diff --git a/client/src/pdd_collect_service.py b/client/src/pdd_collect_service.py
index 2177221..ef15785 100644
--- a/client/src/pdd_collect_service.py
+++ b/client/src/pdd_collect_service.py
@@ -31,6 +31,7 @@ from .pdd_page_classifier import (
PAGE_HOME,
PAGE_LOGIN_REQUIRED,
PAGE_NETWORK_ERROR,
+ PAGE_ORDER_CONFIRMATION,
PAGE_PAYMENT,
PAGE_RISK_CONTROL,
GoodsOpenTracker,
@@ -466,6 +467,8 @@ def _dimension_key(name: str, used: set[str]) -> str:
def _is_dimension_heading(label: str) -> bool:
compact = label.replace(" ", "")
+ if compact.startswith(("请选择", "請選擇", "已选", "已選")):
+ return False
# 新版页面会把可选数量写进标题，例如“颜色 (6)”或“颜色（6）”。
# 数量不是规格名称的一部分，只在判断标题类型时去掉，最终展示仍保留原文。
compact = re.sub(r"[（(]\d+[)）]$", "", compact)
@@ -496,37 +499,92 @@ def _find_non_scrollable_spec_panel(
) -> Optional[ET.Element]:
"""用多项强证据定位不暴露 scrollable 属性的自绘规格面板。"""
- headings: list[ET.Element] = []
- summaries: list[ET.Element] = []
- panel_cues: list[ET.Element] = []
- confirms: list[ET.Element] = []
- for node in root.iter("node"):
- label = _preferred_node_label(node).strip()
- compact = label.replace(" ", "")
- if not label or _parse_bounds(node.get("bounds", "")) is None:
- continue
- if _is_dimension_heading(label):
- headings.append(node)
- if compact.startswith(("已选", "已選", "请选择", "請選擇")):
- summaries.append(node)
- if compact in ("确认款式", "確認款式", "关闭", "關閉"):
- panel_cues.append(node)
- if compact in ("确定", "確定") and node.get("clickable") == "true":
- confirms.append(node)
-
- if not headings or not summaries or not panel_cues or not confirms:
- return None
-
- required = [*headings, summaries[0], panel_cues[0], confirms[0]]
- common = set([required[0], *_ancestors(required[0], parents)])
- for node in required[1:]:
- common.intersection_update([node, *_ancestors(node, parents)])
-
all_bounds = [
bounds
for node in root.iter("node")
if (bounds := _parse_bounds(node.get("bounds", ""))) is not None
]
+ screen_bottom = max((item[3] for item in all_bounds), default=0)
+ pdd_node_count = sum(
+ node.get("package") == PDD_PACKAGE_NAME
+ for node in root.iter("node")
+ )
+
+ headings: list[ET.Element] = []
+ summaries: list[ET.Element] = []
+ panel_cues: list[ET.Element] = []
+ confirms: list[ET.Element] = []
+ submit_hints: list[ET.Element] = []
+ quantity_editors: list[ET.Element] = []
+ decreases: list[ET.Element] = []
+ increases: list[ET.Element] = []
+ for node in root.iter("node"):
+ label = _preferred_node_label(node).strip()
+ compact = label.replace(" ", "")
+ bounds = _parse_bounds(node.get("bounds", ""))
+ if bounds is None:
+ continue
+ is_summary = compact.startswith(
+ ("已选", "已選", "请选择", "請選擇")
+ )
+ if label and not is_summary and _is_dimension_heading(label):
+ headings.append(node)
+ if is_summary:
+ summaries.append(node)
+ if compact in ("确认款式", "確認款式", "关闭", "關閉"):
+ panel_cues.append(node)
+ if compact in ("确定", "確定") and node.get("clickable") == "true":
+ confirms.append(node)
+ descendant_label = _preferred_or_descendant_label(node).replace(" ", "")
+ if (
+ node.get("clickable") == "true"
