mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-16 02:52:00 +00:00
数据目录闭环第一步(后端 A)复核修正:模型提议的有效记录条件存整理过的写法——原文照旧(NOT IN、IS NOT NULL 不被改写),只把列换成表结构里的真名、不带表名、按数据源的方言要加引号的加引号,剖析核对的就是存下的这一条。以前存的是模型的原文,核对时却按真名加了引号:PostgreSQL 上混合大小写的列("OrderStatus")存成 orderstatus <> 'X',一跑就报列不存在,SQL 检查还会让模型照抄。列名先照原样对表结构,只差大小写的几列写不清是哪一列的不收;加了引号超过 500 字的不收(写进去会让整张表写入失败);不合规的说法用界面上的叫法「有效记录条件」
This commit is contained in:
@@ -1738,6 +1738,8 @@ def _clean_text(value: Any, limit: int = _TEXT_MAX) -> str | None:
|
||||
|
||||
#: 模型给的依据(有效行过滤、去重规则的 reason)写进这一项的 note
|
||||
FACT_REASON_NOTE = "依据:{reason}"
|
||||
#: 有效行过滤按数据源的写法整理(列名加了引号)以后超过目录里这一项的长度上限
|
||||
FACT_FILTER_TOO_LONG = "{label}按数据源的写法整理后超过 {limit} 字"
|
||||
|
||||
|
||||
def _model_fact(raw: Any, key: str) -> tuple[str, str | None] | None:
|
||||
@@ -1765,17 +1767,24 @@ def _model_facts(entry: Mapping[str, Any], meta: Mapping[str, Any], *, kind: str
|
||||
"""模型提议的有效行过滤、去重规则(llm / proposed,依据写进 note)。
|
||||
|
||||
有效行过滤按数据源的方言解析成一个条件、只能用这张表的列(sqlcheck.parse_valid_filter),不过就丢掉、日志记一条,
|
||||
不算这张表失败;存的是模型写的原文(只去掉首尾空白,字符串里的空格不动)。去重规则是一句话,同文字项的规矩。"""
|
||||
不算这张表失败。存的是 sqlcheck.stored_valid_filter 整理过的写法:原文照旧,只把列换成表结构里的真名、不带
|
||||
表名,按数据源的需要加引号——剖析核对的、SQL 检查让模型照着加的都是这一条。去重规则是一句话,同文字项的规矩。"""
|
||||
from app.data import sqlcheck
|
||||
|
||||
out: dict[str, Any] = {}
|
||||
found = _model_fact(entry.get("valid_filter"), "condition")
|
||||
if found is not None:
|
||||
text, reason = found
|
||||
problem = sqlcheck.valid_filter_problem(text, meta, kind=kind, table=table)
|
||||
parsed = sqlcheck.parse_valid_filter(text, meta, kind=kind, table=table)
|
||||
problem = parsed.problem
|
||||
if problem is None:
|
||||
out["valid_filter"] = _fact_item(text, reason)
|
||||
else:
|
||||
stored = sqlcheck.stored_valid_filter(parsed, kind)
|
||||
if _text_ok(stored):
|
||||
out["valid_filter"] = _fact_item(stored, reason)
|
||||
else:
|
||||
# 加了引号以后超过上限:写进去会让整张表写入失败(validate_notes),不收
|
||||
problem = FACT_FILTER_TOO_LONG.format(label=TABLE_FIELD_LABEL["valid_filter"], limit=_TEXT_MAX)
|
||||
if problem is not None:
|
||||
import logging
|
||||
|
||||
logging.getLogger(__name__).info("起草数据目录:表 %s 的有效行过滤没有通过校验,已丢掉:%s(%s)",
|
||||
|
||||
@@ -79,10 +79,11 @@
|
||||
一段被缩小的上限分开成几块、被拆成两半,块数就多几块;一块问完(写没写出来都算)算结束一块。
|
||||
|
||||
**防错事实。** 表级的项里模型可以提议有效行过滤(valid_filter)和去重规则(dedup),各附一句依据(写进这一项的 note),
|
||||
校验见 catalog._from_model(有效行过滤按方言解析,不过就丢掉)。开了数据剖析的源(catalog_profile.profile_settings):
|
||||
目录里记下的码值交给模型做背景(遮罩的列不给,「已确认的目录项」里遮罩列的码值同样不给),表级那段写完后经剖析的只读
