fix: 合并只读核单多视口证据 (#122)

This commit is contained in:
chengma
2026-08-10 19:46:00 +08:00
parent 7e4adf2a1c
commit a8f8e7360b
2 changed files with 316 additions and 28 deletions
+182 -26
View File
@@ -50,6 +50,8 @@ _PAYMENT_ACTION_MARKERS = ("立即支付", "确认支付", "输入支付密码")
_UNPAID_MARKERS = ("待付款", "待支付")
_NON_UNPAID_MARKERS = ("已付款", "交易成功", "交易完成", "已取消", "退款")
_SAFE_NAVIGATION_LABELS = ("个人中心", "我的订单", "待付款")
_ORDER_DETAIL_PAY_LABELS = ("去支付",)
_ORDER_DETAIL_CANCEL_LABELS = ("取消订单",)
class PddPurchaseReconcileError(RuntimeError):
@@ -83,6 +85,38 @@ def _subtree_text(node: ET.Element) -> str:
)
def _package_hint_from_tree(root: ET.Element) -> str:
"""可靠控件树可以证明 PDD 在前台,避免错误的前台查询结果。"""
pdd_node_count = sum(
1
for node in root.iter("node")
if node.get("package") == PDD_PACKAGE_NAME
)
return PDD_PACKAGE_NAME if pdd_node_count >= 3 else ""
def _label_count(root: ET.Element, targets: tuple[str, ...]) -> int:
return sum(
1
for node in root.iter("node")
if _label(node) in targets
)
def _is_single_order_detail(root: ET.Element) -> bool:
"""只在页面上只有一组订单操作时允许跨视口累积字段。"""
pay_count = _label_count(root, _ORDER_DETAIL_PAY_LABELS)
cancel_count = _label_count(root, _ORDER_DETAIL_CANCEL_LABELS)
order_numbers = {
match.group(1)
for node in root.iter("node")
if (match := _ORDER_NO_PATTERN.search(_label(node))) is not None
}
return pay_count == 1 and cancel_count <= 1 and len(order_numbers) <= 1
def _parse_bounds(value: str) -> Optional[Bounds]:
match = _BOUNDS_PATTERN.fullmatch(str(value or "").strip())
if match is None:
@@ -141,6 +175,16 @@ def _parse_ordered_at(text: str) -> tuple[str, str]:
return "", raw
def _payment_status(text: str) -> str:
"""只在页面有明确状态文字时返回状态。"""
if any(marker in text for marker in _NON_UNPAID_MARKERS):
return "other"
if any(marker in text for marker in _UNPAID_MARKERS):
return "unpaid"
return ""
def parse_order_candidates(
xml_data: str | bytes, query: PurchaseReconcileQuery
) -> tuple[PurchaseOrderCandidate, ...]:
@@ -184,12 +228,6 @@ def parse_order_candidates(
for key, value in query.options.items()
if str(value).strip() and str(value).strip() in card_text
}
payment_status = (
"unpaid"
if any(marker in card_text for marker in _UNPAID_MARKERS)
and not any(marker in card_text for marker in _NON_UNPAID_MARKERS)
else "other"
)
candidate = PurchaseOrderCandidate(
order_no=order_no,
goods_id=goods_match.group(1) if goods_match else "",
@@ -198,7 +236,7 @@ def parse_order_candidates(
total_price_cent=_parse_money_cent(card_text),
ordered_at=ordered_at,
ordered_at_raw=ordered_at_raw,
payment_status=payment_status,
payment_status=_payment_status(card_text),
)
previous = parsed.get(order_no)
if previous is None or previous == candidate:
@@ -206,6 +244,81 @@ def parse_order_candidates(
return tuple(parsed.values())
def _parse_order_detail_evidence(
root: ET.Element, query: PurchaseReconcileQuery
) -> PurchaseOrderCandidate:
"""读取单个订单详情当前视口;字段可以暂时不完整。"""
text = _subtree_text(root)
order_numbers = tuple(dict.fromkeys(_ORDER_NO_PATTERN.findall(text)))
goods_match = _GOODS_ID_PATTERN.search(text)
ordered_at, ordered_at_raw = _parse_ordered_at(text)
options = {
key: value
for key, value in query.options.items()
if str(value).strip() and str(value).strip() in text
}
return PurchaseOrderCandidate(
order_no=order_numbers[0] if len(order_numbers) == 1 else "",
goods_id=goods_match.group(1) if goods_match else "",
options=options,
quantity=_parse_quantity(text),
total_price_cent=_parse_money_cent(text),
ordered_at=ordered_at,
ordered_at_raw=ordered_at_raw,
payment_status=_payment_status(text),
)
def _merge_order_evidence(
previous: PurchaseOrderCandidate,
current: PurchaseOrderCandidate,
) -> Optional[PurchaseOrderCandidate]:
"""合并同一订单的相邻视口;关键字段冲突时拒绝合并。"""
scalar_fields = (
"order_no",
"goods_id",
"quantity",
"total_price_cent",
"ordered_at",
"ordered_at_raw",
)
values: dict[str, object] = {}
for field_name in scalar_fields:
old_value = getattr(previous, field_name)
new_value = getattr(current, field_name)
if old_value and new_value and old_value != new_value:
return None
values[field_name] = old_value or new_value
options = dict(previous.options)
