"""文本列按文本渲染：查询快照带着 column_types（查询时从驱动的原始值记下的）。 以前快照里没有列类型，VARCHAR 的 "2026" 和小数位为 0 的 DECIMAL "45678" 长得一模一样，只能 「像数字就当数字」：报告里 [[v:Q1.r0.week]] 渲染成「2,026」。现在 text 列一律按文本： render_cell、[[v:]] 的渲染和取值、写作目录、cite_fields 没声明类型的字段。拿不到 column_types 的老快照保持原来的启发式。 口径卡表达式里的 cell() 照旧按值猜：那是在算数，数字写成文本的列（CAST 成文本、VARCHAR 存金额） 多半就是要拿来算的；要文本写 str(cell(...))。 """ from __future__ import annotations import asyncio import sqlite3 import pytest from langchain_core.messages import AIMessage from sqlalchemy import select from app.core import artifact_store from app.data.engine import engines from app.db.base import SessionLocal from app.db.models import DataSource, Run from app.engine.evidence import ( RenderError, build_catalog, catalog_prompt, compose_doc, iter_segments, render_cell, verify_doc, ) from app.engine.expressions import cell_value, column_kind, eval_expression from app.engine.runner import run_manager BASE = {"columns": ["week", "gmv", "code"], "rows": [["2026", "45678", "10001"], ["2027", "12.50", "10002"]], "row_count": 2, "truncated": False, "elapsed_ms": 1, "sql": "SELECT week, gmv, code FROM orders", "source": "shop"} TYPED = {**BASE, "column_types": {"week": "text", "gmv": "number", "code": "text"}} def _store(snap: dict) -> str: text = artifact_store.canonical_json(snap) artifact = artifact_store.content_hash(text) path = artifact_store._path_of(artifact) path.parent.mkdir(parents=True, exist_ok=True) path.write_text(text, encoding="utf-8") return artifact def catalog_of(snap: dict) -> dict: artifact = _store(snap) return build_catalog(nodes={}, ledger=[{ "kind": "query", "node_id": "fetch", "exec": 1, "artifact": artifact, "via": "b" * 64, "tool": "db_query__shop", "source": "shop", "columns": snap["columns"], "rows": len(snap["rows"]), "truncated": False}], inputs={}) def test_cell_value_keeps_text_columns_as_text(): assert cell_value("2026", "text") == "2026" assert cell_value("45678", "number") == 45678 assert cell_value("2026") == 2026 # 老快照：照旧按值猜 assert cell_value("2026", None) == 2026 assert column_kind(TYPED, "week") == "text" and column_kind(TYPED, "gmv") == "number" assert column_kind(BASE, "week") is None and column_kind(TYPED, "nope") is None assert column_kind({"column_types": "bad"}, "week") is None def test_render_cell_reads_the_column_type(): assert render_cell("2026", kind="text") == "2026" assert render_cell("10001", kind="text") == "10001" assert render_cell("45678", kind="number") == "45,678" assert render_cell("2026") == "2,026" with pytest.raises(RenderError): render_cell("2026", "万", kind="text") def test_cell_refs_render_text_columns_as_text(): doc = compose_doc("[[v:Q1.r0.week]] 周销售额 [[v:Q1.r0.gmv]]，编号 [[v:Q1.r1.code]]。", catalog_of(TYPED)) assert doc["markdown"] == "2026 周销售额 45,678，编号 10002。" week, gmv, code = [s for s in iter_segments(doc) if s.get("ref")] assert week["kind"] == "value" and week["cite"]["value"] == "2026" assert gmv["kind"] == "number" and gmv["cite"]["value"] == 45678 assert code["kind"] == "value" and code["cite"]["value"] == "10002" assert doc["violations"] == [] assert verify_doc(doc, catalog_of(TYPED))["violations"] == [] def test_old_snapshots_keep_the_heuristic(): doc = compose_doc("[[v:Q1.r0.week]] 周", catalog_of(BASE)) assert doc["markdown"] == "2,026 周" def test_a_text_column_cannot_be_converted(): doc = compose_doc("[[v:Q1.r0.week|万]]", catalog_of(TYPED)) [bad] = [v for v in doc["violations"] if v["code"] == "unresolved_ref"] assert "不是数" in bad["message"] def test_tables_render_text_columns_as_text(): doc = compose_doc("[[table:Q1 cols=week,gmv]]", catalog_of(TYPED)) assert "| 2026 | 45,678 |" in doc["markdown"] and "| 2027 | 12.5 |" in doc["markdown"] def test_the_writing_catalog_shows_what_the_report_will_show(): prompt = catalog_prompt(catalog_of(TYPED)) assert "week=2026，gmv=45,678，code=10001" in