26ai新特性实战：WITH子句嵌套

胖头鱼的鱼缸

发布于 2026-07-07 14:50:09

特性介绍

Oracle AI Database 23.26.2 解除了此前对 WITH 子句嵌套的限制，现在支持在其他 WITH 子句内部嵌套 WITH 子句，并允许在 WITH 子句有效的任何位置使用。

此前的限制：

WITH 子句（CTE，公共表表达式）只能在查询的最外层定义
子查询内部不能使用 WITH 子句
WITH 子句内部不能再嵌套 WITH 子句

新特性允许：

在 WITH 子句的 SELECT 主体中定义内层 WITH 子句
在 FROM 子句的子查询中使用 WITH 子句
在 INSERT … SELECT 中使用嵌套 WITH 子句
内层 WITH 子句引用外层 WITH 子句的数据（关联引用）
多层嵌套（三层及以上）

仍然不支持的：

递归嵌套（WITH 内部直接引用自身）
嵌套 WITH 子句内部定义 PL/SQL 函数
嵌套 WITH 子句内部使用表值构造器（Table Value Constructors）

关于内联提示：现有提示（inline/materialize）仍可使用，但在嵌套场景下可能被忽略，不影响非嵌套查询块的行为。

适用场景

复杂报表：将多步骤的数据转换逻辑分解为层次化的 CTE，提高可读性
数据管道：在 INSERT … SELECT 中使用嵌套 WITH 实现复杂的数据加工
分析查询：对中间结果进行排名后分类，避免重复子查询
代码复用：将公共计算逻辑封装在内层 WITH 中，供外层多处引用
迁移兼容：从 PostgreSQL、SQL Server 等支持嵌套 CTE 的数据库迁移

实战演示

同之前一样，使用用户NFTEST，已授权DB_DEVELOPER_ROLE角色，测试操作均在NFTEST用户下执行。

创建测试表并添加测试数据

CREATE TABLE nwc_sales (
  sale_id   NUMBER PRIMARY KEY,
  product   VARCHAR2(30),
  region    VARCHAR2(30),
  amount    NUMBER,
  sale_date DATE
);

INSERT INTO nwc_sales VALUES (1, 'Widget', 'East', 100, DATE'2026-01-15');
INSERT INTO nwc_sales VALUES (2, 'Widget', 'West', 150, DATE'2026-01-20');
INSERT INTO nwc_sales VALUES (3, 'Gadget', 'East', 200, DATE'2026-02-10');
INSERT INTO nwc_sales VALUES (4, 'Gadget', 'West', 120, DATE'2026-02-15');
INSERT INTO nwc_sales VALUES (5, 'Widget', 'East',  80, DATE'2026-03-01');
INSERT INTO nwc_sales VALUES (6, 'Gadget', 'West', 300, DATE'2026-03-05');

COMMIT;

测试1：基本测试

在一个WITH子句中嵌套的WITH子句

WITH outer_cte AS (
WITH inner_cte AS (
SELECT product, region, SUM(amount) AS region_total
FROM nwc_sales
GROUP BY product, region
  )
SELECT product, SUM(region_total) AS total,
MAX(region_total) AS top_region_total
FROM inner_cte
GROUP BY product
)
SELECT product, total, top_region_total
FROM outer_cte
ORDER BY total DESC;

测试2：多层嵌套

3层：

第一级：按产品和区域分类的销售
第二级：对产品内的区域进行排名（包含嵌套的WITH语句）

WITH level1 AS (
SELECT product, region, SUM(amount) AS region_total
FROM nwc_sales GROUP BY product, region
),
level2 AS (
WITH region_rank AS (
SELECT product, region, region_total,
RANK() OVER (PARTITIONBY product ORDER BY region_total DESC) AS rnk
FROM level1
  )
SELECT product, region, region_total, rnk
FROM region_rank
)
SELECT product, region, region_total,
CASEWHEN rnk = 1 THEN'TOP' ELSE 'OTHER' END AS tier
FROM level2
ORDE RBY product, rnk;

测试3：在带有相关性的FROM子句子查询中嵌套使用WITH

WITH outer_cte AS (
SELECT product, SUM(amount) AS total
FROM nwc_sales GROUP BY product
)
SELECT sub.product, sub.total, sub.sale_count
FROM (
WITH inner_cte AS (
SELECT product, COUNT(*) AS sale_count
FROM nwc_sales GROUP BY product
  )
SELECT o.product, o.total, i.sale_count
FROM outer_cte o JOIN inner_cte i ON o.product = i.product
) sub
ORDER BY sub.total DESC;

测试4：在insert into中嵌套WITH子句

CREATETABLE nwc_summary (
  product   VARCHAR2(30),
  total     NUMBER,
  avg_sale  NUMBER,
  num_sales NUMBER
);

INSERT INTO nwc_summary
WITH product_stats AS (
WITH sale_details AS (
SELECT product, SUM(amount) AS total, AVG(amount) AS avg_amt, COUNT(*) AS num_sales
FROM nwc_sales GROUP BY product
  )
SELECT * FROM sale_details
)
SELECT product, total, avg_amt, num_sales FROM product_stats;

COMMIT;

-- 验证数据
SELECT product, total, avg_sale, num_sales
FROM nwc_summary;