mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 05:21:57 +00:00
ThearchyHelios
2d5e50fa88
配方语言（agentlab-recipe/2，新 recipe_types、recipe） - 封闭、不含正则、保存时静态校验：常量只能从表里的候选文字里选，占位符只能映射成空值，单位从固定词表里取 - 交叉表和列表两种版式；分段各自定位、不依赖先后顺序；合计按标签语义区间核对，关系恒等式、口径不同的声明都写在配方里 - 固定的解析器集合（parsers/2）：全角、各类横线、千分位、月日写法宽容归一，规范写法存库、原文另记 执行器（新 xlsx_cells、recipe_engine） - 两遍读取逐行对账，每个非空格都要有去向（格子账）；交叉表逆透视成长表，按统计期补年份并断言覆盖整期 - 合计行改作核对并另存一张表，不和明细混在一列；区域外文字、隐藏行列、合并单元格、空行之后的文字都单独报出 - STRICT 加主键写临时库；按锚点段记溯源，供后续证据下钻用；大表流式读取，扫描在 sheetData 处停止 校验与说明（新 recipe_checks、recipe_notes） - SQL 核对分通过、不一致、无法核对三种结果，失败分结构、数据质量、需确认三类；结构类直接拒收 - 数据质量类可以写理由接受，接受不改库、只改状态和说明 - 表的说明按核对结果生成、不含数字，经过数字检查；没核对过的不写「已核对」 起草（新 recipe_suggest、recipe_ai） - 先用规则起草：建议卡片加待确认问题（关系要不要登记、占位符表示什么） - 规则起草不完整时才提示 AI 兜底：界面先展示要发出的全部文字，用户同意才发；只发结构，数字、隐藏行列、遮罩列不发，公式先脱敏；同意绑定模型和服务方；AI 只写配方，有限次修订，记 token 和费用；截断单独识别 - 无论规则还是 AI 起草，都必须人工确认才能启用 导入流程（新 recipe_imports、recipe_confirm、recipe_diff、api/table_imports） - 暂存 → 起草 → 回答问题或改配方 → 试运行 → 勾确认项、写接受理由 → 确认并启用；接入期 1 的构建、导入记录、快照和运行固定版本 - 上传新一期按已确认的配方重放，不调用模型，出回执和差异卡，手动启用 - 改配方报破坏性变更（breaking:*）、单位变化、工作表改名的兜底，都要逐项勾选确认 前端导入向导（新 pages/import） - 原始网格（行列号、去向着色）、建议卡片、配方面板（表单，不露正则）、试运行回执、确认清单、差异卡、AI 起草的同意对话框
445 lines
15 KiB
Python
445 lines
15 KiB
Python
"""model_message 契约测试（P2-SPEC 6.2、第 10 节 WP-2）。
|
||
|
||
AI 修订时回灌给模型的是 Problem.model_message，绝不退回 message（message 可以带格子里的值）。这里为执行器
|
||
能产出的**每个** problem code 各造一个触发夹具，触发格里放特征数（987654）或特征文字，断言：
|
||
- 这个 code 真的出现了；
|
||
- 它的 model_message 非空；
|
||
- model_message 里没有触发格的值。
|
||
测试里维护一份全量 code 表，和 recipe_engine.PROBLEM_CODES 比对：执行器新增 code 不补用例，这个测试就失败。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import datetime as dt
|
||
import re
|
||
from collections.abc import Callable
|
||
from pathlib import Path
|
||
from typing import Any
|
||
|
||
import pytest
|
||
|
||
from app.data import recipe_engine, xlsx_cells, xlsx_scan
|
||
from app.data.recipe_types import Extraction, Recipe
|
||
from tests.test_recipe_engine_crosstab import FLOW, flow_book, recipe_dict
|
||
from tests.test_recipe_engine_list import LIST, list_recipe, sales_book
|
||
|
||
N = 987654
|
||
|
||
|
||
def _flow(edit: Callable[[Any, dict[str, Any]], None] | None = None, **kw: Any) -> tuple[bytes, str]:
|
||
return flow_book(days=3, edit=edit, **kw)
|
||
|
||
|
||
def _set(**cells: Any) -> Callable[[Any, dict[str, Any]], None]:
|
||
def edit(ws: Any, info: dict[str, Any]) -> None:
|
||
for ref, value in cells.items():
|
||
ws[ref] = value
|
||
return edit
|
||
|
||
|
||
def _run(book: tuple[bytes, str], recipe: dict[str, Any], *, scan_fix: Callable[[Any], None] | None = None,
