mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-16 04:11:58 +00:00
一张草稿图支持多个取数节点
一份文档里的指标常常分属不同粒度、不同表,一条 SQL 装不下。build_graph 改成接收多组
{sql, entry_fields, metrics}:一组仍叫 fetch,多组叫 fetch_1…,每组指标的表达式改写到
自己的取数节点,指标 id 全局重排,入口字段合并(同名不同类型直接报错,不静默取舍)。
同时把「文档被截」和「目录被截」拆开:以前目录超长也报「文档太长」,界面在骗人。
This commit is contained in:
+117
-35
@@ -12,8 +12,13 @@ SELECT,服务端把 SQL、入口字段、口径卡、报告串成图,后台
|
||||
这里按 sqlglot 解析后的语句数判定,不靠字符串里有没有分号(字符串字面量里的分号不算)。注释里的第二条照样会被解析成
|
||||
第二条语句,所以也拒。
|
||||
|
||||
**为什么表达式必须是 cell(nodes.fetch, 0, '列名')。** 试跑和之后按口径出具都从口径卡读值。取数节点固定叫 fetch(见
|
||||
build_graph),单行结果取第 0 行的某一列;写成别的样子,卡上就没有值,图能建但出不了数。
|
||||
**为什么表达式必须是 cell(nodes.fetch, 0, '列名')。** 试跑和之后按口径出具都从口径卡读值。一次 plan 只起草一组,这组的
|
||||
取数节点固定叫 fetch(多组拼图时由 build_graph 改成 fetch_1、fetch_2…),单行结果取第 0 行的某一列;写成别的样子,卡上
|
||||
就没有值,图能建但出不了数。
|
||||
|
||||
**为什么 build_graph 收多组 blocks。** 一张草稿图原来只能挂一条 SQL,业务文档几十条指标得分几十次起草、几十张图。取数
|
||||
节点多几个、口径卡读全部,试跑一次就跑完;入口字段按 name 合并,同名但类型/必填对不上就抛 ValueError——那是两份草案自己
|
||||
打架,调用方当成问题记下来,比悄悄取一个更能让人发现。
|
||||
|
||||
**为什么引用核对还是照 docdraft 的老规矩。** 定义的事实来源还是文档原文。模型转述得再像,也要能在原文里逐字定位;一条
|
||||
引用都核不上的指标整条丢弃,写进 problems 交给人看。这里只做校验,不改 docdraft 的规则。
|
||||
@@ -25,7 +30,7 @@ from __future__ import annotations
|
||||
|
||||
import json
|
||||
import re
|
||||
from collections.abc import Mapping
|
||||
from collections.abc import Mapping, Sequence
|
||||
from typing import Any
|
||||
|
||||
from app.ledger import docdraft
|
||||
@@ -45,6 +50,8 @@ METRIC_FIELDS = ("id", "name", "synonyms", "unit", "decimals", "expression", "de
|
||||
_INPUT = re.compile(r"\{\{\s*input\.([A-Za-z_][\w-]*)\s*\}\}")
|
||||
#: 口径卡的取值表达式,列名可以用单引号或双引号
|
||||
_EXPRESSION = re.compile(r"^cell\(\s*nodes\.fetch\s*,\s*0\s*,\s*['\"]([^'\"]+)['\"]\s*\)$")
|
||||
#: 指标表达式里的取数节点引用;后面紧跟字母数字下划线的不算(nodes.fetch_2 已经带下标,不改)
|
||||
_FETCH_REF = re.compile(r"nodes\.fetch(?![A-Za-z0-9_])")
|
||||
#: 只读查询的根节点:普通 SELECT,或各分支都是 SELECT 的 UNION
|
||||
_WRITE_NODES = ("Insert", "Update", "Delete", "Create", "Drop", "Alter", "Merge", "TruncateTable", "Command")
|
||||
|
||||
@@ -62,6 +69,7 @@ _SYSTEM = """你是业务口径与取数模板的起草助手。读给定的业
|
||||
6. 拿不准的不要猜,写进 open_questions,用一句问得清的话。不要编造文档里没有的数字、单位、生效日期。
|
||||
7. **一次只起草能共用同一条 SQL 的一组指标,最多 6 条。** 文档里其余的指标不要硬凑进这一条 SQL,把名称写进
|
||||
uncovered(字符串数组),下一轮再起草。不同粒度、不同表、不同时间字段的指标本来就不该共用一条 SQL。
|
||||
human 里会列出「已经起草过的指标」:这些不要重复起草,这一组只从未起草的指标里选,剩下的未起草指标都写进 uncovered。
|
||||
8. **sql 要短。** 能用 GROUP BY 和聚合表达的,不要为每个属性枚举一长串 CASE WHEN 列;输出额度有限,SQL 写长了,
|
||||
后面的指标就写不出来了。
|
||||
9. 写不下时宁可少写几条指标、把剩下的写进 uncovered,**也不要让 JSON 半截断掉**。
|
||||
@@ -127,7 +135,7 @@ def _chunk_block(chunk_id: str, chunk: Any) -> str:
|
||||
|
||||
|
||||
def _human(docs: list[dict], chunks: Mapping[str, dict], guidance: str, datasource: dict | None,
|
||||
tables_text: str) -> str:
|
||||
tables_text: str, covered: Sequence[str] = ()) -> str:
|
||||
parts = ["文档清单:", *(_doc_line(doc) for doc in docs), "",
|
||||
"逐段原文(引用时 chunk 写方括号里的 id,quote 逐字抄):"]
|
||||
parts.extend(_chunk_block(chunk_id, chunks[chunk_id]) for chunk_id in _order(chunks))
|
