假设您有一个排序列——比如 id——那么您可以在 SQL Server 2012 中执行以下操作:

Assuming you have an ordering column -- say id -- then you can do the following in SQL Server 2012:

select col,
       col - coalesce(lag(col) over (order by id), 0) as diff
from t;

在早期版本的 SQL Server 中，您可以使用相关子查询执行几乎相同的操作:

In earlier versions of SQL Server, you can do almost the same thing using a correlated subquery:

select col,
       col - isnull((select top 1 col
                     from t t2
                     where t2.id < t.id
                     order by id desc
                    ), 0)
from t

这使用 isnull() 而不是 coalesce() 因为 SQL Server 中的错误"在使用 coalesce() 时会计算第一个参数两次代码>.

This uses isnull() instead of coalesce() because of a "bug" in SQL Server that evaluates the first argument twice when using coalesce().

你也可以用 row_number() 做到这一点:

You can also do this with row_number():

with cte as (
      select col, row_number() over (order by id) as seqnum
      from t
     )
select t.col, t.col - coalesce(tprev.col, 0) as diff
from cte t left outer join
     cte tprev
     on t.seqnum = tprev.seqnum + 1;

所有这些都假设您有一些用于指定排序的列.它可能是 id，或创建日期或其他内容.SQL 表本质上是无序的，因此没有指定排序的列就没有前一行"这样的东西.

All of these assume that you have some column for specifying the ordering. It might be an id, or a creation date or something else. SQL tables are inherently unordered, so there is no such thing as a "previous row" without a column specifying the ordering.