Weird failing spec

Testing RSpec
1

Hi guys,

I have a weird failing spec, for which I just cannot figure out the
reason of failure. I’m now rewriting my controller specs based on the
advice of David and Ashley, and I got stuck on this (see: [rspec-
users] specing rescue, ensure and else blocks of an Exception).

http://pastie.caboo.se/111221

I’ve tried everything, like stubbing out :update_attributes! , even
tested it in real-life and it works like a charm: updates / creates
new records nicely and everything else is working properly.

Any ideas would be greatly appreciated, have a nice day and thanks,
András

2

On Oct 26, 2007, at 5:39 pm, Tarsoly András wrote:

Parked at Loopia

I’ve tried everything, like stubbing out :update_attributes! , even
tested it in real-life and it works like a charm: updates / creates
new records nicely and everything else is working properly.

First, like David said earlier, you’re putting too much into one spec:

it “should retrieve the Option instance identified by ID” do
CalciumOption.should_receive(:find).with(“1”).and_return(@option)
@option.should_receive(:update_attributes!).with
(invalid_option_attributes).and_raise(ActiveRecord::RecordInvalid)
post :update, :id => 1, :option => invalid_option_attributes
end

should really be

it “should retrieve the Option instance identified by ID” do
CalciumOption.should_receive(:find).with(“1”).and_return(@option)
post :update, :id => 1, :option => invalid_option_attributes
end

it “should update the Option instance’s attributes” do
@option.should_receive(:update_attributes!).with
(invalid_option_attributes).and_raise(ActiveRecord::RecordInvalid)
post :update, :id => 1, :option => invalid_option_attributes
end

Also I don’t think you need '@option = ’ in front of
@option.update_attributes!, as that method call modifies the object
in-place. But that’s a Rails issue.

One other thing - you aren’t stubbing CalcumOption.find in
‘Admin::OptionsController, “#update (POST) with valid params”’. Are
you sure this isn’t loading real models?

As for the actual problem you asked about… sorry but I am lost.
I’ve looked over it about 5 times and everything I thought was wrong
was me not paying attention. I can’t see ANY reason it should fail.
Is it still not working?

Ashley

–
blog @ http://aviewfromafar.net/
linked-in @ http://www.linkedin.com/in/ashleymoran
currently @ home

3

On 2007.10.26., at 21:28, Ashley M. wrote:

First, like David said earlier, you’re putting too much into one spec:

Also I don’t think you need '@option = ’ in front of
@option.update_attributes!, as that method call modifies the object
in-place. But that’s a Rails issue.

One other thing - you aren’t stubbing CalcumOption.find in
‘Admin::OptionsController, “#update (POST) with valid params”’. Are
you sure this isn’t loading real models?

Yes, I was aware of this as I was just messing around, trying to
figure out where could this failed specs could come from, I pastied
the real stuff with a rake spec --trace output here:

http://pastie.caboo.se/111379

As for the actual problem you asked about… sorry but I am lost.
I’ve looked over it about 5 times and everything I thought was wrong
was me not paying attention. I can’t see ANY reason it should fail.
Is it still not working?

In the link above, you can see, that the specs still fail. Looks like
it has to do something with exception handling as only those examples
fail which should raise errors (eg. doing :create!
or :update_attributes! with invalid and/or missing attributes and
specing :and_raise())

The test.log sample in the pastie shows unfinished requests in all
the :update! specs, where the :update_attributes! is stubbed on the
mock.

However, one example at #create raises an Exception properly, which
can be seen in the log too, only where I’m using :and_raise().

I believe I stubbed everything out properly in the above example. I’m
using edge rspec with the latest rspec_on_rails and using autotest
for convenient coding.

The problem is, I’m rather new to rspec / ruby / rails, so I can’t
just start to track this down, however I went into the rspec source
already, but got lost pretty much :slight_smile:

Thanks,
András

Tarsoly
Andrá[email protected]

4

On 2007.10.27., at 7:18, David C. wrote:

and_raise(ActiveRecord::RecordInvalid.new(@option))

Sorry if I led you astray on that one.

No problems, actually it makes perfect sense, since you want @option
to have all the proper error messages and such.