+ and node.get("enabled", "true") != "false"
+ and node.get("visible-to-user", "true") != "false"
+ and screen_bottom
+ and (bounds[1] + bounds[3]) // 2 >= screen_bottom * 0.6
+ and "提交订单" in descendant_label
+ and any(
+ word in descendant_label
+ for word in ("选择", "颜色", "尺码", "规格")
+ )
+ ):
+ submit_hints.append(node)
+ if (
+ node.get("class") == "android.widget.EditText"
+ and node.get("text", "").strip().isdigit()
+ and int(node.get("text", "0")) > 0
+ ):
+ quantity_editors.append(node)
+ if compact == "减少数量" and node.get("clickable") == "true":
+ decreases.append(node)
+ if compact == "增加数量" and node.get("clickable") == "true":
+ increases.append(node)
+
+ if not headings or not summaries or not panel_cues:
+ return None
+ if confirms:
+ action_nodes = [confirms[0]]
+ elif (
+ pdd_node_count >= 3
+ and len(submit_hints) == 1
+ and len(quantity_editors) == 1
+ and len(decreases) == 1
+ and len(increases) == 1
+ ):
+ action_nodes = [
+ submit_hints[0],
+ quantity_editors[0],
+ decreases[0],
+ increases[0],
+ ]
+ else:
+ return None
+
+ required = [*headings, summaries[0], panel_cues[0], *action_nodes]
+ common = set([required[0], *_ancestors(required[0], parents)])
+ for node in required[1:]:
+ common.intersection_update([node, *_ancestors(node, parents)])
+
screen_area = 0
if all_bounds:
screen_area = max(item[2] for item in all_bounds) * max(
@@ -1223,21 +1281,34 @@ class PddCollectService:
"""等待点击后的规格面板真正出现，不能只依赖固定延时。"""
deadline = self._monotonic() + self._spec_panel_timeout
+ saw_confirmation_page = False
+ incomplete_reads = 0
while self._monotonic() < deadline:
self._check_cancelled()
xml_data = self._dump_hierarchy(device)
+ root = _parse_xml(xml_data)
+ if classify_pdd_page(root, "").kind == PAGE_ORDER_CONFIRMATION:
+ saw_confirmation_page = True
try:
snapshot = parse_spec_panel(xml_data)
except PddCollectError as exc:
if exc.code != "PDD_DATA_SPEC_INCOMPLETE":
raise
self._last_invalid_spec_xml = xml_data
+ incomplete_reads += 1
else:
if _is_spec_panel_open(xml_data):
self._last_valid_spec_xml = xml_data
return snapshot
self._last_invalid_spec_xml = xml_data
+ incomplete_reads += 1
self._sleep(0.25)
+ if saw_confirmation_page:
+ raise PddCollectError(
+ "PDD_DATA_SPEC_INCOMPLETE",
+ "规格面板已经打开，但页面结构无法识别",
+ {"incomplete_spec_reads": incomplete_reads},
+ )
raise PddCollectError(
"PDD_PAGE_SPEC_PANEL_TIMEOUT",
"点击规格入口后，等待规格面板加载超时",
diff --git a/client/test/fixtures/pdd_spec_panel_submit_hint.xml b/client/test/fixtures/pdd_spec_panel_submit_hint.xml
new file mode 100644
index 0000000..97aa1ec
--- /dev/null
+++ b/client/test/fixtures/pdd_spec_panel_submit_hint.xml
@@ -0,0 +1,62 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
diff --git a/client/test/test_pdd_collect_service.py b/client/test/test_pdd_collect_service.py
index 27b477f..7809cab 100644
--- a/client/test/test_pdd_collect_service.py
+++ b/client/test/test_pdd_collect_service.py
@@ -609,6 +609,9 @@ class PddCollectParserTest(unittest.TestCase):
cls.non_scrollable_spec_xml = (