||||
@@ -139,17 +147,21 @@ def _human(docs: list[dict], chunks: Mapping[str, dict], guidance: str, datasour
|
||||
parts.append("数据目录(sql 只能用这里的表和列,表名列名照抄):")
|
||||
parts.append(tables_text.strip() or "(没有给出表结构,不能写 sql)")
|
||||
parts.append("")
|
||||
parts.append("已经起草过的指标(不要重复起草):")
|
||||
names = [name.strip() for name in covered if isinstance(name, str) and name.strip()]
|
||||
parts.append("、".join(names) if names else "(无)")
|
||||
parts.append("")
|
||||
parts.append("用户补充要求:{}".format(guidance.strip() if guidance and guidance.strip() else "(无)"))
|
||||
return "\n".join(parts)
|
||||
|
||||
|
||||
def prompt(docs: list[dict], chunks: dict[str, dict], guidance: str = "", datasource: dict | None = None,
|
||||
tables_text: str = "") -> list[Any]:
|
||||
"""构造 system + human:文档原文 + 数据目录。目录是 sql 的唯一依据,和原文一样原样交给模型。"""
|
||||
tables_text: str = "", covered: Sequence[str] = ()) -> list[Any]:
|
||||
"""构造 system + human:文档原文 + 数据目录 + 已经起草过的指标。目录是 sql 的唯一依据,和原文一样原样交给模型。"""
|
||||
from langchain_core.messages import HumanMessage, SystemMessage
|
||||
|
||||
return [SystemMessage(content=_SYSTEM),
|
||||
HumanMessage(content=_human(docs, chunks, guidance, datasource, tables_text))]
|
||||
HumanMessage(content=_human(docs, chunks, guidance, datasource, tables_text, covered))]
|
||||
|
||||
|
||||
def _chars(messages: list[Any]) -> int:
|
||||
@@ -157,25 +169,30 @@ def _chars(messages: list[Any]) -> int:
|
||||
|
||||
|
||||
def _fit(docs: list[dict], chunks: dict[str, dict], guidance: str,
|
||||
datasource: dict | None, tables_text: str) -> tuple[dict[str, dict], str, bool]:
|
||||
datasource: dict | None, tables_text: str,
|
||||
covered: Sequence[str] = ()) -> tuple[dict[str, dict], str, bool, bool]:
|
||||
"""把材料压到 MAX_INPUT_CHARS 以内:先截表目录尾部,再从后往前按 chunk 边界丢,最后截仅剩的那一段。
|
||||
|
||||
表目录先截:SQL 必须写在目录里的表上,目录比文档的细枝末节更硬;被截掉的是表详情,表名索引还在最前面。
|
||||
|
||||
截断来源分开报:doc_truncated 是文档段落被丢/被截,catalog_truncated 是表目录被截。调用方要能说清「指标没覆盖
|
||||
完,是因为原文没给全,还是因为目录没给全」——改文档、改目录是两件不同的事。
|
||||
"""
|
||||
kept = dict(chunks)
|
||||
tables = tables_text
|
||||
truncated = False
|
||||
doc_truncated = False
|
||||
catalog_truncated = False
|
||||
if len(tables) > MAX_TABLES_CHARS:
|
||||
tables = tables[:MAX_TABLES_CHARS]
|
||||
truncated = True
|
||||
catalog_truncated = True
|
||||
|
||||
def size() -> int:
|
||||
return _chars(prompt(docs, kept, guidance, datasource, tables))
|
||||
return _chars(prompt(docs, kept, guidance, datasource, tables, covered))
|
||||
|
||||
order = _order(kept)
|
||||
while size() > MAX_INPUT_CHARS and len(order) > 1:
|
||||
kept.pop(order.pop())
|
||||
truncated = True
|
||||
doc_truncated = True
|
||||
if size() > MAX_INPUT_CHARS and order:
|
||||
chunk_id = order[0]
|
||||
chunk = dict(kept[chunk_id]) if isinstance(kept[chunk_id], Mapping) else {}
|
||||
@@ -183,12 +200,12 @@ def _fit(docs: list[dict], chunks: dict[str, dict], guidance: str,
|
||||
excess = size() - MAX_INPUT_CHARS
|
||||
chunk["content"] = content[:max(0, len(content) - excess)]
|
||||
kept[chunk_id] = chunk