|
||
**kw: Any) -> Extraction:
|
||
raw, filename = book
|
||
scan = xlsx_scan.scan(raw)
|
||
if scan_fix is not None:
|
||
scan_fix(scan)
|
||
return recipe_engine.execute(Recipe.model_validate(recipe), raw, filename, scan, None, **kw)
|
||
|
||
|
||
# --------------------------------------------------------------------------
|
||
# 每个 code 的触发夹具：返回 (Extraction, 触发格的值)
|
||
# --------------------------------------------------------------------------
|
||
|
||
Case = Callable[[pytest.MonkeyPatch], tuple[Extraction, list[Any]]]
|
||
|
||
|
||
def c_table_shape_conflict(mp):
|
||
data = recipe_dict()
|
||
data["sheets"][0]["blocks"][0]["segments"][2]["table"] = "日客流"
|
||
return _run(_flow(_set(C11=N)), data), [N]
|
||
|
||
|
||
def c_measures_keys_mismatch(mp):
|
||
data = recipe_dict()
|
||
seg = data["sheets"][0]["blocks"][0]["segments"][0]
|
||
seg["measures"] = {"全日客流(人次)": "全日客流", "分区甲（人次）": "分区甲", "分区乙（人次）": "分区乙"}
|
||
return _run(_flow(_set(C5=N)), data), [N]
|
||
|
||
|
||
def c_sheet_missing(mp):
|
||
data = recipe_dict()
|
||
data["sheets"][0]["match"] = {"name": "另一张表", "fallback": "none"}
|
||
return _run(_flow(_set(C5=N)), data), [N]
|
||
|
||
|
||
def c_sheet_unexpected(mp):
|
||
def edit(ws, info):
|
||
ws.parent.create_sheet("说明")["A1"] = N
|
||
return _run(_flow(edit), recipe_dict(other_visible_sheets="reject")), [N]
|
||
|
||
|
||
def c_sheet_too_large_for_layout(mp):
|
||
mp.setattr(xlsx_cells, "GRID_MAX_CELLS", 10)
|
||
return _run(_flow(_set(C11=N)), FLOW), [N]
|
||
|
||
|
||
def c_merged_too_many(mp):
|
||
def fix(scan):
|
||
scan.sheets[0].merged_total = len(scan.sheets[0].merged) + 1
|
||
return _run(_flow(_set(C11=N)), FLOW, scan_fix=fix), [N]
|
||
|
||
|
||
def c_sparse(mp):
|
||
return _run(_flow(_set(Z100000=N)), FLOW), [N]
|
||
|
||
|
||
def c_pass_mismatch(mp):
|
||
def fix(scan):
|
||
a, b, n = scan.sheets[0].row_runs[0]
|
||
scan.sheets[0].row_runs[0] = (a, b, n + 1)
|
||
return _run(_flow(_set(C11=N)), FLOW, scan_fix=fix), [N]
|
||
|
||
|
||
def c_cell_unclaimed(mp):
|
||
return _run(_flow(_set(G11=N)), FLOW), [N]
|
||
|
||
|
||
def c_outside_number(mp):
|
||
return _run(_flow(_set(B1=N)), FLOW), [N]
|
||
|
||
|
||
def c_outside_digits(mp):
|
||
text = f"注：{N}号闸机故障"
|
||
return _run(_flow(_set(B32=text)), FLOW), [text, N]
|
||
|
||
|
||
def c_hidden_rows(mp):
|
||
def edit(ws, info):
|
||
ws["C23"] = N
|
||
ws.row_dimensions[23].hidden = True
|
||
return _run(_flow(edit), FLOW), [N]
|
||
|
||
|
||
def c_hidden_cols(mp):
|
||
def edit(ws, info):
|
||
ws["D11"] = N
|
||
ws.column_dimensions["D"].hidden = True
