运行对比里的分析卡只认做完了的那次执行:analysis.end 先按卡记在这一次执行里,见到这张卡的 node.finished 才算数(新的 node.started 清掉上一次没做完的)。分析卡执行到一半失败时前几项已经发了 analysis.end,以前运行对比会把这次失败执行的判定当成结果列出来、也算进执行次数。测试:中途失败的那次不列、不算次数,失败后重来做完的按做完的那次对比

This commit is contained in:
ThearchyHelios
2026-10-21 15:50:15 +08:00
parent 27ac11fa5f
commit 9339facf9a
2 changed files with 32 additions and 6 deletions
+11 -6
View File
@@ -336,17 +336,22 @@ def _queries(side: Side) -> list[_Query]:
def _analysis_items(side: Side) -> dict[str, tuple[dict[str, dict[str, Any]], int]]:
"""封存范围内每张分析卡各项最后一次的结论(analysis.end,engine/nodes/analysis.py 每项一条):{卡: ({项: 事件数据},
执行了几次)},项按记下的先后。循环里执行了几次的,同一项后一次的盖过前一次。"""
"""封存范围内每张分析卡各项最后一次做完的结论(analysis.end,engine/nodes/analysis.py 每项一条):{卡: ({项: 事件数据},
执行了几次)},项按记下的先后。只认后面跟着这张卡 node.finished 的那几次执行:执行到一半失败的(前几项已经发了
analysis.end,后面一项让卡失败),那次的判定不是这次运行的结果,不列,也不算执行次数。循环里执行了几次的,同一项后
一次的盖过前一次。"""
out: dict[str, tuple[dict[str, dict[str, Any]], int]] = {}
pending: dict[str, dict[str, dict[str, Any]]] = {} # 卡 → 这一次执行里已经做完的项(还没见到 node.finished)
for _, etype, node_id, data in side.events:
if not node_id or not isinstance(data, dict):
continue
if etype == EventType.ANALYSIS_END and isinstance(data.get("item"), str):
if etype == EventType.NODE_STARTED:
pending.pop(node_id, None)
elif etype == EventType.ANALYSIS_END and isinstance(data.get("item"), str):
pending.setdefault(node_id, {})[data["item"]] = data
elif etype == EventType.NODE_FINISHED and node_id in pending:
items, runs = out.setdefault(node_id, ({}, 0))
items[data["item"]] = data
elif etype == EventType.NODE_FINISHED and node_id in out:
items, runs = out[node_id]
items.update(pending.pop(node_id))
out[node_id] = (items, runs + 1)
return out
+21
View File
@@ -40,3 +40,24 @@ def test_compare_aligns_analysis_items():
assert wow["change"] == "phrase"
# 没有分析卡的两次运行:analyses 是空的
assert compare.compare(week("r5").side(), week("r6").side())["analyses"] == []
def test_a_card_that_failed_half_way_lists_only_what_a_finished_execution_decided():
"""分析卡执行到一半失败:前几项已经发了 analysis.end,后一项让卡失败(node.failed,没有 node.finished)。那次执行的判定
不是这次运行的结果:只认后面跟着 node.finished 的执行,失败的那次不列、也不算执行次数。"""
base = _with_card(week("r7"), [("wow", "within_band", "在常规波动范围内")])
failed = week("r8")
failed.add("node.started", "insight", {"node_type": "analysis"})
failed.add("analysis.end", "insight", {"item": "wow", "op": "baseline", "mode": None, "status": "ok",
"verdict": "outside_band", "direction": "down", "phrase": "超出常规波动范围",
"analysis_artifact": "wow-failed"})
failed.add("node.failed", "insight", {"error": "分析项「where」(变化拆解):比率不能按段相加"})
[wow] = compare.compare(base.side(), failed.side())["analyses"]
assert wow["change"] == "removed" and wow["target"] is None and wow["base"]["verdict"] == "within_band"
# 失败后重来一次做完了:按做完的那次对比,执行次数只算做完的
_with_card(failed, [("wow", "far_outside_band", "明显超出常规波动范围")])
result = compare.compare(base.side(), failed.side())
[wow] = result["analyses"]
assert wow["change"] == "verdict" and wow["target"]["verdict"] == "far_outside_band"
assert not [n for n in result["notes"] if "执行了" in n]