for key, value in current.options.items():
if key in options and options[key] != value:
return None
options[key] = value
statuses = {
value
for value in (previous.payment_status, current.payment_status)
if value
}
if "other" in statuses:
payment_status = "other"
else:
payment_status = "unpaid" if "unpaid" in statuses else ""
return PurchaseOrderCandidate(
order_no=str(values["order_no"]),
goods_id=str(values["goods_id"]),
options=options,
quantity=int(values["quantity"]),
total_price_cent=int(values["total_price_cent"]),
ordered_at=str(values["ordered_at"]),
ordered_at_raw=str(values["ordered_at_raw"]),
payment_status=payment_status,
)
class U2PddPurchaseReconcileAdapter(PddPurchaseReconcileAdapter):
"""只导航和滚动读取待付款订单,不包含任何订单写操作。"""
@@ -237,19 +350,47 @@ class U2PddPurchaseReconcileAdapter(PddPurchaseReconcileAdapter):
device = self._connect()
self._open_unpaid_orders(device)
found: dict[str, PurchaseOrderCandidate] = {}
previous_signature = ""
detail_evidence: Optional[PurchaseOrderCandidate] = None
detail_context_valid = True
previous_xml = ""
pages_scanned = 0
for _ in range(self._max_pages):
self._check_cancelled()
xml_data = device.dump_hierarchy(compressed=False)
self._raise_for_special_page(_parse_xml(xml_data))
root = _parse_xml(xml_data)
self._raise_for_special_page(root)
pages_scanned += 1
for candidate in parse_order_candidates(xml_data, query):
found.setdefault(candidate.order_no, candidate)
signature = "|".join(sorted(found)) + f":{len(str(xml_data))}"
if signature == previous_signature:
previous = found.get(candidate.order_no)
merged = (
_merge_order_evidence(previous, candidate)
if previous is not None
else candidate
)
if merged is not None:
found[candidate.order_no] = merged
if detail_context_valid and _is_single_order_detail(root):
current_evidence = _parse_order_detail_evidence(root, query)
if detail_evidence is None:
detail_evidence = current_evidence
else:
merged = _merge_order_evidence(
detail_evidence, current_evidence
)
if merged is None:
detail_context_valid = False
detail_evidence = None
else:
detail_evidence = merged
else:
detail_context_valid = False
detail_evidence = None
current_xml = str(xml_data)
if current_xml == previous_xml:
break
previous_signature = signature
previous_xml = current_xml
width, height = device.window_size()
device.swipe(
width // 2,
@@ -259,6 +400,15 @@ class U2PddPurchaseReconcileAdapter(PddPurchaseReconcileAdapter):
0.35,
)
self._settle()
if detail_context_valid and detail_evidence and detail_evidence.order_no:
previous = found.get(detail_evidence.order_no)
merged = (
_merge_order_evidence(previous, detail_evidence)
if previous is not None
else detail_evidence
)
if merged is not None:
found[detail_evidence.order_no] = merged
return PurchaseReconcileScan(
candidates=tuple(found.values()),
diagnostics={
@@ -281,26 +431,32 @@ class U2PddPurchaseReconcileAdapter(PddPurchaseReconcileAdapter):
return self._device
def _open_unpaid_orders(self, device: Any) -> None:
current = device.app_current()
if str(current.get("package") or "") != PDD_PACKAGE_NAME:
device.app_start(PDD_PACKAGE_NAME)
self._settle()
root = _parse_xml(device.dump_hierarchy(compressed=False))
if not _package_hint_from_tree(root):
current = device.app_current()
if str(current.get("package") or "") != PDD_PACKAGE_NAME:
device.app_start(PDD_PACKAGE_NAME)
self._settle()
opened_order_area = False
for _ in range(6):
self._check_cancelled()
current = device.app_current()
if str(current.get("package") or "") != PDD_PACKAGE_NAME:
raise PddPurchaseReconcileError(
"RECONCILE_WRONG_APP", "核单时 PDD 不在前台"
)
root = _parse_xml(device.dump_hierarchy(compressed=False))
if not _package_hint_from_tree(root):
current = device.app_current()
if str(current.get("package") or "") != PDD_PACKAGE_NAME:
raise PddPurchaseReconcileError(
"RECONCILE_WRONG_APP", "核单时 PDD 不在前台"
)
self._raise_for_special_page(root)
combined = _subtree_text(root)
if opened_order_area and (
_ORDER_NO_PATTERN.search(combined)
or "暂无订单" in combined
or "暂无相关订单" in combined
if _is_single_order_detail(root) or (
opened_order_area
and (
_ORDER_NO_PATTERN.search(combined)