|
||
return _run(_flow(edit), FLOW), [N]
|
||
|
||
|
||
def c_period_missing(mp):
|
||
text = f"统计时间：第{N}期"
|
||
return _run(_flow(period=text, filename=f"客流{N}.xlsx"), FLOW), [text, N]
|
||
|
||
|
||
def c_axis_not_found(mp):
|
||
return _run(_flow(_set(C4=N, D4=N + 1, E4=N + 2)), FLOW), [N, N + 1, N + 2]
|
||
|
||
|
||
def c_axis_ambiguous(mp):
|
||
def edit(ws, info):
|
||
for j in range(3):
|
||
ws.cell(40, 3 + j).value = ws.cell(4, 3 + j).value
|
||
ws["B40"] = N
|
||
return _run(_flow(edit), FLOW), ["8月1日", "8月3日", N]
|
||
|
||
|
||
def c_axis_gap(mp):
|
||
return _run(flow_book(days=5, edit=_set(E4=N)), FLOW), [N]
|
||
|
||
|
||
def c_axis_extra_cells(mp):
|
||
text = f"合计{N}"
|
||
def edit(ws, info):
|
||
ws["F4"] = text
|
||
ws["F5"] = N
|
||
return _run(_flow(edit), FLOW), [text, N]
|
||
|
||
|
||
def c_axis_year_missing(mp):
|
||
book = _flow(period="统计时间范围：2026年9月1日至2026年9月3日")
|
||
return _run(book, FLOW), ["8月1日", "8月2日", "2026年9月1日"]
|
||
|
||
|
||
def c_axis_year_ambiguous(mp):
|
||
book = flow_book(start=dt.date(2026, 1, 5), days=3, period="统计时间范围：2026年1月1日至2027年1月31日")
|
||
return _run(book, FLOW), ["1月5日", "1月6日"]
|
||
|
||
|
||
def c_axis_duplicate(mp):
|
||
return _run(flow_book(days=5, edit=_set(E4="8月2日")), FLOW), ["8月2日", "2026-08-02"]
|
||
|
||
|
||
def c_axis_not_contiguous(mp):
|
||
return _run(flow_book(days=5, edit=_set(D4="8月3日", E4="8月2日")), FLOW), ["8月3日", "2026-08-03",
|
||
"2026-08-02"]
|
||
|
||
|
||
def c_axis_coverage(mp):
|
||
book = flow_book(days=5, period="统计时间范围：2026年8月1日至2026年8月6日")
|
||
return _run(book, FLOW), ["8月1日", "8月5日", "2026-08-01", "2026-08-06"]
|
||
|
||
|
||
def c_row_without_label(mp):
|
||
return _run(_flow(_set(D32=N)), FLOW), [N]
|
||
|
||
|
||
def c_row_unclaimed(mp):
|
||
label = f"补录{N}"
|
||
return _run(_flow(_set(B32=label, C32=N + 1)), FLOW), [label, N + 1]
|
||
|
||
|
||
def c_title_not_found(mp):
|
||
title = f"日间{N}时段"
|
||
return _run(_flow(_set(B9=title)), FLOW), [title]
|
||
|
||
|
||
def c_title_ambiguous(mp):
|
||
title = "日间时段客流 （人次）" # 与配方的标题 match_key 相同，原文不同
|
||
def edit(ws, info):
|
||
ws.unmerge_cells("B8:E8")
|
||
ws["B8"] = title
|
||
return _run(_flow(edit), FLOW), [title]
|
||
|
||
|
||
def c_segment_not_found(mp):
|
||
data = recipe_dict()
|
||
seg = data["sheets"][0]["blocks"][0]["segments"][0]
|
||
seg["labels"]["expect"] = ["甲（人次）", "乙（人次）"]
|
||
seg["measures"] = {"甲（人次）": "甲", "乙（人次）": "乙"}
|
||
data["tables"][0]["units"] = {}
|
||
data["relations"] = []
|
||
return _run(_flow(_set(B5=f"全日{N}")), data), [f"全日{N}"]
|
||
|
||
|
||