|
||||
truncated = True
|
||||
doc_truncated = True
|
||||
if size() > MAX_INPUT_CHARS:
|
||||
excess = size() - MAX_INPUT_CHARS
|
||||
tables = tables[:max(0, len(tables) - excess)]
|
||||
truncated = True
|
||||
return kept, tables, truncated
|
||||
catalog_truncated = True
|
||||
return kept, tables, doc_truncated, catalog_truncated
|
||||
|
||||
|
||||
def _salvage_value(text: str, key: str) -> Any:
|
||||
@@ -225,11 +242,15 @@ def _salvage_reply(text: str) -> dict:
|
||||
|
||||
|
||||
async def plan(session: Any, *, docs: list[dict], chunks: dict[str, dict], guidance: str = "",
|
||||
datasource: dict | None = None, tables_text: str = "", model: Any = None) -> dict:
|
||||
"""一次模型调用,返回 {entry_fields, sql, metrics, open_questions, model_id, truncated}(未校验)。
|
||||
datasource: dict | None = None, tables_text: str = "", model: Any = None,
|
||||
covered: Sequence[str] = ()) -> dict:
|
||||
"""一次模型调用,返回 {entry_fields, sql, metrics, open_questions, model_id, doc_truncated,
|
||||
catalog_truncated, truncated}(未校验)。
|
||||
|
||||
不落库。model 可注入(测试用假对象:有 async ainvoke(messages),回复有 .content);不给就按设置解析,模型不可用时抛
|
||||
docdraft.DraftUnavailable。SQL 的对错、引用能不能核对,交给 normalize_plan。
|
||||
covered 是前几轮已经起草过的指标名,写进提示词让模型不要重复;调用方按 uncovered 续跑,直到 covered 不再增长。
|
||||
truncated 仍是两者的「或」,老前端只认这一个键。不落库。model 可注入(测试用假对象:有 async ainvoke(messages),
|
||||
回复有 .content);不给就按设置解析,模型不可用时抛 docdraft.DraftUnavailable。SQL 的对错、引用能不能核对,交给
|
||||
normalize_plan。
|
||||
"""
|
||||
from app.engine.state import message_text
|
||||
|
||||
@@ -243,8 +264,9 @@ async def plan(session: Any, *, docs: list[dict], chunks: dict[str, dict], guida
|
||||
named = getattr(model, "model_name", None) or getattr(model, "model", None)
|
||||
model_id = named if isinstance(named, str) else ""
|
||||
|
||||
kept, tables, truncated = _fit(docs, chunks, guidance, datasource, tables_text)
|
||||
reply = await model.ainvoke(prompt(docs, kept, guidance, datasource, tables))
|
||||
kept, tables, doc_truncated, catalog_truncated = _fit(docs, chunks, guidance, datasource, tables_text,
|
||||
covered)
|
||||
reply = await model.ainvoke(prompt(docs, kept, guidance, datasource, tables, covered))
|
||||
text = message_text(reply)
|
||||
salvaged = False
|
||||
try:
|
||||
@@ -276,7 +298,9 @@ async def plan(session: Any, *, docs: list[dict], chunks: dict[str, dict], guida
|
||||
"salvaged": salvaged,
|
||||
"diagnostics": diagnostics,
|
||||
"model_id": model_id,
|
||||
"truncated": truncated,
|
||||
"doc_truncated": doc_truncated,
|
||||
"catalog_truncated": catalog_truncated,
|
||||
"truncated": doc_truncated or catalog_truncated,
|
||||
}
|
||||
|
||||
|
||||
@@ -462,12 +486,44 @@ def normalize_plan(raw: Any, chunks: Mapping[str, dict], *, allowed_tables: set[
|
||||
# ==========================================================================
|
||||
|
||||
|
||||
def build_graph(plan: dict, *, name: str, question: str = "") -> dict:
|
||||
"""草案 → 草稿图:start(input) → fetch(tool) → card(metrics) → write(report) → out(output),过 auto_layout。
