mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 05:01:57 +00:00
SQL 差异说明收尾（第 2 轮）：PostgreSQL 符号运算符和 || 混用不判等价；条件写在聚合里、时间窗口、反连接、经主键定死的列不再说错
- 保真守卫（_misread）：PostgreSQL 写成符号、sqlglot 却当成 LIKE 那一级来读的运算符（~、~*、~~、~~*、@@、-|-、@> 等） 和别的二元运算混用时拦下（PostgreSQL 里它们和 ||、& 同级、从左往右，status ~ 'p' || 'x' 是 (status ~ 'p') || 'x'）； 另拦 PostgreSQL 的前缀 @、|/、||/、COLLATE 后面的 ::；不知道是哪种库时拦块注释里套着的 /*、紧跟着字的 --、#、双引号 括起来的名字（MySQL 默认的 sql_mode 里是字符串）。都只给兜底，附原文的逐行对照。 - 条件写在聚合的参数里（SUM(status = 'paid')、COUNT(… OR NULL)、COUNT(NULLIF(…))、SUM(amount * (1 - is_deleted))） 算作这一边对这一列有条件（_Usage 的 case），另一边多出同一列上的 WHERE 条件时不再说「这一边没有这个条件」，只给兜底。 - 时间：timed 认 BETWEEN 和时间列上 LIKE 像日期的模式（paid_at LIKE '2026-09%'），时间窗口不再被说成有效行过滤； 「另一边没有按时间筛选」之前看另一边可能是时间窗口的条件（聚合参数里的时间比较、date_key、stat_month 这类列、 20260901 / 202609 这类整数、上下界一对），有就不说；认不准是不是时间列的窗口（_window_like）不说成有效行过滤。 两边的时间列在不同的表上时写成「表.列」。 - 反连接：用到 LEFT JOIN 可空一侧的列的条件（WHERE r.id IS NULL、r.order_id IS NULL）不说成有效行过滤，时间列在 可空一侧时也不说按它筛时间。 - 给了表结构时，聚合参数所在的表的键在组里定死了（按 u.id 分组时的 u.city），MAX 对 MIN、AVG(DISTINCT) 对 AVG 不说。 - 另一边原文也 LEFT JOIN 了、规范化因为结果不怕重复去掉了的表，不说「另一边没有关联」。 - 回归语料：收尾第 1 轮之后复验报的全部 blocker / major 收进 sql_pairs（_DIFF_R5、_DIFF_TIME_R5、_SAME_HARD_R5、 _SAME_SCHEMA_R5 执行；REGRESSIONS_R5、EXPLAIN_TRAPS_R5 只查结论），新增对应的测试。 - 模块说明照实补上：新增的守卫、时间条件的认法、只给兜底的情况、收窄换来的代价、MySQL 按默认 sql_mode 读的假设。 认出等价的比例：语料（SQLite，给表结构）169 对里 157 对，和改动前一样；PostgreSQL、MySQL 也不变；不知道是哪种库时 少认 1 对（same:quoted-names，双引号括起来的名字）。
This commit is contained in:
ThearchyHelios
@@ -1439,6 +1439,14 @@ def test_differences_that_cannot_change_the_result_are_not_described(a, b):
|
||||
("SELECT COUNT(DISTINCT status) FROM orders WHERE id = 5", "SELECT COUNT(status) FROM orders WHERE id = 5"),
|
||||
("SELECT o.id FROM orders o LEFT JOIN users u ON u.id = o.user_id", "SELECT o.id FROM orders o"),
|
||||
("SELECT COUNT(*) FROM orders o JOIN users u ON u.id = o.user_id", "SELECT COUNT(*) FROM orders o"),
|
||||
# 第 5 轮：按主键分组（或者经关联条件等于主键），同一组里这张表的列只有一个值——LEFT JOIN 把行复制了也一样
|
||||
("SELECT u.id, MAX(u.city), COUNT(o.id) FROM users u LEFT JOIN orders o ON o.user_id = u.id GROUP BY u.id",
|
||||
"SELECT u.id, MIN(u.city), COUNT(o.id) FROM users u LEFT JOIN orders o ON o.user_id = u.id GROUP BY u.id"),
|
||||
("SELECT o.id, AVG(DISTINCT o.amount), COUNT(r.id) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id "
|
||||
"GROUP BY o.id",
|
||||
"SELECT o.id, AVG(o.amount), COUNT(r.id) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id GROUP BY o.id"),
|
||||
("SELECT o.user_id, MAX(u.name), SUM(o.amount) FROM orders o JOIN users u ON u.id = o.user_id GROUP BY o.user_id",
|
||||
"SELECT o.user_id, MIN(u.name), SUM(o.amount) FROM orders o JOIN users u ON u.id = o.user_id GROUP BY o.user_id"),