def c_segment_overlap(mp):
|
||
data = recipe_dict()
|
||
data["sheets"][0]["blocks"][0]["segments"].append({
|
||
"id": "日间按标签", "role": "dimension", "table": "另表", "locate": {"by": "labels"},
|
||
"labels": {"expect": ["8-9", "9-10"]}, "dim": {"name": "时段", "parser": "hour_range"}, "value": "客流"})
|
||
data["tables"].append({"name": "另表", "grain": ["日期", "时段"]})
|
||
return _run(_flow(_set(C11=N)), data), [N]
|
||
|
||
|
||
def c_label_missing(mp):
|
||
return _run(flow_book(days=3, day_hours=tuple(range(8, 18)), edit=_set(C10=N)), FLOW), [N]
|
||
|
||
|
||
def c_label_unexpected(mp):
|
||
label = f"分区{N}（人次）"
|
||
def edit(ws, info):
|
||
ws.unmerge_cells("B8:E8")
|
||
ws["B8"] = label
|
||
for j in range(3):
|
||
ws.cell(8, 3 + j, N + 1)
|
||
return _run(_flow(edit), FLOW), [label, N + 1]
|
||
|
||
|
||
def c_label_duplicate(mp):
|
||
return _run(_flow(_set(B12="8－9")), FLOW), ["8－9"]
|
||
|
||
|
||
def c_label_unparsed(mp):
|
||
label = "7-18 时合计"
|
||
def edit(ws, info):
|
||
ws.insert_rows(21)
|
||
ws["B21"] = label
|
||
for j in range(3):
|
||
ws.cell(21, 3 + j, N)
|
||
return _run(flow_book(days=3, edit=edit, uncached=True), FLOW), [label, N]
|
||
|
||
|
||
def c_const_missing(mp):
|
||
data = recipe_dict()
|
||
data["sheets"][0]["blocks"][0]["segments"][2]["const"] = {"时段类别": {"pick": "晚上"}}
|
||
return _run(_flow(), data), ["夜间时段客流（人次）"]
|
||
|
||
|
||
def c_merged_in_values(mp):
|
||
def edit(ws, info):
|
||
ws["C12"] = N
|
||
ws.merge_cells("C12:D12")
|
||
return _run(_flow(edit), FLOW), [N]
|
||
|
||
|
||
def c_header_not_found(mp):
|
||
def edit(ws):
|
||
for c in range(1, 6):
|
||
ws.cell(3, c).value = f"列{N}{c}"
|
||
return _run(sales_book(edit=edit), LIST), [f"列{N}1", f"列{N}5"]
|
||
|
||
|
||
def c_header_ambiguous(mp):
|
||
def edit(ws):
|
||
for c, h in enumerate(("地区 ", "产品", "日期", "销量", "金额（万元）"), 1):
|
||
ws.cell(20, c, h)
|
||
ws["D21"] = N
|
||
return _run(sales_book(edit=edit), LIST), ["地区 ", N]
|
||
|
||
|
||
def c_column_missing(mp):
|
||
header = f"销售额{N}"
|
||
return _run(sales_book(edit=lambda ws: ws.cell(3, 5, header)), LIST), [header]
|
||
|
||
|
||
def c_column_extra(mp):
|
||
header = f"备注{N}"
|
||
def edit(ws):
|
||
ws["F3"] = header
|
||
ws["F4"] = N
|
||
return _run(sales_book(edit=edit), LIST), [header, N]
|
||
|
||
|
||
def c_rows_after_stop(mp):
|
||
text = f"名单{N}"
|
||
def edit(ws):
|
||
ws["A11"], ws["B11"] = text, "名单乙"
|
||
return _run(sales_book(total=False, edit=edit), list_recipe(rows={})), [text, "名单乙"]
|
||
|
||
|
||
def c_list_empty(mp):
|
||
"""AU-1：只有表头、没有数据行。触发格是表头之下本该有数据的地方，这里放不了值；断言表头文字之外的特征不外露。"""