|
||||
守卫数一数有效行过滤排除了多少行(catalog_profile.check_valid_filter),排除 0% 或 100% 的不收,收下的 note 写上排除
|
||||
比例;没开剖析、连不上的照收,note 写没有核对的原因。
|
||||
校验见 catalog._from_model(有效行过滤按方言解析,不过就丢掉;存的是整理过的写法:列换成表结构里的真名、按方言加引号,
|
||||
剖析核对的也是这一条)。开了数据剖析的源(catalog_profile.profile_settings):目录里记下的码值交给模型做背景(遮罩的列
|
||||
不给,「已确认的目录项」里遮罩列的码值同样不给),表级那段写完后经剖析的只读守卫数一数有效行过滤排除了多少行
|
||||
(catalog_profile.check_valid_filter),排除 0% 或 100% 的不收,收下的 note 写上排除比例;没开剖析、连不上的照收,
|
||||
note 写没有核对的原因。
|
||||
|
||||
交给模型的是表结构(列、类型、主键、外键、唯一约束、注释)和使用者补充的背景;开了数据剖析的源另给目录里记下的码值
|
||||
(剖析取过的取值和含义,遮罩的列不给),别的数据值一概不发。写给模型的文字都在 catalog.py 名字带 _SYSTEM / _PROMPT
|
||||
|
||||
@@ -1428,7 +1428,8 @@ async def check_valid_filter(source: Any, table: str, meta: dict[str, Any] | Non
|
||||
except ValueError as e:
|
||||
return _unchecked(str(e))
|
||||
cap = max(int(settings.sample_size), int(settings.max_scan_rows))
|
||||
cond = sqlcheck.render_valid_filter(parsed, quoted=lambda name: dialect.quote(name) != name)
|
||||
# 和存进目录的是同一条(catalog._model_facts 存的就是这个写法):列换成真名、按方言加引号,别的照原文
|
||||
cond = sqlcheck.canonical_valid_filter(parsed, quote=dialect.quote)
|
||||
cols = ", ".join(dialect.quote(c) for c in parsed.columns)
|
||||
inner = dialect.limit(f"SELECT {cols} FROM {dialect.table(meta, table)}", cap)
|
||||
sql = f"SELECT COUNT(*) AS n, SUM(CASE WHEN {cond} THEN 0 ELSE 1 END) AS x FROM ({inner}) s"
|
||||
|
||||
+102
-15
@@ -1142,7 +1142,9 @@ VALID_FILTER_FUNCTIONS = ("UPPER", "LOWER", "TRIM", "COALESCE")
|
||||
#: 模板占位:条件里不该有,有了也说不清最终拼出来的是什么(注释在解析之后按语法树认,字符串里的 -- 不算)
|
||||
_FILTER_TEMPLATE = re.compile(r"\{\{|\{%")
|
||||
|
||||
_FILTER_PREFIX = "有效行过滤"
|
||||
#: 不合规的说法以这一项在界面上的叫法开头(catalog.TABLE_FIELD_LABEL:「有效记录条件」):没收下的原因会记进起草任务
|
||||
#: 逐表的日志,界面上看得到
|
||||
_FILTER_PREFIX = catalog.TABLE_FIELD_LABEL["valid_filter"]
|
||||
|
||||
|
||||
@dataclass(frozen=True)
|
||||
@@ -1150,7 +1152,9 @@ class ValidFilter:
|
||||
"""一段有效行过滤解析的结果。
|
||||
|
||||
expression:sqlglot 的条件(不合规时为 None);columns:条件里用到的列(表结构里的真名,按出现的先后去重);
|
||||
names:小写列名 → 表结构里的真名;problem:不合规时给人看的一句话,合规是 None;dialect:解析用的 sqlglot 方言。
|
||||
names:条件里写的列名 → 表结构里的真名;problem:不合规时给人看的一句话,合规是 None;dialect:解析用的 sqlglot
|
||||
方言。text:解析的原文(去掉首尾空白);spans:每处列引用(连同表名前缀)在原文里的位置 (起, 止(含), 真名),按
|
||||
书写的先后,有一处拿不到位置时为 None(canonical_valid_filter 退回按语法树重写)。
|
||||
"""
|
||||
|
||||
expression: Any = None
|
||||
@@ -1158,6 +1162,8 @@ class ValidFilter:
|
||||
names: Mapping[str, str] = field(default_factory=dict)
|
||||
problem: str | None = None
|
||||
dialect: str | None = None
