From 63c54e1c204055682fa46d06eb145ea9d09cda44 Mon Sep 17 00:00:00 2001 From: ThearchyHelios Date: Mon, 5 Oct 2026 09:00:09 +0800 Subject: [PATCH] =?UTF-8?q?=E8=BF=90=E8=A1=8C=E5=AF=B9=E6=AF=94=EF=BC=9Aco?= =?UTF-8?q?mpare()=20=E7=9A=84=E8=AF=B4=E6=98=8E=E9=87=8C=E5=86=99?= =?UTF-8?q?=E5=85=A8=E8=BF=94=E5=9B=9E=E7=9A=84=E5=BD=A2=E7=8A=B6=EF=BC=8C?= =?UTF-8?q?=E6=8E=A5=E5=8F=A3=E8=AF=B4=E6=98=8E=E6=8C=87=E8=BF=87=E6=9D=A5?= =?UTF-8?q?=E6=97=B6=E5=89=8D=E7=AB=AF=E7=85=A7=E7=9D=80=E5=B0=B1=E8=83=BD?= =?UTF-8?q?=E8=AF=BB?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 除接口约定的字段外，几处补充也写明：模板节点的 labels（配置项的中文叫法）、输入的 masked 和两边的快照 id、查询的 两边快照 id、differences 和 problem；归不了因时 rows 为空、total_delta 为空、原因在 note。 --- backend/app/engine/compare.py | 21 +++++++++++++++++++-- 1 file changed, 19 insertions(+), 2 deletions(-) diff --git a/backend/app/engine/compare.py b/backend/app/engine/compare.py index c6e88e93..6fe89179 100644 --- a/backend/app/engine/compare.py +++ b/backend/app/engine/compare.py @@ -125,8 +125,25 @@ class Side: def compare(base: Side, target: Side) -> dict[str, Any]: - """对比两次运行（base 是基准，target 是这次）：{target, base, summary, template, metrics, queries, notes}， - 形状见接口约定（api/compare.py）。只读，不碰两次运行的任何东西。""" + """对比两次运行（base 是基准，target 是这次），只读，不碰两次运行的任何东西。返回： + + - target / base：两次运行的概要（Side.run 原样） + - summary：各类指标的个数，键是 CLASSES 里的八类 + - template：{base_version, target_version, same_graph, nodes: [{node_id, title, change: added | removed | config, + keys: [配置键名，值是字典的往下一层写 args.sql], labels: [同序的中文叫法]}]}。只列键名，不展示配置的值 + - metrics：每个指标一项 {card_id, card_title, metric_id, name, unit, decimals, base, target（有限的数，否则 + None）, base_rendered, target_rendered（卡上记的写法）, delta, delta_rendered（为 0、不能相减时为 None）, + delta_pct（相对基准的百分比，两位小数；基准为 0 时为 None）, classification, reasons: [说明], decomposition: + {caliber_effect, data_effect, caliber_rendered, data_rendered} 或 None, decomposition_note（为什么没拆）, + inputs: [{path, base, target, masked?, base_artifact?, target_artifact?}], attribution} + - attribution：{columns, measure, rows: [{key: [维度值…], base, target, delta, share}], total_delta, note} 或 + None。归不了因的（快照截断、度量列遮罩、快照读不出来、维度列每期都不同）rows 为空、total_delta 为 None，原因 + 写在 note；一边没有的维度值那一边写 None + - queries：每对查询一项 {node_id, node_title, ordinal, structure, base_sql, target_sql, literals: [{base, + target}]（只列不同的字面量）, base_rows, target_rows, base_truncated, target_truncated, base_artifact, + target_artifact, differences: [结构上差在哪], problem?（快照读不出来、被改过）} + - notes：整体的说明：连线、默认设置、运行输入的变化，口径卡执行了几次、读不出来，遮罩 + """ return _Compare(base, target).result()