|
||
def edit(ws):
|
||
for row in ws.iter_rows(min_row=4):
|
||
for cell in row:
|
||
cell.value = None
|
||
return _run(sales_book(total=False, edit=edit), list_recipe(rows={})), [N]
|
||
|
||
|
||
def c_total_row_missing(mp):
|
||
def edit(ws):
|
||
ws["D4"] = N
|
||
return _run(sales_book(total=False, edit=edit), LIST), [N]
|
||
|
||
|
||
def c_total_row_blank(mp):
|
||
def edit(ws):
|
||
ws["D4"] = N
|
||
ws["D9"] = ws["E9"] = None # 第 9 行是合计行：标签还在，数字格全空
|
||
return _run(sales_book(edit=edit), LIST), [N]
|
||
|
||
|
||
def c_value_not_integer(mp):
|
||
return _run(_flow(_set(C11=9876.54)), FLOW), [9876.54]
|
||
|
||
|
||
def c_value_text_number(mp):
|
||
return _run(_flow(_set(C11="987,654")), FLOW), ["987,654", N]
|
||
|
||
|
||
def c_value_blank(mp):
|
||
return _run(_flow(_set(C11=None, D11=N)), FLOW), [N]
|
||
|
||
|
||
def c_value_error(mp):
|
||
return _run(_flow(_set(C11="#DIV/0!")), FLOW), ["#DIV/0!"]
|
||
|
||
|
||
def c_value_not_number(mp):
|
||
return _run(_flow(_set(C11=f"约{N}")), FLOW), [f"约{N}", N]
|
||
|
||
|
||
def c_value_not_date(mp):
|
||
return _run(sales_book(edit=lambda ws: ws.cell(4, 3, f"{N}日")), LIST), [f"{N}日", N]
|
||
|
||
|
||
def c_value_formula(mp):
|
||
def edit(ws, info):
|
||
ws["C11"] = f"=C12+{N}"
|
||
info["cache"]["C11"] = N + 1
|
||
return _run(_flow(edit), FLOW), [N, N + 1]
|
||
|
||
|
||
def c_formula_uncached(mp):
|
||
data = recipe_dict(sheets__0__blocks__0__values__formula="accept_cached")
|
||
return _run(_flow(_set(C11=f"=C12+{N}")), data), [N]
|
||
|
||
|
||
def c_formula_full_calc(mp):
|
||
data = recipe_dict(sheets__0__blocks__0__values__formula="accept_cached")
|
||
def edit(ws, info):
|
||
ws["C11"] = f"=C12+{N}"
|
||
info["cache"]["C11"] = N + 1
|
||
return _run(flow_book(days=3, edit=edit, keep_full_calc=True), data), [N, N + 1]
|
||
|
||
|
||
def c_pk_duplicate(mp):
|
||
def edit(ws):
|
||
ws["A4"] = ws["A5"] = f"地区{N}"
|
||
ws["B5"] = ws["B4"].value
|
||
return _run(sales_book(edit=edit), LIST), [f"地区{N}", "产品000"]
|
||
|
||
|
||
CASES: dict[str, Case] = {
|
||
name[2:]: fn for name, fn in globals().items() if name.startswith("c_") and callable(fn)
|
||
}
|
||
|
||
|
||
def test_case_table_covers_every_problem_code():
|
||
assert set(CASES) == set(recipe_engine.PROBLEM_CODES)
|
||
assert set(recipe_engine.PROBLEM_FIX) <= set(recipe_engine.PROBLEM_CODES)
|
||
|
||
|
||
def _forms(value: Any) -> list[str]:
|
||
if isinstance(value, float):
|
||
return [repr(value), str(value)]
|
||
return [str(value)]
|
||
|
||
|
||
@pytest.mark.parametrize("code", sorted(recipe_engine.PROBLEM_CODES))