|
||||
text: str = ""
|
||||
spans: tuple[tuple[int, int, str], ...] | None = None
|
||||
|
||||
|
||||
def _bad_filter(problem: str, dialect: str | None = None) -> ValidFilter:
|
||||
@@ -1237,7 +1243,10 @@ def parse_valid_filter(condition: Any, meta: Mapping[str, Any] | None, *, kind:
|
||||
|
||||
规矩:一段文字、不超过 VALID_FILTER_MAX 字;不能有分号、模板占位、注释;解析出来恰好是一个条件,只由比较、IN(值的
|
||||
列表)、BETWEEN、LIKE、IS NULL、AND / OR / NOT 和 UPPER / LOWER / TRIM / COALESCE 组成(不能有子查询、聚合、窗口、
|
||||
别的函数、算术、参数占位);列不带表名或带这张表的表名(table),都在表结构里(不分大小写),至少用到一列。
|
||||
别的函数、算术、参数占位);列不带表名或带这张表的表名(table),都在表结构里,至少用到一列。
|
||||
|
||||
列名先照原样对表结构,对不上再不分大小写对;表结构里有只差大小写的几列(PostgreSQL 上 "Status" 和 status 可以
|
||||
并存)、写法又和哪一列都不一样时说不清是哪一列,不收。
|
||||
"""
|
||||
from app.data.names import name_key as _key
|
||||
|
||||
@@ -1268,24 +1277,55 @@ def parse_valid_filter(condition: Any, meta: Mapping[str, Any] | None, *, kind:
|
||||
return _bad_filter(problem, dialect)
|
||||
from sqlglot import exp
|
||||
|
||||
lookup = {str(c.get("name")).lower(): str(c.get("name")) for c in (meta or {}).get("columns") or []
|
||||
if isinstance(c, Mapping) and c.get("name") is not None}
|
||||
real_names = [str(c.get("name")) for c in (meta or {}).get("columns") or []
|
||||
if isinstance(c, Mapping) and c.get("name") is not None]
|
||||
exact = set(real_names)
|
||||
by_lower: dict[str, list[str]] = {}
|
||||
for name in real_names:
|
||||
by_lower.setdefault(name.lower(), []).append(name)
|
||||
own = {_key(str(n).rsplit(".", 1)[-1]) for n in (table, (meta or {}).get("qualified")) if n}
|
||||
used: dict[str, str] = {}
|
||||
spans: list[tuple[int, int, str]] | None = []
|
||||
for column in expression.find_all(exp.Column, bfs=False): # 按书写的先后
|
||||
if column.is_star:
|
||||
return _bad_filter("里不能写 *", dialect)
|
||||
qualifier = column.table
|
||||
if column.args.get("db") or column.args.get("catalog") or (qualifier and _key(qualifier) not in own):
|
||||
return _bad_filter(f"只能引用这张表的列,不能引用 {column.sql(dialect=dialect)}", dialect)
|
||||
real = lookup.get(column.name.lower())
|
||||
if real is None:
|
||||
return _bad_filter(f"里的列「{column.name}」不在表结构里", dialect)
|
||||
used.setdefault(column.name.lower(), real)
|
||||
written = column.name
|
||||
if written in exact:
|
||||
real = written
|
||||
else:
|
||||
found = by_lower.get(written.lower()) or []
|
||||
if not found:
|
||||
return _bad_filter(f"里的列「{written}」不在表结构里", dialect)
|
||||
if len(found) > 1:
|
||||
return _bad_filter(f"里的列「{written}」在表结构里有只差大小写的几列({'、'.join(found)}),"
