Back when we added a facility to add new labels to an enum, some technical considerations made us adopt a restriction that this must not take place within a transaction.block. David Wheeler has been complaining about this lately, including cornering some of us in the pub last night. His chagrin is understandable, since all the other changes in his migration scripts.are transactional and can be rolled back. Unfortunately this isn't something that is easy to overcome. What we can do, and should for 9.3, is to make the label addition command more resilient by adding an "if not exists" option. That way if the label has already been added to the type the command won't generate an error.
↧