|
||
def test_model_message_is_filled_and_has_no_cell_values(code, monkeypatch):
|
||
ex, triggers = CASES[code](monkeypatch)
|
||
hits = [p for p in ex.problems if p.code == code]
|
||
assert hits, f"{code} 没有出现：{[(p.code, p.message) for p in ex.problems]}"
|
||
for p in ex.problems: # 同一夹具里顺带出现的别的 code 也要守这条
|
||
assert p.model_message.strip(), (p.code, p.message)
|
||
assert p.category == recipe_engine.PROBLEM_CODES[p.code]
|
||
assert len(p.cells) <= recipe_engine.CELLS_MAX
|
||
for p in hits:
|
||
for value in triggers:
|
||
for form in _forms(value):
|
||
assert form not in p.model_message, (code, form, p.model_message)
|
||
assert p.message # 给人看的那句照样有
|
||
|
||
|
||
def test_message_may_carry_values_but_model_message_does_not():
|
||
ex, _ = c_value_not_number(None)
|
||
p = next(p for p in ex.problems if p.code == "value_not_number")
|
||
assert f"约{N}" in p.message and f"约{N}" not in p.model_message
|
||
assert "客流汇总!C11" in p.cells and "C11" in p.model_message
|
||
|
||
|
||
#: 下一步指到配方面板上真实存在的「字段」和「选项」（terms.ts：报错写界面上的中文标签）。check-copy 看不出这种
|
||
#: 跨包的不一致，这里把引用的标签逐个钉住
|
||
NEXT_STEPS = {
|
||
"value_text_number": "在配方面板的「文本写的数字」中选择「千分位写法按数字保存」",
|
||
"value_blank": "在配方面板的「空格」中选择「存为空值」",
|
||
"value_formula": "在配方面板的「公式」中选择「按保存值导入」",
|
||
"value_not_number": "在配方面板中添加占位符",
|
||
"value_not_integer": "在配方面板中把「类型」改为「小数」",
|
||
"value_not_date": "在配方面板中把这一列的「类型」改为「文字」",
|
||
"sheet_unexpected": "在配方面板的「配方没有列出的其他工作表」中选择「记入回执，需确认」",
|
||
"rows_after_stop": "在配方面板的「数据中间的空行」中选择「跳过继续」",
|
||
"hidden_rows": "在配方面板的「隐藏行」中选择「照常导入」或「跳过」",
|
||
"hidden_cols": "在配方面板的「隐藏列」中选择「照常导入」",
|
||
"total_row_missing": "在配方面板的「合计行」中取消「有合计行」",
|
||
"total_row_blank": "在配方面板的「合计行」中取消「有合计行」",
|
||
}
|
||
|
||
|
||
@pytest.mark.parametrize("code", sorted(NEXT_STEPS))
|
||
def test_next_steps_name_recipe_panel_labels(code, monkeypatch):
|
||
ex, _ = CASES[code](monkeypatch)
|
||
p = next(p for p in ex.problems if p.code == code)
|
||
assert NEXT_STEPS[code] in p.message, p.message
|
||
|
||
|
||
def test_next_step_labels_exist_in_the_recipe_panel():
|
||
"""引用的每个标签都要在前端的配方面板文案里（RECIPE_TEXT / RECIPE_CHOICE_LABEL）。前端的配方面板合入之前跳过。"""
|
||
terms = Path(__file__).resolve().parents[2] / "frontend" / "src" / "lib" / "terms.ts"
|
||
text = terms.read_text(encoding="utf-8") if terms.exists() else ""
|
||
if "RECIPE_CHOICE_LABEL" not in text:
|
||
pytest.skip("前端的配方面板文案还没有合入")
|
||
labels = {x for step in NEXT_STEPS.values() for x in re.findall(r"「(.+?)」", step)} | {"主键"}
|
||
missing = sorted(x for x in labels if f"'{x}" not in text)
|
||
assert not missing, missing
|