|
||||
"要照表结构的写法写清是哪一列", dialect)
|
||||
real = found[0]
|
||||
used.setdefault(written, real)
|
||||
span = _column_span(column, text)
|
||||
spans = None if spans is None or span is None else [*spans, (*span, real)]
|
||||
if not used:
|
||||
return _bad_filter("没有用到这张表的任何列", dialect)
|
||||
return ValidFilter(expression=expression, columns=tuple(dict.fromkeys(used.values())), names=used,
|
||||
dialect=dialect)
|
||||
dialect=dialect, text=text, spans=tuple(spans) if spans is not None else None)
|
||||
|
||||
|
||||
def _column_span(column: Any, text: str) -> tuple[int, int] | None:
|
||||
"""一处列引用(连同表名前缀)在原文里的位置 (起, 止(含)):sqlglot 在标识符的 meta 里记下了它在原文里的起止。
|
||||
拿不到、或者那一截原文里认不出这个列名时为 None。"""
|
||||
ident = column.this
|
||||
meta = getattr(ident, "meta", None) or {}
|
||||
start, end = meta.get("start"), meta.get("end")
|
||||
prefix = column.args.get("table")
|
||||
if prefix is not None:
|
||||
start = (getattr(prefix, "meta", None) or {}).get("start")
|
||||
if not (isinstance(start, int) and isinstance(end, int) and 0 <= start <= end < len(text)):
|
||||
return None
|
||||
if not text[start:end + 1].rstrip("\"`]").lower().endswith(str(column.name).lower()):
|
||||
return None
|
||||
return start, end
|
||||
|
||||
|
||||
def valid_filter_problem(condition: Any, meta: Mapping[str, Any] | None, *, kind: str | None,
|
||||
@@ -1296,7 +1336,8 @@ def valid_filter_problem(condition: Any, meta: Mapping[str, Any] | None, *, kind
|
||||
|
||||
def render_valid_filter(parsed: ValidFilter, *, quoted: Any) -> str:
|
||||
"""把解析好的条件按数据源的方言重新写出来,列换成表结构里的真名、不带表名;quoted(真名) 为真的加引号(和剖析查询
|
||||
其余部分的写法一致:SqlDialect.quote 要加引号的才加)。拼进剖析的核对查询用,不改目录里存的原文。"""
|
||||
其余部分的写法一致:SqlDialect.quote 要加引号的才加)。按语法树重写,NOT IN、IS NOT NULL 会写成 NOT x IN、
|
||||
NOT x IS NULL:canonical_valid_filter 拿不到原文里的位置时才退回用它。"""
|
||||
from sqlglot import exp
|
||||
|
||||
if parsed.expression is None:
|
||||
@@ -1304,14 +1345,60 @@ def render_valid_filter(parsed: ValidFilter, *, quoted: Any) -> str:
|
||||
|
||||
def swap(node: Any) -> Any:
|
||||
if isinstance(node, exp.Column):
|
||||
real = parsed.names.get(node.name.lower(), node.name)
|
||||
real = parsed.names.get(node.name, node.name)
|
||||
return exp.Column(this=exp.Identifier(this=real, quoted=bool(quoted(real))))
|
||||
return node
|
||||
|
||||
return parsed.expression.copy().transform(swap).sql(dialect=parsed.dialect)
|
||||
|
||||
|
||||
def canonical_valid_filter(parsed: ValidFilter, *, quote: Any) -> str:
|
||||