|
||||
def _blocks(value: Any) -> list[dict]:
|
||||
"""blocks 可以是 [{"sql", "entry_fields", "metrics"}],也可以是老调用的单个 plan 字典(当成只有一组)。"""
|
||||
if isinstance(value, Mapping):
|
||||
return [dict(value)]
|
||||
return [dict(block) for block in value or [] if isinstance(block, Mapping)]
|
||||
|
||||
plan["datasource"] 是取数节点指向的数据源名(调用方按这次的数据源补,不采信模型回复里的名字)。调用两次结果逐字一致:
|
||||
节点 id、坐标都由输入决定,没有时间戳。报告配置、出口契约照 certify.build_graph 的写法,发布门禁认的就是这一套
|
||||
(REPORT_POLICY、metrics_from、required)。
|
||||
|
||||
def _merge_fields(target: dict[str, dict], value: Any) -> None:
|
||||
"""入口字段按 name 合并;同名但类型或必填对不上就抛 ValueError,不拿其中一组去覆盖另一组。"""
|
||||
for field in value if isinstance(value, list) else []:
|
||||
if not isinstance(field, Mapping):
|
||||
continue
|
||||
name = str(field.get("name") or "").strip()
|
||||
if not name:
|
||||
continue
|
||||
item = dict(field)
|
||||
item["name"] = name
|
||||
old = target.get(name)
|
||||
if old is None:
|
||||
target[name] = item
|
||||
continue
|
||||
if (str(old.get("type") or "string") != str(item.get("type") or "string")
|
||||
or bool(old.get("required")) != bool(item.get("required"))):
|
||||
raise ValueError(f"入口字段「{name}」在各组里的类型或必填不一致,不能合成一个入口")
|
||||
|
||||
|
||||
def build_graph(blocks: Any, *, name: str, question: str = "", datasource: str = "") -> dict:
|
||||
"""各组草案 → 一张草稿图:start(input) → fetch…(tool,每组一个) → card(metrics) → write(report) → out(output)。
|
||||
|
||||
blocks 是 [{"sql", "entry_fields", "metrics"}],也兼容老调用的单个 plan 字典(当成只有一组)。一组时取数节点仍叫
|
||||
fetch,多组叫 fetch_1、fetch_2…(和 certify.build_graph 一致);第 i 组指标表达式里的 nodes.fetch 改写成对应的
|
||||
节点 id,已经带下标的(nodes.fetch_2)不动。指标 id 跨组重名时按 m1…mN 全局重排,口径卡和出口写重排后的 id——
|
||||
重名会让口径卡读错值,宁可换号也不能让两个指标共用一个 id。入口字段按 name 合并,同名但类型/必填对不上抛
|
||||
ValueError(两份草案自己打架,调用方当成问题记下来)。datasource 是取数节点指向的数据源名,调用方按这次的数据源
|
||||
补,也可以写在某一组里;各组指向不同数据源时用第一组的(草稿图只连一个数据源)。
|
||||
|
||||
调用两次结果逐字一致:节点 id、坐标都由输入决定,没有时间戳。报告配置、出口契约照 certify.build_graph 的写法,
|
||||
发布门禁认的就是这一套(REPORT_POLICY、metrics_from、required)。
|
||||
"""
|
||||
from app.engine.governance import REPORT_POLICY
|
||||
from app.engine.layout import auto_layout
|
||||
@@ -475,9 +531,38 @@ def build_graph(plan: dict, *, name: str, question: str = "") -> dict:
|
||||
from app.ledger.certify import REPORT_INSTRUCTIONS, REPORT_INSTRUCTIONS_PARAMS
|
||||
from app.tools.datasource import QUERY_PREFIX
|
||||
|
||||
fields = [dict(f) for f in plan.get("entry_fields") or []]
|
||||
metrics = [dict(m) for m in plan.get("metrics") or []]
|
||||
source = str(plan.get("datasource") or "")
|
||||
groups = _blocks(blocks) or [{}]
|
||||
multi = len(groups) > 1
|
||||
|
||||
def nid(index: int) -> str:
|
||||
return f"fetch_{index + 1}" if multi else "fetch"
|
||||
|
||||
merged: dict[str, dict] = {}
|
||||
metrics: list[dict] = []
|
||||
fetch_nodes: list[dict[str, Any]] = []
|
||||
source = str(datasource or "")
|
||||
for index, group in enumerate(groups):
|
||||
_merge_fields(merged, group.get("entry_fields"))
|
||||
node_id = nid(index)
|
||||
for metric in group.get("metrics") or []:
|
||||
if not isinstance(metric, Mapping):
|
||||
continue
|
||||
item = dict(metric)
|
||||
item["expression"] = _FETCH_REF.sub(f"nodes.{node_id}", str(item.get("expression") or ""))
|
||||
metrics.append(item)
|
||||
source = source or str(group.get("datasource") or "")
|
||||
fetch_nodes.append({"id": node_id, "type": "tool", "position": {"x": 0, "y": 0},
|
||||
"data": {"label": f"取数 {index + 1}" if multi else "取数",
|
||||
"config": {"tool": f"{QUERY_PREFIX}{source}",
|
||||
"args": {"sql": str(group.get("sql") or "")}}}})
|
||||
|
||||
ids = [str(metric.get("id") or "").strip() for metric in metrics]
|
||||
if "" in ids or len(set(ids)) != len(ids):
|
||||
for position, metric in enumerate(metrics, start=1):
|
||||
metric["id"] = f"m{position}"
|
||||
ids = [metric["id"] for metric in metrics]
|
||||
|
||||
fields = list(merged.values())
|
||||
report_cfg: dict[str, Any] = {}
|
||||
if fields:
|
||||
report_cfg["instructions"] = REPORT_INSTRUCTIONS_PARAMS.format(
|
||||
@@ -486,14 +571,11 @@ def build_graph(plan: dict, *, name: str, question: str = "") -> dict:
|
||||
report_cfg["instructions"] = REPORT_INSTRUCTIONS.format(question=question or name or "(问题)")
|
||||
report_cfg.update({key: value for key, value, _ in REPORT_POLICY})
|
||||
report_cfg["metrics_from"] = ["card"]
|
||||
contract: dict[str, Any] = {"report_from": "write", "metrics_from": "card",
|
||||
"required": [m["id"] for m in metrics]}
|
||||
contract: dict[str, Any] = {"report_from": "write", "metrics_from": "card", "required": ids}
|
||||
chain = [
|
||||
{"id": "start", "type": "input", "position": {"x": 0, "y": 0},
|
||||
"data": {"label": "入口", "config": {"fields": fields}}},
|
||||
{"id": "fetch", "type": "tool", "position": {"x": 0, "y": 0},
|
||||
"data": {"label": "取数", "config": {"tool": f"{QUERY_PREFIX}{source}",
|
||||
"args": {"sql": str(plan.get("sql") or "")}}}},
|
||||
*fetch_nodes,
|
||||
{"id": "card", "type": "metrics", "position": {"x": 0, "y": 0},
|
||||
"data": {"label": "口径卡", "config": {"caliber": (name.strip() or question or "口径")[:40],
|
||||
"caliber_version": "v1", "metrics": metrics}}},
|
||||
|
||||
@@ -158,6 +158,86 @@ def test_build_graph_is_deterministic_and_passes_validation():
|
||||
assert result.ok, [issue.message for issue in result.issues if issue.level == "error"]
|
||||
|
||||
|
||||
# ==========================================================================
|
||||
# build_graph:多组
|
||||
# ==========================================================================
|
||||
|
||||
START = {"name": "start", "type": "string", "required": True, "default": "2026-09-01",
|
||||
"description": "开始日期"}
|
||||
|
||||
|
||||
def block(sql: str, metrics: list[dict], fields: list[dict] | None = None) -> dict:
|
||||
return {"sql": sql, "entry_fields": [dict(f) for f in ([START] if fields is None else fields)],
|
||||
"metrics": [dict(m) for m in metrics]}
|
||||
|
||||
|
||||
def test_build_graph_gives_each_group_its_own_fetch_node_and_rewrites_expressions():
|
