Strange Little Problem

I just created a simple model and started to write unit tests for it
(I’ve cleaned out extraneous stuff).

The migration is:

class CreateMembers < ActiveRecord::Migration
def self.up
create_table :members do |t|
t.column :name, :string
t.column :number, :string
end
end

def self.down
drop_table :members
end
end

The model is:

class Member < ActiveRecord::Base
validates_uniqueness_of :number
end

The testing fixture is:

numeric_only_member:
id: 1
name: John S.
number: 15171914
alpha_numeric_member:
id: 2
name: Joanne Jones
number: wke1234

So if I write a little test case (yes I know hard coded values are
fragile):

require File.dirname(FILE) + ‘/…/test_helper’
require ‘member’

class ExampleTest < Test::Unit::TestCase
fixtures :members

def test_uniqueness_of_member
dup_member = Member.new(:name => ‘duplicate’, :number =>
‘15171914’)

assert !dup_member.save
assert_equal 1, dup_member.errors.count
assert_equal "has already been taken",

dup_member.errors.on(:number)
end
end

Problem:

“assert_equal 1, dup_member.errors.count” will fail with:
test_uniqueness_of_member(SimpleTest) [test/unit/simple_test.rb:11]:
<1> expected but was <2>.

The following line would also fail with:
test_uniqueness_of_member(SimpleTest) [test/unit/simple_test.rb:12]:
<“has already been taken”> expected but was
<[“has already been taken”, “has already been taken”]>.

Solution:

If I remove " require ‘member’ " from the test case then everything
passes. I have no idea why. (Yes I’m a newbie, please don’t laugh and
point). I’m digging through ActiveRecord, UnitTest, and Ruby docs and
source but haven’t found an answer yet. Any ideas?

Thanks…qb

On Sep 25, 2007, at 10:55 PM, qb wrote:

end

validates_uniqueness_of :number
end

The testing fixture is:

numeric_only_member:
id: 1
name: John S.
number: 15171914

You might need:
number: “15171914”
So that YAML keeps this a string and not an integer. You’re not
using members(:numeric_only_member).number so it might not matter
(the database should be a string regardless).

alpha_numeric_member:
id: 2
name: Joanne Jones
number: wke1234

So if I write a little test case (yes I know hard coded values are
fragile):

require File.dirname(FILE) + ‘/…/test_helper’
require ‘member’

You shouldn’t need this as Rails will auto-load it when Member is
first used.

assert_equal "has already been taken",

The following line would also fail with:

Thanks…qb

I think that Member has already been defined when your ‘require
“member”’ is executed and since the idempotency of ‘require’ is based
on the argument (i.e., require ‘foo’ and require ‘./foo’ will load
foo.rb from the current directory twice), the explicit require re-
opens the Member class and hits the validates_uniqueness_of a second
time. In addition to removing your “require ‘member’”, I’d also
suggest writing your test like this:

class ExampleTest < Test::Unit::TestCase
fixtures :members

def test_uniqueness_of_member
dup_member = Member.new(:name => ‘duplicate’,
:number => members
(:numeric_only_member).number)

 assert ! dup_member.valid?
 assert dup_member.errors[:number]
 assert_match /has already been taken/, dup_member.errors.on

(:number)
end
end

You want to know that the record isn’t valid, there’s no need to
actually try to save it.
You should check that the field has an error message, rather than the
total error count.
I tend to test for things like error message text with regexps that
will allow some flexibility in the actual value. (e.g., “has already
been taken. Please choose another and try again.”)

However, all you’re doing is testing that the validates_uniqueness_of
works. Don’t you think that the Rails team already does that? You
could make a case for verifying that something like a custom format
is behaving correctly or that you’re using the :scope option as you
expect, but unless you’re just practicing and convincing yourself how
things work, I’d not normally test the things that Rails does.

-Rob

Rob B. http://agileconsultingllc.com
[email protected]

On 9/25/07, qb [email protected] wrote:

If I remove " require ‘member’ " from the test case then everything
passes. I have no idea why. (Yes I’m a newbie, please don’t laugh and
point). I’m digging through ActiveRecord, UnitTest, and Ruby docs and
source but haven’t found an answer yet. Any ideas?

As Rob pointed out, it’s being caused by your model file being loaded
twice, and thus the validates_uniqueness_of being run twice. You end
up getting two (identical) error messages.

The “second” load of your model file is actually coming from the
“fixtures” line, which ends up calling:

require_dependency ‘member’

In order for require_dependency to do it’s magic, it has to turn your
file name into an absolute path. So you get the effect of:

require ‘member’
require ‘/some/absolute/path/to/member.rb’

Ruby doesn’t know that these are the same file, so it ends up loading
the file twice, thus doubling the validates declarations.

So, the short answer is: don’t “require” your models in your test
cases (or use require_dependency).

On 9/26/07, Rob B. [email protected] wrote:

However, all you’re doing is testing that the validates_uniqueness_of
works.

Rob, is that all he’s doing? The test is too elaborate, but a case
could be made that this is a regression test for the business rule
that “Member numbers must be unique”.

It would protect him from coming along later and inadvertently
removing that validation rule.

I agree that we shouldn’t be testing Rails itself.

On Sep 26, 2007, at 2:44 PM, Bob S. wrote:

I agree that we shouldn’t be testing Rails itself.

Stu Halloway addressed this at erubycon in Columbus,OH this summer.

Blog: http://relevancellc.com/2007/7/17/posted-slides-how-not-to-test-
validations
Slides: http://relevancellc.com/assets/2007/7/17/KeepingTestsDry.pdf

If you’re interested in a test to alert you when/if a validation is
removed accidentally, then you should be devising a test for that
(presence) and letting Rails make sure that validates_*_of does the
right thing.

-Rob

Rob B. http://agileconsultingllc.com
[email protected]

On 9/26/07, Rob B. [email protected] wrote:

It would protect him from coming along later and inadvertently
If you’re interested in a test to alert you when/if a validation is
removed accidentally, then you should be devising a test for that
(presence) and letting Rails make sure that validates_*_of does the
right thing.

Well, that sounds good in theory. But in those slides, slide 18 looks
great but the stuff on 19/20 is a mess!

Devising a test for the presence of validates_uniqueness_of (instead
of exercising it) is non-trivial. And I’m not sure that’s somehow
“better” or “DRY’er” than just trying to create a duplicate and seeing
it fail.

I’m really not trying to test whether I have a validates_uniqueness_of
declaration, and I’m not trying to test whether
validates_uniqueness_of works in the abstract (the test doesn’t prove
that anyway). Instead, I’m trying to test that my model doesn’t allow
the number to be duplicated (regardless of the details of the
implementation).

IMHO, something like slide 17 is the correct approach, and not 18.

Cheers,
Bob