Tuesday, August 28, 2007

My take on Enums in SQL Server

I am a registered user on SQL Server Central, and I read a good article there on "Enums in SQL Server". I thought that I would quickly present my own humble solution to this problem, although I do not mean to imply that my way is better - you can decide for yourself.

First, the problem to be solved is to represent a limited set of choices, where new choices would typically require a developer's intervention to implement the business logic. An example would be a list of task priorities in a to-do list.

CREATE TABLE Priorities(
ID tinyint not null identity(1,1) PRIMARY KEY,
Code char(3) not null unique,
Priority varchar(20) not null
) ;

INSERT INTO Priorities (Code, Priority) VALUES ('911', 'Emergency') ;
INSERT INTO Priorities (Code, Priority) VALUES ('HIG', 'High') ;
INSERT INTO Priorities (Code, Priority) VALUES ('NOR', 'Normal') ;
INSERT INTO Priorities (Code, Priority) VALUES ('LOW', 'Low') ;
INSERT INTO Priorities (Code, Priority) VALUES ('ATA', 'As Time Allows') ;

The best feature of this lookup table style is that it allows you to choose a key size that is appropriate for the number of data elements, and tiny int is 1 byte. I feel confident that an integer key comparisons will always be as fast as or faster than a char(x) string, since we never have to worry about collation sequences. We do not use the description field or the primary key in our business logic, we use the "Code" field exclusively. This allows us to change the descriptions based on user dictates (and it will happen!), minimize storage requirements in our task table where it matters, and if we ever need to archive the data, just convert Priorities.ID to Priorities.Code. It also means that our business logic doesn't care what data type is used for the primary key, minimizing code disturbances if/when we move from tinyint to smallint.

Having presented my take on the problem, the approach taken in the Enum article has its advantages. Archiving is dead simple - just copy the lookup key. And there is no need to force the code list to be unique since it is a primary key and this property is guaranteed. Also, debugging is simpler, as the developers will not usually have to translate the codes to determine their meaning.

Either way, this avoids a classic problem in programming - mixing usages. I have always felt that we should not let a user assign primary keys to tables, as they are not likely to make unique values, and they tend to want to change their mind, which makes for some miserable modified key update code that ripples through your database, touching related tables that otherwise would not care that (for example) a user has changed their name.



Post a Comment

Subscribe to Post Comments [Atom]

<< Home