||||
"""一张草稿图挂多个取数节点:第 i 组指标读 fetch_i,指标 id 跨组重排,出口要求全部指标。"""
|
||||
first = block("SELECT SUM(amount) AS gmv FROM orders",
|
||||
[dict(METRIC), {**METRIC, "id": "m2", "name": "订单数",
|
||||
"expression": "cell(nodes.fetch, 0, 'cnt')"}])
|
||||
second = block("SELECT COUNT(DISTINCT customer_id) AS buyers FROM orders",
|
||||
[{**METRIC, "id": "m1", "name": "买家数",
|
||||
"expression": "cell(nodes.fetch, 0, 'buyers')"}])
|
||||
graph = draftplan.build_graph([first, second], name="销售口径", question="上周", datasource="shop")
|
||||
assert draftplan.build_graph([first, second], name="销售口径", question="上周",
|
||||
datasource="shop") == graph
|
||||
assert [node["id"] for node in graph["nodes"]] == ["start", "fetch_1", "fetch_2", "card",
|
||||
"write", "out"]
|
||||
assert [edge["source"] for edge in graph["edges"]] == ["start", "fetch_1", "fetch_2", "card",
|
||||
"write"]
|
||||
nodes = {node["id"]: node for node in graph["nodes"]}
|
||||
assert nodes["fetch_1"]["data"]["config"] == {
|
||||
"tool": "db_query__shop", "args": {"sql": first["sql"]}}
|
||||
assert nodes["fetch_2"]["data"]["config"]["args"]["sql"] == second["sql"]
|
||||
metrics = nodes["card"]["data"]["config"]["metrics"]
|
||||
assert [m["id"] for m in metrics] == ["m1", "m2", "m3"]
|
||||
assert [m["expression"] for m in metrics] == ["cell(nodes.fetch_1, 0, 'gmv')",
|
||||
"cell(nodes.fetch_1, 0, 'cnt')",
|
||||
"cell(nodes.fetch_2, 0, 'buyers')"]
|
||||
contract = nodes["out"]["data"]["config"]["contract"]
|
||||
assert contract == {"report_from": "write", "metrics_from": "card",
|
||||
"required": ["m1", "m2", "m3"]}
|
||||
result = validate_graph(GraphSpec.model_validate(graph))
|
||||
assert result.ok, [issue.message for issue in result.issues if issue.level == "error"]
|
||||
|
||||
|
||||
def test_build_graph_leaves_expressions_that_already_carry_an_index_alone():
|
||||
"""已经写成 nodes.fetch_2 的式子不能再被改坏(改成了 nodes.fetch_1_2 这类)。"""
|
||||
first = block("SELECT SUM(amount) AS gmv FROM orders",
|
||||
[{**METRIC, "expression": "cell(nodes.fetch_2, 0, 'gmv')"}])
|
||||
second = block("SELECT 1 AS gmv FROM orders", [dict(METRIC)])
|
||||
graph = draftplan.build_graph([first, second], name="x")
|
||||
metrics = next(node for node in graph["nodes"] if node["id"] == "card")["data"]["config"]["metrics"]
|
||||
assert metrics[0]["expression"] == "cell(nodes.fetch_2, 0, 'gmv')"
|
||||
assert metrics[1]["expression"] == "cell(nodes.fetch_2, 0, 'gmv')"
|
||||
|
||||
|
||||
def test_build_graph_accepts_one_block_as_a_single_group():
|
||||
"""老调用传一份草案(含 datasource)也要能建图:取数节点仍叫 fetch。"""
|
||||
graph = draftplan.build_graph({**block("SELECT 1 AS gmv FROM orders", [dict(METRIC)]),
|
||||
"datasource": "shop"}, name="销售额口径")
|
||||
assert [node["id"] for node in graph["nodes"]] == ["start", "fetch", "card", "write", "out"]
|
||||
fetch = next(node for node in graph["nodes"] if node["id"] == "fetch")
|
||||
assert fetch["data"]["config"]["tool"] == "db_query__shop"
|
||||
card = next(node for node in graph["nodes"] if node["id"] == "card")["data"]["config"]
