sentence sounds weird on site...

I read this…

“Other refactorings such as �drop table� and �insert data� have no corresponding rollback commands that can be automatically generated.”

I could understand not being able to rollback from drop table or even delete from table where xxx returns way too many rows, BUT insert data???  If I have some insert commands, the delete commands should be generated in the oppossite order where pk=x.  How come that can’t be generated?

Is this a typo or does insert data have no rollback commands?

The documentation is correct.  The reason is because from the changelog file we cannot know which column is the primary key, and there are many times where you are not even inserting into the primary key column (like when it is an autoincrement column) and so we do not know the primary key to delete.  We cannot delete off the columns that were passed in because they may not be unique. 

You can easily add a tag for your insert statements based on what you know of your data and what has been inserted.