FIXTURES / "pdd_spec_panel_non_scrollable.xml"
).read_text(encoding="utf-8")
+ cls.submit_hint_spec_xml = (
+ FIXTURES / "pdd_spec_panel_submit_hint.xml"
+ ).read_text(encoding="utf-8")
cls.sold_out_xml = (
FIXTURES / "pdd_sold_out_recommendations.xml"
).read_text(encoding="utf-8")
@@ -789,6 +792,41 @@ class PddCollectParserTest(unittest.TestCase):
self.assertEqual(result.raw_price, "¥8.17")
self.assertNotIn("确认款式", [item.name for item in result.dimensions])
+ def test_parse_non_scrollable_submit_hint_spec_panel(self):
+ result = parse_spec_panel(self.submit_hint_spec_xml)
+
+ self.assertTrue(_is_spec_panel_open(self.submit_hint_spec_xml))
+ self.assertEqual(
+ [(item.key, item.name) for item in result.dimensions],
+ [("color", "颜色分类"), ("size", "尺码")],
+ )
+ self.assertEqual(
+ [value.text for value in result.dimensions[0].values],
+ ["测试黑色", "测试红色"],
+ )
+ self.assertEqual(
+ [value.text for value in result.dimensions[1].values],
+ ["M", "L"],
+ )
+ self.assertNotIn(
+ "请选择： 颜色分类 尺码",
+ [item.name for item in result.dimensions],
+ )
+
+ def test_isolated_submit_hint_is_not_a_spec_panel(self):
+ xml_data = """
+
+
+
+
+ """
+
+ self.assertFalse(_is_spec_panel_open(xml_data))
+
def test_non_scrollable_spec_panel_exposes_three_color_rows(self):
service = PddCollectService(
PddDeviceService(lambda _serial: object()),
@@ -1790,6 +1828,32 @@ class PddCollectParserTest(unittest.TestCase):
artifacts = raised.exception.diagnostics["artifacts"]
self.assertTrue(Path(artifacts[0]["path"]).is_file())
+ def test_open_confirmation_with_unrecognized_structure_is_not_timeout(self):
+ malformed_panel = self.submit_hint_spec_xml.replace(
+ '',
+ "",
+ )
+ device = FakeCollectDevice(self.home_xml, malformed_panel)
+ device.opened_url = "https://mobile.yangkeduo.com/goods.html?goods_id=123"
+ device.panel_open = True
+ clock = FakeClock()
+ service = PddCollectService(
+ PddDeviceService(lambda _serial: device),
+ "USB-001",
+ "client-001",
+ sleeper=clock.sleep,
+ monotonic=clock.monotonic,
+ spec_panel_timeout=1.0,
+ )
+
+ with self.assertRaises(PddCollectError) as raised:
+ service._wait_spec_panel(device)
+
+ self.assertEqual(raised.exception.code, "PDD_DATA_SPEC_INCOMPLETE")
+ self.assertIn("已经打开", raised.exception.message)
+
def test_transient_empty_spec_trees_are_skipped_during_full_collection(self):
device = TransientSpecTreeDevice(
self.home_xml, keep_only_one_sku(self.spec_xml)
diff --git a/docs/client/02-architecture.md b/docs/client/02-architecture.md
index 2e45fa3..925a0a7 100644
--- a/docs/client/02-architecture.md
+++ b/docs/client/02-architecture.md
@@ -434,6 +434,12 @@ PDD URL，包名和页面类型仍以最新控件树确认。
PDD 页面可能出现登录失效、验证码、控件树不完整、A/B 页面、库存变化和价格变化。适配层必须返回结构化错误，不得把这些情况统一返回 `False`。
+规格面板既可能是带 `scrollable=true` 的滚动容器，也可能是自绘的非滚动容器。
+非滚动容器使用组合证据识别：关闭或确认标题、选择摘要、独立规格标题，以及
+“确定”按钮；另一种全高面板则必须同时具有唯一数量输入框、唯一加减按钮和底部
+提交提示。以“请选择/已选”开头的摘要不是规格标题。已经确认进入规格确认页但结构
+仍无法解析时返回规格数据不完整，不得继续误报规格入口或面板加载超时。
+
采集商品时按“首页就绪 → 读取当前首页摘要 → 有限滚动补采评价和店铺 →
点击规格入口 → 确认规格面板出现 → 颜色列表归左 → 按行蛇形逐色点击并采价 →
向下滚动并只读尺码 → 组装结果”的顺序执行。颜色点击可能改变列表位置，因此每次点击后必须重新读取