I’m working on a project where I need to convert Firebird SQL code into T-SQL code. No schema, just the modules. There are more than 1000 objects between stored procedures, views, triggers, user-defined data types, etc.

First - the pain...

While checking the Firebird reference manuals I saw a lot of different concepts (Selectable Stored Procedures - Yes you can do SELECT FROM StoredProcedure) and different functions names with different syntax compared to T-SQL.

With this in mind, can you imagine doing a code migration between two different SQL flavours that contains more than 60000 lines of code by hand? 😨

and...the false alarm - My approach

When I started to think about the possible approaches to this huge amount of work I will have, one of the ideas (and actually the one I ended up doing) was write a find/replace function in PowerShell that can receive an hashtable with “pattern to search” and the “pattern to replace”.

Because the -replace method allows the use of regular expressions this become very powerful.

Don't reinvent the wheel

I did a search and found a gist from Matthew Steeples that has a function to do a find/replace of a single string on a file using regular expressions.

I picked this script and I have adapted to my reality. This means, I have changed the code in order to:

  • Accept an ordered hashtable and this way do multiple changes.
  • Open the file and do all the changes before save it again.

You can find on my GitHub repository the Set-Expression PowerShell function I ended with.

Filling the hastable

To fill the hashtable I had to make a match between the two SQL flavours.

Just a couple of examples to ilustrate the differences:

Firebird has the similar function which translate to LIKE '%%' on T-SQL Other example is the SUBSTRING function which has the following structure SUBSTRING([field] from [start_position] for [number of chars]) which, in order to translate for T-SQL, we just need to replace the “from” and “for” words by a comma (“,”).

Everything was going well, work was progressing at a good pace until...

The pain strikes back

These databases make a heavy use of triggers. The main use of it is to generate a new ID to a column based on a Firebird DOMAIN object (SEQUENCES in T-SQL) on BEFORE INSERT (INSTEAD OF in T-SQL) triggers.

Note: “But they can use IDENTITY columns…” - Exactly what I have proposed, and for some tables they have implemented but for the majority it wasn’t possible.

Firebird make this somehow easy. This is a code example from an Firebird trigger:

  if (new.id is null) then
    new.id = gen_id(gen_Customers_id,1);

in T-SQL syntax, we need to set the whole INSERT statement with all columns from the INSERTED table. The equivalent code can be something like this:

ON dbo.Customers
          IF EXISTS
                  FROM INSERTED
                 WHERE Inserted.ID IS NULL
                    INSERT INTO dbo.Customers (ID, <other columns>
                           NEXT VALUE FOR dbo.gen_Customers_id
                         , <other columns>
                    FROM INSERTED;

"so you need to replace 1 line with an `INSERT` statement with all columns from the table..."

Yes that much work..on almost 300 triggers! Feeling the pain, right?

Here is where the SQL Server metadata joins the party!

This is just an example, but with it you can imagine the others :-)

I wrote a piece of T-SQL (not pretty but it works) that creates, for each table, the hashtable entry for my find/replace function.

    , '''(FOR )(\b' + A.TABLE_NAME
      + '\b)((.|\r\n){1,})(\bACTIVE BEFORE INSERT POSITION 0\b)(\r\n)(AS)(\r\n)(BEGIN)(\r\n)((.|\n){1,})(gen_id\()(\w{1,})((.|\r\n)*?)(?=end;)(.*)'' = "ON `$2`$3INSTEAD OF INSERT`$3`$7`$8`$9`$10`$11`$12 `$3INSERT INTO `$2 ( '
      + A.ColumnList + ' ) `$3SELECT NEXT VALUE FOR dbo.`$14, ' + A.ColumnList + ' FROM INSERTED `$3 --`$13`$14`$15`$16`$17"'
    , '''(FOR )(\b' + A.TABLE_NAME
      + '\b)((.|\r\n){1,})(\bACTIVE BEFORE UPDATE POSITION 0\b)(\r\n)(AS)(\r\n)(BEGIN)(\r\n)((.|\r\n){1,})((.|\r\n)*?)(?=end;)(.*)'' = "ON `$2`$3INSTEAD OF UPDATE`$3`$7`$8`$9 `$3UPDATE `$2 `$3SET '
      + A.ColumnListUpdate + ', `$10`$11 FROM INSERTED `$3 `$15"'
               , STUFF((
                                  ', ' + ISC1.COLUMN_NAME
                             FROM INFORMATION_SCHEMA.COLUMNS AS ISC1
                            ORDER BY ISC1.ORDINAL_POSITION
                           FOR XML PATH('')
                     , 1
                     , 1
                     , ''
                      ) AS ColumnList
               , STUFF((
                                  ', ' + ISC1.COLUMN_NAME + ' = ' + ISC1.COLUMN_NAME
                             FROM INFORMATION_SCHEMA.COLUMNS AS ISC1
                            ORDER BY ISC1.ORDINAL_POSITION
                           FOR XML PATH('')
                     , 1
                     , 1
                     , ''
                      ) AS ColumnListUpdate
                 INNER JOIN sys.sysobjects AS so
                    ON ISC.TABLE_NAME = so.name
                   AND so.type = 'U'
           WHERE ISC.TABLE_NAME NOT IN ('sysdiagrams', 'Numbers')
      ) AS A

