validate_uniqueness_of matcher incorrectly calls #next on UUIDs, causing spec failures

[alisdair]

I have a Rails 4/Postgres app using UUID primary keys, with a uniqueness validation on a column which is scoped to a UUID foreign key. My specs fail occasionally because shoulda-matchers is trying to insert an invalid UUID into the database.

This happens when a UUID used to test the scope of the validation happens to end in "f". The validate_after_scope_change? method calls #next on the UUID, which makes it end in "g", which is obviously invalid hex and an invalid UUID.

See this code in ValidateUniquenessOfMatcher:

previous_value = all_records.map(&scope).max

# Assume the scope is a foreign key if the field is nil
previous_value ||= correct_type_for_column(@subject.class.columns_hash[scope.to_s])

next_value =
  if previous_value.respond_to?(:next)
    previous_value.next
  elsif previous_value.respond_to?(:to_datetime)
    previous_value.to_datetime.next
  else
    previous_value.to_s.next
  end

@subject.send("#{scope}=", next_value)

The fix for the bug should be relatively straightforward: when generating the next value, first check if the column type is UUID; if so, generate a new UUID instead of calling #next.

However, I cannot put together a failing test, because none of the existing tests are actually using UUIDs. The test suite uses sqlite, which does not have a ":uuid" column type. This breaks the correct_type_for_column method:

def correct_type_for_column(column)
  if column.type == :string
    '0'
  elsif column.type == :datetime
    DateTime.now
  elsif column.type == :uuid
    SecureRandom.uuid
  else
    0
  end
end

The column type for the :uuid scopes is nil, so the case statement falls through and the UUID-related tests are actually using integers for the scope values. This means that when I try to validate the UUID for these tests, it always fails.

I can put together a pull request that fixes the bug, but I don't know how to go about adding a test for the problem. What should I do?

/cc @yabawock, who implemented the existing UUID support (thanks!)

[mcmire]

Hmm. Good call on this. Let me think about it and get back to you on this.

[mcmire]

@alisdair Hey... sorry for the delay in getting back to you.

So... okay. I guess right now you can't really add a test for it, and in order to test this stuff properly, we need to set up alternate databases and run the tests against them, similar to how ActiveRecord tests are run against multiple databases.

If you would like to tackle this, that would be awesome... otherwise I would say try to fix this locally for your app and we'll get to setting this up at some point in the future (I can't say when as we have quite a backlog right now).

[kidlab]

I had the same issue.

[smber1]

Me too. When can we expect an official fix for this?

[mcmire]

I don't have a timetable for this, but I am trying to make it a priority in the next release to address some of these older issues.