|
||||
], ids=lambda v: None)
|
||||
def test_constraints_only_make_the_explanation_say_less(a, b):
|
||||
"""表结构里的约束不拿来判等价，只拿来少说：主键不为空（id IS NOT NULL 恒为真）、主键定下来每组只有一行（SUM 和
|
||||
@@ -1453,3 +1461,161 @@ def test_constraints_only_make_the_explanation_say_less(a, b):
|
||||
diff = explain_sql_diff(x, y, "sqlite")
|
||||
check_shape(diff)
|
||||
assert any(d.aspect != "other" for d in diff.items), (x, y, diff.items)
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------
|
||||
# 第 5 轮：PostgreSQL 写成符号的运算符、各库读法不同的注释；条件写在聚合里、时间窗口、反连接、经主键定下来的列
|
||||
# --------------------------------------------------------------------------
|
||||
|
||||
|
||||
@pytest.mark.parametrize("dialect,sql", [
|
||||
("postgres", "SELECT id, status ~ 'p' || 'x' FROM orders"),
|
||||
("postgres", "SELECT id, status ~* 'p' || 'x' FROM orders"),
|
||||
("postgres", "SELECT id, status ~~ 'p%' || 'x' FROM orders"),
|
||||
("postgres", "SELECT id, status ~~* 'P%' || 'x' FROM orders"),
|
||||
("postgres", "SELECT id, to_tsvector(status) @@ 'paid' || '' FROM orders"),
|
||||
("postgres", "SELECT id, tags @> 'a' || 'b' FROM orders"),
|
||||
("postgres", "SELECT id, @ amount - 20 FROM orders"),
|
||||
("postgres", "SELECT id, |/ amount FROM orders"),
|
||||
("postgres", "SELECT id, status COLLATE \"C\"::text FROM orders"),
|
||||
(None, "SELECT id /* a /* b */ + 1 -- */\nFROM orders"),
|
||||
(None, "SELECT id, amount--1\nFROM orders"),
|
||||
(None, "SELECT COUNT(*) FROM orders WHERE \"status\" = 'paid'"),
|
||||
], ids=["regex-concat", "iregex-concat", "like-op-concat", "ilike-op-concat", "text-search-concat", "contains-concat",
|
||||
"absolute-value", "square-root", "collate-cast", "unknown-nested-comment", "unknown-dashes",
|
||||
"unknown-double-quotes"])
|
||||
def test_operators_postgres_reads_differently_from_sqlglot_are_held_back(dialect, sql):
|
||||
"""PostgreSQL 里 ~、~*、~~、~~*、@@、@> 这些写成符号的运算符和 ||、&、# 同级（「别的运算符」）、从左往右结合，sqlglot
|
||||
把它们当成 LIKE、IN 那一级、比 || 松：status ~ 'p' || 'x' 在 PostgreSQL 里是 (status ~ 'p') || 'x'。前缀的 @ 是取
|
||||
绝对值（sqlglot 读成参数），|/ 是双精度的平方根（sqlglot 写成 SQRT），x COLLATE "C"::text 的 :: 套在整个 COLLATE 上。
|
||||
不知道是哪种库时，块注释里的 /*、紧跟着字的 --、双引号括起来的名字（MySQL 默认的 sql_mode 里是字符串）各库读法不同。
|
||||
这些都不在语法树上判等价。"""
|
||||
from app.data.sql_semantics import _lost_in_parsing, _parse, _read
|
||||
|
||||
read = _read(dialect)
|
||||
tree, error = _parse(sql, read)
|
||||
assert error is None and _lost_in_parsing(sql, read, tree)
|
||||
|
||||
|
||||
@pytest.mark.parametrize("dialect,sql", [
|
||||
("postgres", "SELECT id FROM orders WHERE status LIKE 'p' || '%'"),
|
||||