"""有效行过滤存进目录、拼进剖析核对查询的写法(两处用的是同一条):原文照旧,只把每处列引用换成 quote(真名)——表结构
|
||||
里的真名、不带表名,按数据源的需要加引号(quote 是 catalog_profile.SqlDialect.quote,和剖析查询其余部分的写法一致)。
|
||||
|
||||
以前目录里存的是模型的原文,核对时却按真名加了引号:PostgreSQL 上混合大小写的列("OrderStatus")不加引号会被折成
|
||||
小写,存下的 orderstatus <> 'X' 一跑就报列不存在,核对过的却能跑通,SQL 检查还会让模型照抄存下的那一句。
|
||||
|
||||
拿不到原文里的位置、换完以后认不出原来那几列时,退回按语法树重写(render_valid_filter)。"""
|
||||
if parsed.expression is None:
|
||||
raise ValueError(parsed.problem or "有效行过滤不合规")
|
||||
if parsed.spans:
|
||||
text = parsed.text
|
||||
for start, end, real in sorted(parsed.spans, reverse=True):
|
||||
text = text[:start] + quote(real) + text[end + 1:]
|
||||
if _same_columns(text, parsed):
|
||||
return text
|
||||
return render_valid_filter(parsed, quoted=lambda name: quote(name) != name)
|
||||
|
||||
|
||||
def _same_columns(text: str, parsed: ValidFilter) -> bool:
|
||||
"""换过列名的原文还是一个条件,列依次是表结构里的那几列、都不带表名。"""
|
||||
import sqlglot
|
||||
from sqlglot import exp
|
||||
|
||||
try:
|
||||
tree = sqlglot.parse_one(text, read=parsed.dialect)
|
||||
except Exception: # noqa: BLE001 - 换坏了就退回按语法树重写
|
||||
return False
|
||||
columns = list(tree.find_all(exp.Column, bfs=False))
|
||||
return ([c.name for c in columns] == [real for _, _, real in parsed.spans or ()]
|
||||
and not any(c.table for c in columns))
|
||||
|
||||
|
||||
def stored_valid_filter(parsed: ValidFilter, kind: str | None) -> str:
|
||||
"""模型提议的有效行过滤存进目录的写法:canonical_valid_filter,按这个数据源的方言加引号(catalog_profile.SqlDialect:
|
||||
SQLAlchemy 的写法,要加才加——PostgreSQL、SQLite 的 "OrderStatus",MySQL 的 `OrderStatus`,Oracle 探查出来的小写名
|
||||
不加)。"""
|
||||
from app.data.catalog_profile import SqlDialect
|
||||
|
||||
try:
|
||||
dialect = SqlDialect(kind)
|
||||
except ValueError:
|
||||
return parsed.text
|
||||
return canonical_valid_filter(parsed, quote=dialect.quote)
|
||||
|
||||
|
||||
__all__ = ["CHECK_CODES", "LEVELS", "MAX_CHECKS", "MAX_SQL_CHARS", "SqlCheck", "SqlChecker", "VALID_FILTER_FUNCTIONS",
|
||||
"VALID_FILTER_MAX", "ValidFilter", "check_sql", "dialect_of", "graph_checks", "graph_sources",
|
||||
"load_checkers", "load_checkers_by_name", "parse_valid_filter", "render_valid_filter", "tool_query",
|
||||
"valid_filter_problem"]
|
||||
"VALID_FILTER_MAX", "ValidFilter", "canonical_valid_filter", "check_sql", "dialect_of", "graph_checks",
|
||||
"graph_sources", "load_checkers", "load_checkers_by_name", "parse_valid_filter", "render_valid_filter",
|
||||
"stored_valid_filter", "tool_query", "valid_filter_problem"]
|
||||
|
||||
@@ -3,6 +3,7 @@
|
||||
