Question

Initially I have a column (partner_email) of type varchar. Now a recent change has come where it needs to be changed to be changed to the XML datatype but the existing rows needs to be converted into the new column.

Help needed

Answer 1

Stepping through the scripts will help understand the problem:

Script 1

1. Use the Test database
2. If the column partner_email_temp does not exist in the Partner
   table, add partner_email_temp as an XML column
3. Update the Partner table. Change the value of partner_email_temp to
   be an XML-friendly string version of partner_email. Then convert it to XML. Use REPLACE to convert all & to the encoded &amp
4. Drop the column partner_email from Partner
5. Rename the column partner_email_temp to partner_email

All of this works the first time. When you run the script a second time, SQL Server gets to Step 2 and executes the ALTER TABLE statement again. This is because the column partner_email_temp does not exist (it was renamed in the first run to partner_email).

Now in step 3, you are attempting to update the partner_email column again. The problem is that this column is now an XML data type and your use of REPLACE is illegal. So SQL Server throws an error.

Script 2

1. Capture the column data type for the partner_email column in
   Partner.
2. If the column data type is varchar (meaning this is the first pass
   of the script), execute the code in the BEGIN and END
3. If the column partner_email_temp does not exist, add the
   partner_email_temp column as an XML
4. Update the Partner table by changing the value of partner_email_temp
   to be an XML-friendly string version of partner_email and then
   convert it to XML. Use REPLACE to convert all & to the encoded &amp
5. Drop the column partner_email from Partner
6. Rename the column partner_email_temp to partner_email

The problem in the second script is that you cannot use the GO statement in your BEGIN and END statement.

According to BOL:

SQL Server utilities interpret GO as a signal that they should send
the current batch of Transact-SQL statements to an instance of SQL
Server.

So in script 2, when SQL Server attempts to compile, it notices that the column partner_email_temp doesn’t exist. It works in the Script 1 because the sections are executed separately. And when the portion that references partner_email_temp is executed, the column does exist.

So how do you fix it?

You can actually bypass all the dropping and renaming of columns if you simply update the data in the partner_email column to the correct format and then convert the column to an XML data type. This can be achieved in one SQL batch statement, but you need to execute it as dynamic SQL so that SQL Server will only try to execute the REPLACE statement when the partner_email column is a varchar data type:

IF NOT EXISTS 
(
    SELECT * 
    FROM INFORMATION_SCHEMA.columns 
    WHERE table_name = 'Partner'
    AND   column_name = 'partner_email'
    AND   data_type = 'xml'
)
BEGIN

  DECLARE @sqlCmd NVARCHAR(4000)

  SET @sqlCmd = N' 
  UPDATE [dbo].[Partner]
  SET partner_email = ''<PartnerEmails><Email>'' 
  + REPLACE(partner_email, ''&'', ''&amp;'') 
  + ''</Email><Email></Email><Email></Email></PartnerEmails>'';
  ALTER TABLE Partner ALTER Column partner_email XML
  '

  EXEC (@sqlCmd) 

END