You shouldn't care enough about the physical column ordering of an existing table that you would want to invite the risks involved with changing it. The only case where this might be a concern is if you're using SELECT * and want that output arranged in a certain way - and you shouldn't be doing that, you should be listing the column names explicitly (and you can then put the columns in any order you want, or create a view that does that).
You can drop and re-create the table, like others have said, with the columns in your preferred order. But this is going to be very disruptive (blocking, I/O) if your table is large, since all of the other users on the system will have to wait for that to complete. If you want to see exactly what Management Studio does, create this table:
CREATE TABLE dbo.foo(b INT, a INT, r INT);
INSERT dbo.foo(b,a,r) VALUES(1,2,3),(4,5,6);
Now, in Object Explorer, right-click the table and choose Design. Select the box to the left of the column name that you want to reorder. Drag the column to another location within the table. I moved a above b because I'm OCD and wanted the columns in alphabetical order:
Then I clicked the "Generate Change Script" button, and here is the monstrosity that it created (and note this is a heap with no keys, constraints, triggers etc. - this re-create can get pretty complex).
/*
To prevent any potential data loss issues, you should review
this script in detail before running it outside the context
of the database designer.
*/
BEGIN TRANSACTION
SET QUOTED_IDENTIFIER ON
SET ARITHABORT ON
SET NUMERIC_ROUNDABORT OFF
SET CONCAT_NULL_YIELDS_NULL ON
SET ANSI_NULLS ON
SET ANSI_PADDING ON
SET ANSI_WARNINGS ON
COMMIT
BEGIN TRANSACTION
GO
CREATE TABLE dbo.Tmp_foo
(
a int NULL,
b int NULL,
r int NULL
) ON [PRIMARY]
GO
ALTER TABLE dbo.Tmp_foo SET (LOCK_ESCALATION = TABLE)
GO
IF EXISTS(SELECT * FROM dbo.foo)
EXEC('INSERT INTO dbo.Tmp_foo (a, b, r)
SELECT a, b, r FROM dbo.foo WITH (HOLDLOCK TABLOCKX)')
GO
DROP TABLE dbo.foo
GO
EXECUTE sp_rename N'dbo.Tmp_foo', N'foo', 'OBJECT'
GO
COMMIT
For fun, let's create something a little more complex:
CREATE TABLE dbo.flarb
(
a INT PRIMARY KEY,
b INT NOT NULL CHECK (b>1),
c ROWVERSION,
d DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
e UNIQUEIDENTIFIER NOT NULL DEFAULT NEWSEQUENTIALID()
);
CREATE TABLE dbo.blarf
(
a INT NOT NULL FOREIGN KEY REFERENCES dbo.flarb(a)
);
GO
CREATE TRIGGER dbo.flarb_tr
ON dbo.flarb
FOR UPDATE
AS
SELECT 1;
GO
CREATE VIEW dbo.flarb_v
WITH SCHEMABINDING
AS
SELECT a FROM dbo.flarb;
Now we try the same thing:
'flarb' table
- Error validating the default for column 'e'.
- Warning: The following schema-bound objects will be modified:
- View 'dbo.flarb_v': schema binding will be removed.
Oops. Thankfully, we can still proceed to see what the script will look like:
/*
To prevent any potential data loss issues, you should review
this script in detail before running it outside the context
of the database designer.