Let’s say I have a Customers table with 3 columns ID, NAME, COMPANY_NAME the output will be:

Example of the full string for the INSTEAD OF INSERT trigger:

'(FOR )(\bCustomers\b)((.|\r\n){1,})(\bACTIVE BEFORE INSERT POSITION 0\b)(\r\n)(AS)(\r\n)(BEGIN)(\r\n)((.|\n){1,})(gen_id\()(\w{1,})((.|\r\n)*?)(?=end;)(.*)' = "ON `$2`$3INSTEAD OF INSERT`$3`$7`$8`$9`$10`$11`$12 `$3INSERT INTO `$2 (  ID, NAME, COMPANY_NAME ) `$3SELECT NEXT VALUE FOR dbo.`$14,  ID, NAME, COMPANY_NAME FROM INSERTED `$3 --`$13`$14`$15`$16`$17"`

This way I can just copy/paste the result to my hashtable and run the PowerShell function on my triggers’ files.

All's well that ends well

This made possible to automate, I would say, 95% of the work I have to do on each trigger object.

The 5% left are purposeful and intended to format the code using Redgate SQL Prompt (formatted the code using the great CTRL + K, Y shortcut) once I open the script on SQL Server Management Studio and test the object compilation.

Other examples?

I applied the same recipe on scripts with user-defined data types (UDDT). I picked the system data type and created entries to the hashtable with the proper replace.

       '''\b' + T2.name + '\b'' = "' + T1.name + CASE
                                                     WHEN T2.system_type_id IN (231, 239) THEN '(' + CASE
                                                                                                         WHEN T2.max_length = -1 THEN 'MAX'
                                                                                                         ELSE CAST(T2.max_length / 2 AS VARCHAR(10))
                                                                                                     END + ')'
                                                     WHEN T2.system_type_id IN (175, 167) THEN '(' + CASE
                                                                                                         WHEN T2.max_length = -1 THEN 'MAX'
                                                                                                         ELSE CAST(T2.max_length AS VARCHAR(10))
                                                                                                     END + ')'
                                                     ELSE ''
                                                 END + '"'
  FROM sys.types AS T1
       INNER JOIN sys.types AS T2
          ON T1.user_type_id = T2.system_type_id
 WHERE T2.user_type_id > 256;
Practical example:

UDDT named INT_VALUE, that represents an INT, used in the following way CAST(column AS INT_VALUE) needs to be replaced as CAST(column as INT).

Why we need to do this replace? I have written about it on my blog post Using CAST() function with User-Defined Data Types…Did you know… take a look.

Wrap up - Life saver and time saved

Leverage on SQL Server metadata to generate quick code and use it inside SQL Server or outside like I did it here!

I did some math and I have estimated that I would need several months to finish this task doing everything by hand.

Everything could go wrong doing that way:

  • This is an heavely repetitive task
  • highly prone to errors
  • a witch hunt when problems arise

After all, with this approach I have made in less than one month! And, half of those days were to tweak the regular expressions/hashtable.

Automation? You gotta to love it! :-)

Thanks for reading!