|
||||
assert card["metrics"][0]["expression"] == "cell(nodes.fetch, 0, 'gmv')"
|
||||
|
||||
|
||||
def test_build_graph_merges_entry_fields_and_refuses_conflicting_ones():
|
||||
same = block("SELECT 1 AS gmv FROM orders", [dict(METRIC)])
|
||||
merged = draftplan.build_graph([same, block("SELECT 2 AS cnt FROM orders", [{**METRIC, "id": "m2"}])],
|
||||
name="x")
|
||||
fields = next(node for node in merged["nodes"] if node["id"] == "start")["data"]["config"]["fields"]
|
||||
assert [f["name"] for f in fields] == ["start"]
|
||||
|
||||
conflicting = block("SELECT 1 AS gmv FROM orders", [dict(METRIC)],
|
||||
fields=[{**START, "required": False}])
|
||||
with pytest.raises(ValueError, match="start"):
|
||||
draftplan.build_graph([same, conflicting], name="x")
|
||||
|
||||
|
||||
|
||||
# ==========================================================================
|
||||
# plan:注入假模型
|
||||
# ==========================================================================
|
||||
@@ -225,3 +305,35 @@ async def test_plan_passes_uncovered_through():
|
||||
model = FakeModel(json.dumps({**raw(), "uncovered": ["月活买家"]}, ensure_ascii=False))
|
||||
result = await draftplan.plan(None, docs=[], chunks={}, model=model)
|
||||
assert result["uncovered"] == ["月活买家"]
|
||||
|
||||
|
||||
async def test_plan_tells_the_model_which_metrics_were_already_drafted():
|
||||
"""续跑时把已经起草过的指标写进提示词,模型才会换下一组,不会重复起草前几轮。"""
|
||||
model = FakeModel(json.dumps(raw(), ensure_ascii=False))
|
||||
await draftplan.plan(None, docs=[], chunks={}, model=model, covered=["销售额", "订单数"])
|
||||
human = model.messages[-1].content
|
||||
assert "已经起草过的指标" in human
|
||||
assert "销售额、订单数" in human
|
||||
assert "不要重复起草" in model.messages[0].content
|
||||
|
||||
|
||||
async def test_plan_reports_document_and_catalog_truncation_separately():
|
||||
"""文档被截(doc_truncated)和表目录被截(catalog_truncated)要分开说:改文档、改目录是两件事。"""
|
||||
text = json.dumps(raw(), ensure_ascii=False)
|
||||
long_tables = "t" * (draftplan.MAX_TABLES_CHARS + 100)
|
||||
result = await draftplan.plan(None, docs=[], chunks={}, model=FakeModel(text), tables_text=long_tables)
|
||||
assert result["catalog_truncated"] is True
|
||||
assert result["doc_truncated"] is False
|
||||
assert result["truncated"] is True
|
||||
|
||||
chunks = {"c1": {"ordinal": 1, "content": CHUNK, "document_id": "d1"},
|
||||
"c2": {"ordinal": 2, "content": "长" * (draftplan.MAX_INPUT_CHARS + 100), "document_id": "d1"}}
|
||||
result = await draftplan.plan(None, docs=[], chunks=chunks, model=FakeModel(text), tables_text="表 orders")
|
||||
assert result["doc_truncated"] is True
|
||||
assert result["catalog_truncated"] is False
|
||||
assert result["truncated"] is True
|
||||
|
||||
result = await draftplan.plan(None, docs=[], chunks=CHUNKS, model=FakeModel(text), tables_text="表 orders")
|
||||
assert result["doc_truncated"] is False
|
||||
assert result["catalog_truncated"] is False
|
||||
assert result["truncated"] is False
|
||||
|
||||
Reference in New Issue
Block a user