("postgres", "SELECT id FROM orders WHERE status ILIKE 'p' || '%' AND amount > 1"),
|
||||
("postgres", "SELECT id FROM orders WHERE status ~ ('p' || 'x')"),
|
||||
("postgres", "SELECT id FROM orders WHERE status ~ '^p' AND amount + 1 > 2"),
|
||||
("postgres", "SELECT id, status::text || 'x' FROM orders WHERE status COLLATE \"C\" > 'a'"),
|
||||
("sqlite", "SELECT id /* a /* b */ + 1 -- */\nFROM orders"),
|
||||
("postgres", "SELECT id /* a /* b */ + 1 -- */\nFROM orders"),
|
||||
("mysql", "SELECT id /* a /* b */ + 1 -- */\nFROM orders"),
|
||||
("sqlite", "SELECT id, amount--1\nFROM orders"),
|
||||
("postgres", "SELECT id, amount--1\nFROM orders"),
|
||||
("postgres", "SELECT COUNT(*) FROM orders WHERE \"status\" = 'paid'"),
|
||||
("mysql", "SELECT COUNT(*) FROM orders WHERE status = \"paid\""),
|
||||
], ids=["pg-like-concat", "pg-ilike-concat", "pg-regex-parenthesized", "pg-regex-own-segment", "pg-cast-collate-apart",
|
||||
"sqlite-comment", "pg-comment", "mysql-comment", "sqlite-dashes", "pg-dashes", "pg-double-quotes",
|
||||
"mysql-double-quotes"])
|
||||
def test_writings_each_engine_reads_like_sqlglot_are_not_held_back(dialect, sql):
|
||||
"""写成关键字的 LIKE / ILIKE 本来就比 || 松（PostgreSQL 和 sqlglot 读得一样）；加了括号、AND 隔开的不算混用；知道是
|
||||
SQLite、PostgreSQL、MySQL 时，注释怎么嵌套、-- 后面要不要空白、双引号括起来的是名字还是字符串，sqlglot 和库读得一样：
|
||||
照常规范化，加个空格照样认得出等价。"""
|
||||
from app.data.sql_semantics import _lost_in_parsing, _parse, _read
|
||||
|
||||
read = _read(dialect)
|
||||
tree, error = _parse(sql, read)
|
||||
assert error is None and not _lost_in_parsing(sql, read, tree)
|
||||
assert explain_sql_diff(sql, sql.replace(" FROM", " FROM"), dialect).equivalent
|
||||
|
||||
|
||||
@pytest.mark.parametrize("pid,expected", [
|
||||
("diff:r5-between-on-two-time-columns-grouped-by-date", "A 按 paid_at 筛时间，B 按 created_at"),
|
||||
("diff:r5-between-on-two-time-columns-top-n", "A 按 paid_at 筛时间，B 按 created_at"),
|
||||
("diff:r5-between-window-on-one-side-grouped-by-date", "A 没有按时间筛选，B 按 paid_at 筛时间"),
|
||||
("diff:r5-like-month-vs-no-window", "A 没有按时间筛选，B 按 paid_at 筛时间"),
|
||||
("diff:r5-range-vs-like-month-on-another-column",
|
||||
"A 按 paid_at 筛时间，B 按 created_at（A 是 paid_at >= '2026-09-01' AND paid_at < '2026-10-01'，"
|
||||
"B 是 created_at LIKE '2026-09%'）"),
|
||||
])
|
||||
def test_time_windows_written_as_between_or_like_are_told_as_time_columns(pid, expected):
|
||||
"""时间列上的 BETWEEN、LIKE '2026-09%'（SQLite 里日期存成文字，取月常这么写）是统计窗口：说成按哪一列筛时间，不说成
|
||||
「只保留 … 的行」（那是有效行过滤，会写进数据目录）。结果看行先后的一段（按日期表达式分组、取前 N 行）里 BETWEEN
|
||||
不拆成两个比较，照样认得出。"""
|
||||
pair = sql_pairs.PAIRS_BY_ID[pid]
|
||||
for schema in (None, sql_pairs.schema_cache()):
|
||||
