...这是通用的全SQL解决方案：

SELECT Stack_Counter,value,Stack_Depth 
FROM
  (SELECT *,RANK() OVER (
      PARTITION BY Stack_Depth
      ORDER BY Stack_Counter DESC) rank
   FROM stack)
WHERE rank=1 AND Stack_Depth > 0;

仅使用SQL的解决方案可能比其价值更大的麻烦（由于其复杂性或乏味性），但这是与SQLite一起使用的解决方案：

SELECT * FROM (SELECT * FROM stack WHERE Stack_Depth=1 ORDER BY Stack_Counter DESC LIMIT 1)
UNION ALL
SELECT * FROM (SELECT * FROM stack WHERE Stack_Depth=2 ORDER BY Stack_Counter DESC LIMIT 1)
UNION ALL
SELECT * FROM (SELECT * FROM stack WHERE Stack_Depth=3 ORDER BY Stack_Counter DESC LIMIT 1)
UNION ALL
SELECT * FROM (SELECT * FROM stack WHERE Stack_Depth=4 ORDER BY Stack_Counter DESC LIMIT 1);

在我自己的评论中，我在注释中提到的窗口函数比我想象的要困难得多。 Peak的答案是使用rank（）窗口函数的优雅解决方案，这正是我最初想要的。同时，sqlite还支持recursive CTE（公用表表达式； WITH语句）：

WITH RECURSIVE
     latest (id,level) AS (
          VALUES (20,(SELECT Stack_Depth FROM stack WHERE Stack_Counter = 20))
          UNION ALL
          SELECT (SELECT max(Stack_Counter) 
                  FROM stack
                  WHERE Stack_Depth = level - 1
                        AND Stack_Counter <= 20),level - 1
          FROM latest
          WHERE level - 1 > 0
     )
SELECT stack.*
FROM stack INNER JOIN latest
     ON stack.Stack_Counter = latest.id
ORDER BY stack.Stack_Counter

我必须在三个位置插入所需的堆栈级别，但是如果您从宿主语言中的准备好的语句调用此参数，则可以用命名的SQL参数替换所有这些位置。

如果您对选择特定的Stack_Counter值不感兴趣，而只是希望从整个表中获取结果，则将VALUES子句替换为SELECT之类的

WITH RECURSIVE
     latest (id,level) AS (
          SELECT * FROM (SELECT Stack_Counter,Stack_Depth FROM stack ORDER BY Stack_Counter DESC LIMIT 1)
          UNION ALL
          SELECT (SELECT max(Stack_Counter) 
                  FROM stack
                  WHERE Stack_Depth = level - 1),level - 1
          FROM latest
          WHERE level - 1 > 0
     )
SELECT stack.*
FROM stack INNER JOIN latest
     ON stack.Stack_Counter = latest.id
ORDER BY stack.Stack_Counter