# b. 视觉：整体 + 局部（优先按实体分布自适应瓦片；数据截断则退回固定策略）
|
||
truncated = _is_truncated(entities, texts, dimensions, blocks, polylines)
|
||
boxes = heatmap.drop_oversized(
|
||
heatmap.entity_boxes(entities), win, config.HEATMAP_MAX_AREA_FRAC
|
||
)
|
||
images = capture_images(
|
||
ctx, win,
|
||
tiles=config.EVIDENCE_DETAIL_TILES,
|
||
frame=frame,
|
||
boxes=boxes,
|
||
text_boxes=_text_boxes(entities, texts, dimensions),
|
||
truncated=truncated,
|
||
)
|
||
|
||
return {
|
||
"source_type": evidence_model.source_type_of(frame),
|
||
"index": index,
|
||
"frame": frame,
|
||
"window": win,
|
||
"entities": entities,
|
||
"texts": texts,
|
||
"dimensions": dimensions,
|
||
"blocks": blocks,
|
||
"polylines": polylines,
|
||
"images": images,
|
||
}
|
||
|
||
|
||
def collect(ctx, index, unit):
|
||
"""按审查单元来源分发取证。
|
||
|
||
当前仅实现 CAD 源；PDF / 图片源预留（未来在对应分支接入离线解析 / 直接读图，
|
||
产出与 CAD **同形**的 evidence：window=None、结构化键给空信封、images 带 label）。
|
||
"""
|
||
source = evidence_model.source_type_of(unit)
|
||
if source == evidence_model.SOURCE_CAD:
|
||
return collect_frame(ctx, index, unit)
|
||
# TODO(pdf/image): 接入离线解析（DWG/PDF）与直接读图。
|
||
return None
|