*/
BEGIN TRANSACTION
SET QUOTED_IDENTIFIER ON
SET ARITHABORT ON
SET NUMERIC_ROUNDABORT OFF
SET CONCAT_NULL_YIELDS_NULL ON
SET ANSI_NULLS ON
SET ANSI_PADDING ON
SET ANSI_WARNINGS ON
COMMIT
BEGIN TRANSACTION
GO
ALTER VIEW dbo.flarb_v
AS
SELECT a FROM dbo.flarb;GO
ALTER TABLE dbo.flarb
DROP CONSTRAINT DF__flarb__d__1B7E091A
GO
ALTER TABLE dbo.flarb
DROP CONSTRAINT DF__flarb__e__1C722D53
GO
CREATE TABLE dbo.Tmp_flarb
(
a int NOT NULL,
b int NOT NULL,
d datetime NOT NULL,
c timestamp NOT NULL,
e uniqueidentifier NOT NULL
) ON [PRIMARY]
GO
ALTER TABLE dbo.Tmp_flarb SET (LOCK_ESCALATION = TABLE)
GO
ALTER TABLE dbo.Tmp_flarb ADD CONSTRAINT
DF__flarb__d__1B7E091A DEFAULT (getdate()) FOR d
GO
ALTER TABLE dbo.Tmp_flarb ADD CONSTRAINT
DF__flarb__e__1C722D53 DEFAULT (newsequentialid()) FOR e
GO
IF EXISTS(SELECT * FROM dbo.flarb)
EXEC('INSERT INTO dbo.Tmp_flarb (a, b, d, e)
SELECT a, b, d, e FROM dbo.flarb WITH (HOLDLOCK TABLOCKX)')
GO
ALTER TABLE dbo.blarf
DROP CONSTRAINT FK__blarf__a__1E5A75C5
GO
DROP TABLE dbo.flarb
GO
EXECUTE sp_rename N'dbo.Tmp_flarb', N'flarb', 'OBJECT'
GO
ALTER TABLE dbo.flarb ADD CONSTRAINT
PK__flarb__3BD0198EF8B8CFAF PRIMARY KEY CLUSTERED
(
a
) WITH( STATISTICS_NORECOMPUTE = OFF, IGNORE_DUP_KEY = OFF,
ALLOW_ROW_LOCKS = ON, ALLOW_PAGE_LOCKS = ON) ON [PRIMARY]
GO
ALTER TABLE dbo.flarb ADD CONSTRAINT
CK__flarb__b__1A89E4E1 CHECK (([b]>(1)))
GO
CREATE TRIGGER dbo.flarb_tr
ON dbo.flarb
FOR UPDATE
AS
SELECT 1;
GO
COMMIT
BEGIN TRANSACTION
GO
ALTER TABLE dbo.blarf ADD CONSTRAINT
FK__blarf__a__1E5A75C5 FOREIGN KEY
(
a
) REFERENCES dbo.flarb
(
a
) ON UPDATE NO ACTION
ON DELETE NO ACTION
GO
ALTER TABLE dbo.blarf SET (LOCK_ESCALATION = TABLE)
GO
COMMIT
Yeah, that really looks like something I want to run in production, even if the table is small.
The other alternative is to do this quite manually. Say you have this table:
CREATE TABLE dbo.Employee
(
EmployeeID INT,
Name NVARCHAR(255),
Salary INT
);
INSERT ...
Theoretically, and leaving out the complications caused by keys, constraints, triggers, and schemabound objects, you could "change" the order of the columns by - for each column to the left of the one you want to move - creating a new column, moving the data, dropping the old column, and renaming the new column. This works because the new column is always added to the "end" of the table. Leaving out error handling and any of the other complications:
SET TRANSACTION ISOLATION LEVEL SERIALIZABLE;
BEGIN TRANSACTION;
ALTER TABLE dbo.Employee ADD Nametmp NVARCHAR(255);
UPDATE dbo.Employee SET Nametmp = Name;
ALTER TABLE dbo.Employee DROP COLUMN Name;
EXEC sp_rename N'dbo.Employee.Nametmp', N'Name', N'COLUMN';
COMMIT TRANSACTION;
This of course has the potential to be as disruptive (or worse) as the Management Studio approach of just swapping out the entire table, and gets really complicated if you want to move more than one column.
Now, do you really need to change the column order?
Best Answer
It is too long for a comment.
Well, that's the problem, then. I take data from
A, then apply overrides fromBto it and end up with a cycle. I think that there is a cycle, you don't think that there should be a cycle.So, there is a bigger problem with understanding how overrides should work.
My logic is the following.
Data from
Acan be represented in the same format asB:So, for
Jonwe have this chain originally:OR this set of rules:
Then we have overrides from
B:OR this set of rules:
Now we merge two sets of rules where second set overrides the first set if there is a conflict.
Rule (d) is the same as (c), so (c) is removed. Rule (e) overrides (a), so (a) is removed. Rule (f) is a new rule that doesn't exist in the first set. If any rules from the first set are still there, they are kept.
Final result of rules:
This is a directional graph. If it represents a tree, it can be ordered. It there are cycles, you need to decide what to do with them. If there are several disconnected trees, you can order each tree individually and then decide the order between trees.
In your example the graph looks like this (if we start from Coke).