diff = explain_sql_diff(pair.a, pair.b, "sqlite", schema=schema)
|
||||
check_shape(diff)
|
||||
assert [(d.aspect, d.text) for d in diff.items] == [("time", expected)], diff.items
|
||||
|
||||
|
||||
@pytest.mark.parametrize("a,b,expected", [
|
||||
# 聚合的参数就是那一列（不是写在参数里的条件）：另一边确实没有按 status 筛
|
||||
("SELECT region, SUM(amount) FROM orders GROUP BY region",
|
||||
"SELECT region, SUM(amount) FROM orders WHERE status = 'paid' GROUP BY region",
|
||||
[("filter", "B 只保留 status = 'paid' 的行，A 没有这个条件")]),
|
||||
("SELECT SUM(amount) FROM orders", "SELECT SUM(amount) FROM orders WHERE amount > 0",
|
||||
[("filter", "B 只保留 amount > 0 的行，A 没有这个条件")]),
|
||||
# 只是选出来给人看的标记（不聚合，行都在结果里）：另一边确实把别的行也列出来了
|
||||
("SELECT id, status = 'paid' AS is_paid FROM orders",
|
||||
"SELECT id, status = 'paid' AS is_paid FROM orders WHERE status = 'paid'",
|
||||
[("filter", "B 只保留 status = 'paid' 的行，A 没有这个条件")]),
|
||||
# 判空、按状态筛这类和日期无关的条件照说；时间列上的判空是有效行过滤，不是时间窗口
|
||||
("SELECT COUNT(*) FROM orders", "SELECT COUNT(*) FROM orders WHERE paid_at IS NOT NULL",
|
||||
[("filter", "B 排除了 paid_at 为空的行，A 没有")]),
|
||||
("SELECT COUNT(*) FROM orders", "SELECT COUNT(*) FROM orders WHERE grade = 3",
|
||||
[("filter", "B 只保留 grade = 3 的行，A 没有这个条件")]),
|
||||
], ids=lambda v: None)
|
||||
def test_filters_the_other_side_really_does_not_have_are_still_told(a, b, expected):
|
||||
"""收窄之后的对照：另一边确实没有这个条件时照说。"""
|
||||
diff = explain_sql_diff(a, b, "sqlite")
|
||||
check_shape(diff)
|
||||
assert [(d.aspect, d.text) for d in diff.items] == expected, diff.items
|
||||
|
||||
|
||||
@pytest.mark.parametrize("a,b", [
|
||||
# 条件、时间窗口写在聚合的参数里：这一边也只算满足条件的行
|
||||
("SELECT region, SUM(CASE WHEN status = 'paid' THEN 1 ELSE 0 END) FROM orders GROUP BY region",
|
||||
"SELECT region, COUNT(*) FROM orders WHERE status = 'paid' GROUP BY region"),
|
||||
("SELECT region, SUM(IIF(status = 'paid', amount, 0)) FROM orders GROUP BY region",
|
||||
"SELECT region, SUM(amount) FROM orders WHERE status = 'paid' GROUP BY region"),
|
||||
("SELECT region, AVG(status = 'paid') FROM orders GROUP BY region",
|
||||
"SELECT region, AVG(status = 'paid') FROM orders WHERE status = 'paid' GROUP BY region"),
|
||||
("SELECT region, SUM(amount * is_paid) FROM orders GROUP BY region",
|
||||
"SELECT region, SUM(amount) FROM orders WHERE is_paid = 1 GROUP BY region"),
|
||||
("SELECT SUM(amount) FILTER (WHERE created_at >= '2026-09-01') FROM orders",
|
||||
"SELECT SUM(amount) FROM orders WHERE paid_at >= '2026-09-01'"),
|
||||
# 分不清是不是日期的列上的窗口（业务日期键、统计月份、分区）：既不说成有效行过滤，也不说另一边「没有按时间筛选」
|
||||
("SELECT SUM(amount) FROM sales WHERE stat_month = 202609", "SELECT SUM(amount) FROM sales"),
|
||||
