Rails fixtures seem to be adding extra unexpected data

Question

Hello, all. I've got a dynamic fixture CSV file that's generating predictable data for a table in order for my unit tests to do their thing. It's working as expected and filling the table with the data, but when I check the table after the tests run, I'm seeing a number of additional rows of "blank" data (all zeros, etc). Those aren't being created by the fixture, and the unit tests are read-only, just doing selects, so I can't blame the code. There doesn't seem to be any logging done during the fixtures setup, so I can't see when the "blank" data is being inserted. Anyone ever run across this before, or have any ideas of how to log or otherwise see what the fixture setup is doing in order to trace down the source of the blank data?

Answer 1

You could turn on ActiveRecord logging (put ActiveRecord::Base.logger = Logger.new(STDOUT) in your test/test_helper.rb file).

Or, instead of using fixtures (which have gone the way of the dodo for most Rails developers) you could use something more reliable like Factories (thoughtbot's factory_girl) or seed_fu (if you have specific data that must be loaded).

Answer 2

I discovered what the problem was, although not the precise way to prevent it. I was placing ERB into the fixture CSV file, which was working fine, but due to the way it was being parsed and processed, it was causing blank lines to be placed into the resulting CSV output. Fixtures doesn't seem to handle that very well, and as a result it was inserting blank rows into the table. I couldn't prevent the blank lines from being placed into the output CSV because for whatever reason, the <% rubycode -%> doesn't work -- having the closing dash caused ERB parsing errors. Not sure why.

In any case, the eventual workaround was to switch to YML instead of CSV. It tolerates the white space just fine, and no blank rows are being inserted into the table anymore.

As an aside, factory___girl seems potentially interesting, but the YML is now doing just fine so it may be overkill. There's not a huge benefit to using seed_fu I think. In this current case I'm testing reporting code so the data is very specific and needs to be structured in a certain way in order to verify output data for the reports.