prompt # 示例挑第一个数：文本列不算 assert "例：[[v:Q1.r0.gmv]]" in prompt assert "例：[[v:Q1.r0.week]]" in catalog_prompt(catalog_of(BASE)) def test_cite_fields_without_a_declared_type_follow_the_column(): from app.engine.nodes.llm import cited_fields, verify_cited_fields queries = [{"alias": "Q1", "artifact": "a" * 64, "via": "b" * 64}] parsed = {"week": {"value": "2026", "from": {"call": "Q1", "row": 0, "column": "week"}}, "gmv": {"value": 45678, "from": {"call": "Q1", "row": 0, "column": "gmv"}}, "weeks": {"value": ["2026", "2027"], "from": {"call": "Q1", "rows": "0-1", "column": "week"}}} schema = {"type": "object", "properties": {"week": {}, "gmv": {}, "weeks": {"type": "array", "items": {}}}} data, ev = verify_cited_fields({"parsed": parsed}, cited_fields(schema), queries, loader=lambda a: TYPED) assert data == {"week": "2026", "gmv": 45678, "weeks": ["2026", "2027"]} assert {e["status"] for e in ev.values()} == {"verified"} # 声明了类型的照声明：写 schema 的人说了它是数 declared = {"type": "object", "properties": {"week": {"type": "integer"}, "gmv": {}, "weeks": { "type": "array", "items": {"type": "number"}}}} data, _ = verify_cited_fields({"parsed": parsed}, cited_fields(declared), queries, loader=lambda a: TYPED) assert data == {"week": 2026, "gmv": 45678, "weeks": [2026, 2027]} # 老快照照旧按值猜 data, _ = verify_cited_fields({"parsed": parsed}, cited_fields(schema), queries, loader=lambda a: BASE) assert data["week"] == 2026 def test_card_expressions_keep_computing_on_number_like_text(): assert eval_expression("cell(vars.q, 0, 'week') + 1", {"vars": {"q": TYPED}}) == 2027 assert eval_expression("str(cell(vars.q, 0, 'week'))", {"vars": {"q": TYPED}}) == "2026" # -------------------------------------------------------------------------- # 端到端：列类型从驱动一路带到报告 # -------------------------------------------------------------------------- @pytest.fixture async def shop(tmp_path): await run_manager.setup() path = tmp_path / "shop.db" db = sqlite3.connect(path) db.executescript("CREATE TABLE orders (id INTEGER PRIMARY KEY, week TEXT, amount REAL);") db.executemany("INSERT INTO orders VALUES (?, ?, ?)", [(1, "2026", 100.5), (2, "2026", 200.0)]) db.commit() db.close() async with SessionLocal() as session: row = (await session.execute(select(DataSource).where(DataSource.name == "shop"))).scalar_one_or_none() if row is None: row = DataSource(name="shop", kind="sqlite", database=str(path), readonly=True, enabled=True) session.add(row) else: row.database, row.kind, row.enabled, row.readonly, row.options = str(path), "sqlite", True, True, {} await session.commit() source_id = row.id await engines.invalidate(source_id) yield await engines.invalidate(source_id) await run_manager.shutdown() async def test_a_report_quotes_a_text_year_as_written(shop, monkeypatch): from app.providers import mock_model monkeypatch.setattr(mock_model.MockChatModel, "_decide", lambda model, messages: AIMessage(content="[[v:Q1.r0.week]] 年销售额 [[v:Q1.r0.gmv]]。")) def node(nid, ntype, **config): return {"id": nid, "type": ntype, "position": {"x": 0, "y": 0}, "data": {"label": nid, "config": config}} nodes = [node("start", "input"), node("fetch", "tool", tool="db_query__shop", args={"sql": "SELECT week, SUM(amount) AS gmv FROM orders GROUP BY week"}), node("write", "report", instructions="写一句话", on_violation="flag"), node("out", "output", fields=[{"name": "答案", "value": "{{ nodes.write.text }}"}])] graph = {"nodes": nodes, "edges": [{"source": a["id"], "target": b["id"]} for a, b in zip(nodes, nodes[1:])]} run = await run_manager.start(graph=graph, input_payload={}) for _ in range(400): await asyncio.sleep(0.05) async with SessionLocal() as session: row = await session.get(Run, run.id) if row.status in ("succeeded", "failed"): break assert row.status == "succeeded", row.error assert row.output["答案"] == "2026 年销售额 300.5。"