or "暂无订单" in combined
or "暂无相关订单" in combined
)
):
return
if any(marker in combined for marker in _PAYMENT_ACTION_MARKERS):
@@ -1,9 +1,20 @@
"""只读订单核对 XML 解析测试;不连接真实手机。"""
import unittest
from contextlib import AbstractContextManager
from src.pdd_purchase_reconcile_adapter import PurchaseReconcileQuery
from src.pdd_u2_purchase_reconcile_adapter import parse_order_candidates
from src.pdd_purchase_reconcile_adapter import (
PurchaseOrderCandidate,
PurchaseReconcileQuery,
)
from src.pdd_u2_purchase_reconcile_adapter import (
U2PddPurchaseReconcileAdapter,
_is_single_order_detail,
_merge_order_evidence,
_parse_xml,
parse_order_candidates,
)
from src.purchase_reconcile_service import PurchaseReconcileService
def query() -> PurchaseReconcileQuery:
@@ -28,6 +39,63 @@ def xml_card(text: str) -> str:
)
def xml_detail(text: str) -> str:
package = "com.xunmeng.pinduoduo"
return (
'<?xml version="1.0" encoding="UTF-8"?>'
'<hierarchy rotation="0">'
f'<node package="{package}" text="{text}" />'
f'<node package="{package}" text="去支付" />'
f'<node package="{package}" text="取消订单" />'
"</hierarchy>"
)
class FakeSession(AbstractContextManager):
def __init__(self, device):
self.device = device
def __enter__(self):
return self.device
def __exit__(self, exc_type, exc_value, traceback):
return None
class FakeDeviceService:
def __init__(self, device):
self.device = device
def connect(self, device_address):
return FakeSession(self.device)
class DetailDevice:
def __init__(self, pages):
self.pages = pages
self.page_index = 0
self.app_current_calls = 0
self.app_start_calls = 0
self.swipes = 0
def dump_hierarchy(self, compressed=False):
return self.pages[self.page_index]
def app_current(self):
self.app_current_calls += 1
return {"package": "com.android.settings"}
def app_start(self, package):
self.app_start_calls += 1
def window_size(self):
return 1080, 2400
def swipe(self, *args):
self.swipes += 1
self.page_index = min(self.page_index + 1, len(self.pages) - 1)
class PddU2PurchaseReconcileAdapterTest(unittest.TestCase):
def test_parser_extracts_only_required_unpaid_order_fields(self):
candidates = parse_order_candidates(
@@ -70,6 +138,70 @@ class PddU2PurchaseReconcileAdapterTest(unittest.TestCase):
self.assertEqual(candidates[0].goods_id, "")
self.assertEqual(candidates[0].ordered_at, "")
def test_adapter_merges_one_order_detail_across_viewports(self):
device = DetailDevice(
[
xml_detail(
"待付款 商品编号:737116531267 黑色 L "
"共2件 合计 ¥84.00"
),
xml_detail(
"订单编号:ORDER-20260810 "
"下单时间:2026-08-10 16:03:00"
),
]
)
adapter = U2PddPurchaseReconcileAdapter(
"serial",
device_service=FakeDeviceService(device),
settle_seconds=0,
max_pages=3,
)
scan = adapter.read_order_candidates(query())
adapter.close()
self.assertEqual(len(scan.candidates), 1)
candidate = scan.candidates[0]
self.assertEqual(candidate.order_no, "ORDER-20260810")
self.assertEqual(candidate.goods_id, "737116531267")
self.assertEqual(candidate.options, {"color": "黑色", "size": "L"})
self.assertEqual(candidate.quantity, 2)
self.assertEqual(candidate.total_price_cent, 8400)
self.assertEqual(candidate.ordered_at, "2026-08-10T08:03:00Z")
self.assertEqual(candidate.payment_status, "unpaid")
self.assertTrue(PurchaseReconcileService._candidate_matches(candidate, query()))
self.assertEqual(device.app_current_calls, 0)
self.assertEqual(device.app_start_calls, 0)
def test_conflicting_order_numbers_are_not_merged(self):
first = PurchaseOrderCandidate(
order_no="ORDER-A",
goods_id="737116531267",
payment_status="unpaid",
)
second = PurchaseOrderCandidate(
order_no="ORDER-B",
quantity=2,
total_price_cent=8400,
payment_status="unpaid",
)
self.assertIsNone(_merge_order_evidence(first, second))
def test_multiple_order_actions_disable_detail_evidence_merge(self):
root = _parse_xml(
'<hierarchy><node package="com.xunmeng.pinduoduo" '
'text="订单编号:ORDER-A" />'
'<node package="com.xunmeng.pinduoduo" text="去支付" />'
'<node package="com.xunmeng.pinduoduo" '
'text="订单编号:ORDER-B" />'
'<node package="com.xunmeng.pinduoduo" text="去支付" />'
"</hierarchy>"
)
self.assertFalse(_is_single_order_detail(root))
if __name__ == "__main__":
unittest.main()