Unit testing a select query

Question

I am returning a set of rows, each representing a desktop machine.

I am stumped on finding a way to unit test this. There's not really any edge cases or criteria I can think of, to test. It's not like share prices where I might want to check I am getting data which is indeed 5 months old. It's not like storing person details where you could check that a certain length always works, or special characters, etc. Or currency and different currencies (£, $, etc) as strings.

How would I test this sort of resultset?

Also, in testing the returnset of a query, there are a few problems:

1) Testing you have the same number of rows as when you run the query on the server is brittle because someone might change the table data. Is this when you have a test server, which nobody changes unless they upload change scripts?

2) Do you test the dataset object is not null? So if it's instantiated as null, but is not after the query's executed, it's holding value (this doesn't prove the data is correct, just that data has been retrieved).

Thanks

Answer 1

You can use a component like NBuilder that will simulate your database. And as you can manage all the aspects of your dataset you can test several aspects of the database interaction: the number of records your query returns, the range of values in some field. And, because the dataset is always created with the arguments you choose the data are always the same so you can reproduce your tests completly decoupled from your database.

Answer 2

1 -

a) Never ever test against the production server, for any reason.

b) Tests should start from a known configuration, which you can achieve either with mock objects or with a test database (some might argue that unit tests should use mock objects and integration tests should use a test database).

As for test cases, start with a basic functionality test - put a "normal" row in and make sure you get it back. You'll appreciate having that test if you later refactor. Make sure the program responds correctly to columns being null, or blank. Put the maximum and minimum values in all the DB fields and make sure the object fields you're storing them in can fit that resolution. Check duplicate records in the DB, or missing records. If you have production data, grab a snapshot of it to put in your test DB and make sure that loads correctly. Is there a value that chronically causes difficulties in other parts of the program? Check it here too. And once you release the code, add to the test list any values you find in production that break the system (Regression testing).