- 提示词写清两项只在表结构、注释、码值能看出来时才写;结构化输出、纯文本两条路都认这两项;
|
||||
- 有效行过滤按数据源的方言用 sqlglot 解析成一个条件:只能引用这张表的列,不能有子查询、聚合、窗口函数、分号,只允许
|
||||
比较、IN、BETWEEN、LIKE、IS NULL、AND / OR / NOT 和 UPPER / LOWER / TRIM / COALESCE,不超过 500 字。不过就丢掉;
|
||||
存的是整理过的写法:原文照旧,列换成表结构里的真名、不带表名、按方言加引号(剖析核对的也是这一条);
|
||||
- 开了数据剖析、能连上的源:经剖析的只读守卫数出排除比例,排除 0% 或 100% 的不收,收下的在 note 里写剖析核对的结果;
|
||||
没开剖析、连不上的照收,note 写没有核对的原因;
|
||||
- 开了剖析的源,目录里记下的码值交给模型做背景(遮罩的列不给;「已确认的目录项」里遮罩列的码值也不给);
|
||||
@@ -154,6 +155,54 @@ def test_the_condition_is_rendered_with_the_columns_quoted_like_the_source():
|
||||
assert rendered == "`isTest` = 0 AND status <> 'C'"
|
||||
|
||||
|
||||
#: 带一列混合大小写列名的订单表(ORM 建的 PostgreSQL 库里常见)
|
||||
MIXED = {**META, "columns": [*META["columns"], _col("OrderStatus", "TEXT")]}
|
||||
|
||||
|
||||
@pytest.mark.parametrize("kind, condition, stored", [
|
||||
# PostgreSQL 把没加引号的名字折成小写:混合大小写的列一律加引号,不论模型写的大小写对不对
|
||||
("postgresql", "orderstatus <> 'X' AND STATUS NOT IN ('C', 'X')",
|
||||
"\"OrderStatus\" <> 'X' AND status NOT IN ('C', 'X')"),
|
||||
("postgresql", "OrderStatus <> 'X'", "\"OrderStatus\" <> 'X'"),
|
||||
("postgresql", "orders.\"OrderStatus\" IS NOT NULL", "\"OrderStatus\" IS NOT NULL"),
|
||||
("mysql", "orderstatus <> 'X'", "`OrderStatus` <> 'X'"),
|
||||
("oracle", "ORDERS.STATUS <> 'C'", "status <> 'C'"),
|
||||
# 不用改的照原文:NOT IN、IS NOT NULL 这些写法不被改写
|
||||
("sqlite", "status NOT IN ('C', 'X') AND updated_at IS NOT NULL",
|
||||
"status NOT IN ('C', 'X') AND updated_at IS NOT NULL"),
|
||||
])
|
||||
def test_the_stored_filter_names_its_columns_the_way_the_source_needs(kind, condition, stored):
|
||||
"""存进目录的有效记录条件:原文照旧,只把列换成表结构里的真名、不带表名,按数据源的需要加引号(和剖析核对查询的写法
|
||||
一致)。以前存的是模型的原文:PostgreSQL 上 orderstatus 会被当成另一列,核对时却按真名加了引号,存下的和核对过的
|
||||
不是同一条,SQL 检查还会让模型照抄这一句。"""
|
||||
entry = {"table": "orders", "valid_filter": {"condition": condition, "reason": "注释写明"}}
|
||||
notes = catalog._from_model(entry, MIXED, system=False, source_kind=kind, table="orders")
|
||||
assert notes["valid_filter"]["value"] == stored
|
||||
again = sqlcheck.parse_valid_filter(stored, MIXED, kind=kind, table="orders")
|
||||
assert again.problem is None and sqlcheck.stored_valid_filter(again, kind) == stored, "存下的再整理一遍不变"
|
||||
|
||||
|
||||
def test_a_filter_that_grows_past_the_limit_once_quoted_is_dropped_not_written():
|
||||
"""整理时加了引号、超过 500 字的不收:目录里这一项最多 500 字,写进去会让整张表写入失败。"""
|
||||
condition = "orderstatus IN (" + ", ".join(f"'{i:03d}'" for i in range(66)) + ") OR updated_at IS NULL"
|
||||