("SELECT SUM(amount) FROM sales WHERE pt = '20260901'", "SELECT SUM(amount) FROM sales"),
|
||||
("SELECT SUM(amount) FROM sales WHERE biz_key BETWEEN 20260901 AND 20260930",
|
||||
"SELECT SUM(amount) FROM sales WHERE created_at >= '2026-09-01' AND created_at < '2026-10-01'"),
|
||||
("SELECT SUM(amount) FROM sales WHERE fiscal_year = 2026",
|
||||
"SELECT SUM(amount) FROM sales WHERE created_at >= '2026-01-01'"),
|
||||
# 另一边对别的列有范围（BETWEEN、上下界一对）：可能是认不出的时间窗口，不说「没有按时间筛选」
|
||||
("SELECT SUM(amount) FROM sales WHERE day_no BETWEEN 20697 AND 20726",
|
||||
"SELECT SUM(amount) FROM sales WHERE created_at >= '2026-09-01' AND created_at < '2026-10-01'"),
|
||||
# 反连接：可空一侧的列为空说的是「没关联上」
|
||||
("SELECT COUNT(*) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id WHERE r.state IS NULL",
|
||||
"SELECT COUNT(*) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id"),
|
||||
("SELECT COUNT(*) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id WHERE r.refunded_at IS NULL",
|
||||
"SELECT COUNT(*) FROM orders o LEFT JOIN refunds r ON r.order_id = o.id WHERE r.id IS NULL"),
|
||||
], ids=lambda v: None)
|
||||
@pytest.mark.parametrize("dialect", DIALECTS)
|
||||
def test_conditions_the_other_side_has_in_another_form_are_not_called_missing(a, b, dialect):
|
||||
"""第 5 轮：另一边其实有这个条件（写在聚合的参数里、写在 CASE / IIF / FILTER 里、乘上 0 / 1 的标记），或者这个条件
|
||||
可能是认不出的时间窗口，或者是反连接——说「另一边没有这个条件」「另一边没有按时间筛选」就说错了，只给兜底。"""
|
||||
for x, y in ((a, b), (b, a)):
|
||||
diff = explain_sql_diff(x, y, dialect)
|
||||
check_shape(diff)
|
||||
assert not diff.equivalent and diff.items == [FacetDiff("other", None, None, FALLBACK)], (x, y, diff.items)
|
||||
|
||||
|
||||
def test_time_columns_on_different_tables_are_written_with_their_tables():
|
||||
"""两边的时间列在不同的表上：光写「B 按 created_at」读起来像订单的下单时间，其实是用户的注册时间。写成「表.列」；
|
||||
说「另一边没有按时间筛选」的一边关联了好几张表时也写表名。detail 照旧给各自的表。"""
|
||||
a = "SELECT COUNT(*) FROM orders o JOIN users u ON u.id = o.user_id WHERE o.paid_at >= '2026-09-01'"
|
||||
b = "SELECT COUNT(*) FROM orders o JOIN users u ON u.id = o.user_id WHERE u.created_at >= '2026-09-01'"
|
||||
plain = "SELECT COUNT(*) FROM orders o JOIN users u ON u.id = o.user_id"
|
||||
[item] = explain_sql_diff(a, b, "sqlite", names={"users.created_at": "注册时间"}).items
|
||||
assert item.text == "A 按 orders.paid_at 筛时间，B 按注册时间 users.created_at"
|
||||
assert item.detail == {"a": [{"table": "orders", "column": "paid_at"}],
|
||||
"b": [{"table": "users", "column": "created_at"}]}
|
||||
[item] = explain_sql_diff(b, plain, "sqlite").items
|
||||
assert (item.aspect, item.text) == ("time", "A 按 users.created_at 筛时间，B 没有按时间筛选")
|
||||
|
||||
Reference in New Issue
Block a user