获取存储过程返回值的几种方式

CREATE PROCEDURE p_test1 
AS 
DECLARE @int int
SET @int = 102400;
RETURN @int;--这里只能返回整型


--执行
DECLARE @p1return INT --声明一个变量
EXECUTE @p1return= p_test1 --使用变量来接收 return回来的值 

SELECT @p1return

CREATE PROCEDURE p_2 @str NVARCHAR(20)=NULL OUTPUT,@int INT=NULL OUTPUT
AS 
SET @str=‘say hi‘;
SET @int =1024;


DECLARE @p1 NVARCHAR(20),@p2 INT --声明两个变量

--执行
EXECUTE p_2 @p1 OUTPUT,@p2 OUTPUT --变量后有OUTPUT
SELECT @p1,@p2

--结果
------------------------
(无列名)    (无列名)
say hi    1024


--执行
EXECUTE p_2 @p1 ,@p2  --注意这里没有OUTPUT
SELECT @p1,@p2

--结果
-----------------------
(无列名)    (无列名)
NULL    NULL

CREATE PROCEDURE p_3
AS 
SELECT 1 id,‘xiaoli‘ name
UNION
SELECT 2 id,‘xiaowang‘ name

--注意如果有两次查询,那么这两次查询返回的数据表的结构必需是一样的.
SELECT 3 id,‘xiaozhang‘ name
UNION
SELECT 4 id,‘xiaozhao‘ name


--执行
--这里使用的是表变量,当然使用临时表,实体表都是可以的.
DECLARE @tab AS  TABLE(
id int 
,name nvarchar(20)
)

INSERT INTO @tab(id,name)
EXECUTE p_3

SELECT * FROM @tab

--结果
----------------------
id    name
1    xiaoli
2    xiaowang
3    xiaozhang
4    xiaozhao

In the life of every software engineer, there should be at least (I’d love to say “at most" actually) one project where bussiness logic is implemented on database side. My current project is one of that kind. We have all logic designed as stored procedures, then there’s a middleware where we use LINQ2SQL for database API bindings, then there are iPhone application and a web-site.

Well, it works. But it’s a hell.

First of all, T-SQL itself is a very primitive low-level language. There are just no features to make components reusable. So, as long as your procedures are just couple of select statements, everything’s fine. But the more execution flows you have to implement, the more unreadable your procedure becomes. I do firmly believe that being extremely pedantic can make things better, but it doesn’t solve the problem completely. It’s just fine to make mistakes when you write a program. In most cases you have an easy way to cover your code with automated tests, at least it’s pretty straighforward in high-level languages. With T-SQL, you just can’t.

In our project, I had to take testing tasks as one of my concerns. I finally came out with idea that the easiest approach technically is just to test the middleware+database together. Test suite is just another client like iPhone application or web-site. This approach proved its adequacy, but the problem is, when Teamcity says things like Test "CanTransferLotsOfMoneys" failed, database guys have no idea what it means.

So, for a long time I’ve been trying to find out a solution to write tests for stored procedure in pure T-SQL. In this case, I could have gotten rid of being the only person to intepret test failure. The major point was, in T-SQL it’s just impossible to access multiple result sets if stored procedure returns them. In our project, there are 2 or may be 3 procedures, which only return 1 result set. That’s the issue.

Always return only one result set

Sure this will never work. Simple example is, I need to get data to render a blog post page. I’ll need:

Current user’s details
Post details
Comments and their details
There are at least 3 result sets.

Use temporary tables

This will probably work, but what do you do if some procedures call themselves recursively?

Table variables

Table variables is a great idea:

declare @Users table(
  UserId int,
  UserName nvarchar(256)
)

exec DoSomething @Users
But unfortunately stored procedures can’t return them.

I finally discovered that T-SQL has features to work with XML. The solution consists of few simple ideas.

There’s a type xml in T-SQL. This type allows storing XML documents and accessing their nodes.
Objects of this type can be either converted to nvarchar() or constructed from it.
Objects of this type can be passed to stored procedures as output parameters.
You can easily generate XML from select queries.
Let’s say we have 2 tables:

create table Blog(
  BlogId int identity(1,1) not null,
  BlogName nvarchar(256)
)

create table Post(
  PostId int identity(1,1) not null,
  BlogId int not null, -- yes, this should be FK
  BlogName nvarchar(256)
)
The task is:

Create a testable stored procedure GetBlogWithDetails @blogId, that will return BlogId, BlogName and all its posts.
Definition of “testable" is: I can access all the data returned and based on this data make some conclusions.
Normally, this procedure would look like this: (no error handling - sorry)

create procedure GetBlogWithDetails(@BlogId int) as 
begin
select BlogId, BlogName 
  from Blog 
  where BlogId = @BlogId

  select PostId, BlogId, PostName 
  from Post 
  where BlogId = @BlogId
end
You can’t access its second result set. The solution is to render same data as XML and return this XML as output parameter.

create procedure GetBlogWithDetailsXml(
  @blogId int, 
  @xml xml output) as 
begin
  -- first, create 2 table variables for blogs and for posts
  declare @blogRows table(BlogId int, BlogName nvarchar(256))
  declare @postRows table(PostId int, BlogId int, PostName nvarchar(256))

  -- (#1) this is where you implement your useful logic
  insert into @blogRows
  select BlogId, BlogName 
  from Blog 
  where BlogId = @blogId

  insert into @postRows
  select PostId, BlogId, PostName 
  from Post
  where BlogId = @blogId

  -- here you render your 2 tables containing useful data as XML
  declare @blogRowsXml xml = (
    select BlogId, BlogName 
    from @blogRows as Blog
    for xml auto)

  declare @postRowsXml xml = (
    select PostId, BlogId, PostName 
    from @postRows as Post
    for xml auto)

  -- here you build a single XML with all the data required
  set @xml = 
    cast(@blogRowsXml as nvarchar(max)) + 
    cast(@postRowsXml as nvarchar(max))

  -- (#2) here you return the data as "raw" result sets.
  select * from @blogRows
  select * from @postRows
end
As you see, “useful" code is only at #1 and #2. The rest is bunch of stuff for XML. You can now play with GetBlogWithDetailsXml like this:

For my test data,

declare @xml xml
exec GetBlogWithDetailsXml 1, @xml output
select @xml
returns XML like this:

<Blog BlogId="1" BlogName="Blog #1" />
<Post PostId="1" BlogId="1" PostName="Post #1 in Blog #1" />
<Post PostId="2" BlogId="1" PostName="Post #2 in Blog #1" />
It’s now pretty easy to extract all the data returned:

select
  Col.value(‘@BlogId‘, ‘int‘) as BlogId,
  Col.value(‘@BlogName‘, ‘nvarchar(256)‘) as BlogName
from @xml.nodes(‘/Blog‘) as Data(Col)

select
  Col.value(‘@PostId‘, ‘int‘) as PostId,
  Col.value(‘@BlogId‘, ‘int‘) as BlogId,
  Col.value(‘@PostName‘, ‘nvarchar(256)‘) as PostName
from @xml.nodes(‘/Post‘) as Data(Col)
with my test data, it returns:

BlogId    BlogName
1    Blog #1
PostId    BlogId    PostName
1    1    Post #1 in Blog #1
2    1    Post #2 in Blog #1
I’m not really sure if returning “raw" result sets even makes sense with this approach. From the viewpoint of LINQ2SQL, yes, you won’t be able to utilize its mapping feature, but deserializing object from XML is a primitive task, so it shouldn’t be an issue.