assert len(condition) == 499
|
||||
notes = catalog._from_model({"table": "orders", "label": "订单", "valid_filter": condition}, MIXED, system=False,
|
||||
source_kind="postgresql", table="orders")
|
||||
assert "valid_filter" not in notes and notes["label"]["value"] == "订单"
|
||||
assert not catalog.validate_notes(notes)
|
||||
|
||||
|
||||
def test_columns_that_differ_only_in_case_must_be_named_exactly():
|
||||
"""表结构里有只差大小写的两列:写得和其中一列一模一样的认那一列,否则说不清是哪一列,不收。"""
|
||||
meta = {**META, "columns": [*META["columns"], _col("Status", "TEXT")]}
|
||||
problem = sqlcheck.valid_filter_problem("STATUS = 'C'", meta, kind="postgresql", table="orders")
|
||||
assert problem == ("有效记录条件里的列「STATUS」在表结构里有只差大小写的几列(status、Status),"
|
||||
"要照表结构的写法写清是哪一列")
|
||||
for condition, stored in (("\"Status\" = 'C'", "\"Status\" = 'C'"), ("status = 'C'", "status = 'C'")):
|
||||
parsed = sqlcheck.parse_valid_filter(condition, meta, kind="postgresql", table="orders")
|
||||
assert parsed.problem is None and sqlcheck.stored_valid_filter(parsed, "postgresql") == stored
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# 模型给的两项:校验、依据
|
||||
# --------------------------------------------------------------------------
|
||||
@@ -318,6 +367,25 @@ async def test_the_check_counts_excluded_rows_through_the_read_only_guard(scenic
|
||||
assert "COUNT(*)" in seen[0][0] and "status = 1" in seen[0][0]
|
||||
|
||||
|
||||
async def test_the_check_runs_the_condition_as_it_is_stored(scenic, monkeypatch):
|
||||
"""剖析核对的就是存进目录的那一条:表名去掉、列换成表结构里的真名,别的照原文(NOT IN 不被改写成 NOT x IN)。"""
|
||||
seen: list[str] = []
|
||||
real = catalog_profile.guard.check
|
||||
|
||||
def spy(sql, **kw):
|
||||
seen.append(sql)
|
||||
return real(sql, **kw)
|
||||
|
||||
monkeypatch.setattr(catalog_profile.guard, "check", spy)
|
||||
source = await scenic()
|
||||
reply = {"label": "入园记录", "valid_filter": {"condition": "VISITS.STATUS NOT IN (0)", "reason": "0 是作废的检票"}}
|
||||
got = await D.run_model_draft(FakeModel({"visits": reply}), source, ["visits"])
|
||||
vf = got["visits"].notes["valid_filter"]
|
||||
assert vf["value"] == "status NOT IN (0)"
|
||||
assert len(seen) == 1 and f"CASE WHEN {vf['value']} THEN 0 ELSE 1 END" in seen[0]
|
||||
assert "剖析核对:排除" in vf["note"]
|
||||
|
||||
|
||||
@pytest.mark.parametrize("condition", ["status IN (0, 1)", "status = 7"])
|
||||
async def test_a_filter_that_excludes_nothing_or_everything_is_dropped(scenic, condition):
|
||||
source = await scenic()
|
||||
|
||||
Reference in New Issue
Block a user