After adjusting the specs like this one, it was all working fine and
dandy as soon as I stubbed the .errors.full_messages() call.

This raises up a question in me about mocking or using real-life
models, because in some of my specs the setups getting extremely
bulky. But this is for another topic :slight_smile:

Cheers,
András

Tarsoly
Andrá[email protected]

5

On Oct 27, 2007, at 9:45 am, Tarsoly András wrote:

This raises up a question in me about mocking or using real-life
models, because in some of my specs the setups getting extremely
bulky. But this is for another topic :slight_smile:

Well while you have everyone’s attention :slight_smile:

Often you can factor out the setup code - if you search through the
archives of this list there’s plenty of examples. But don’t worry
about having long setups - I’ve written some setups that aren’t much
shorter than the specs that follow. Write whatever setup you need to
enable you to write solid specs, then worry after about making it
look nice.

I would say you should ALWAYS mock your models in your controllers
specs. Otherwise you will further complicate your setups, make your
specs run unacceptably slow, and you risk coupling your controllers
and models too tightly. Many people in the BDD community (and at a
guess I am assuming David is among them) would frown on using
database access in your model code, because persistence is a
separate issue to business logic.

What you want is a set of specs for each of model (ideally, database
persistence and business logic separately), controller, and view, and
then an integration test, such as an RSpec story, that tests the full
stack. Without this integration step, you will miss errors when you
change the interface of a model. What you do not want is the Rails
way of testing, where each layer of testing re-tests everything below
it. (I recently saw a blog post where someone asked for comments on
how people test views in isolation - it never occurred to me that
some Rails developers still see that as an issue!)

Having thought about it, you could argue Rail has no unit testing
support built in at all.

Ashley

–
blog @ http://aviewfromafar.net/
linked-in @ http://www.linkedin.com/in/ashleymoran
currently @ home

6

On 10/27/07, Ashley M. [email protected] wrote:

On Oct 27, 2007, at 9:45 am, Tarsoly András wrote:

This raises up a question in me about mocking or using real-life
models, because in some of my specs the setups getting extremely
bulky. But this is for another topic :slight_smile:

Bulky setup is a design smell. That doesn’t mean it never happens, but
when it does you should ask yourself what you can do to improve the
design. If you’re not familiar with the Law of Demeter, search around
for it and give it some thought.

Well while you have everyone’s attention :slight_smile:

Often you can factor out the setup code - if you search through the
archives of this list there’s plenty of examples. But don’t worry
about having long setups - I’ve written some setups that aren’t much
shorter than the specs that follow. Write whatever setup you need to
enable you to write solid specs, then worry after about making it
look nice.

In spite of what I said earlier, this is true. For specs, coming from
TDD, “the way” is to get from red to green and then refactor. The
refactoring step should be considered every time you get to green.
Where is duplication? How can I improve the design to eliminate it?
You won’t always actually do anything, but the thought process is very
important. It’s like a 24/7 design review.

I would say you should ALWAYS mock your models in your controllers
specs. Otherwise you will further complicate your setups, make your
specs run unacceptably slow, and you risk coupling your controllers
and models too tightly. Many people in the BDD community (and at a
guess I am assuming David is among them) would frown on using
database access in your model code, because persistence is a
separate issue to business logic.

If I were writing the framework, this would be true. But Rails
provides efficiencies in exchange for blending these concepts in to
one. It’s a trade off, but I’ve come to appreciate the efficiencies
enough where I live with it.

What you want is a set of specs for each of model (ideally, database
persistence and business logic separately

Do you mean like this?

spec/models-persistence
spec/models-business_logic

Or do you just mean separating the concerns within a single spec
document?

support built in at all.
Absolutely agree. You have to go some distance to do any real unit
testing in Rails. But, on the flip side, you get so much for free that
you get from other more highly-decoupled framework that it might be
worth the trade-off. There’s a balance in there somewhere … let me
know when you find it :slight_smile:

Cheers,
David

7

On Oct 27, 2007, at 3:36 pm, David C. wrote:

Bulky setup is a design smell. That doesn’t mean it never happens, but
when it does you should ask yourself what you can do to improve the
design. If you’re not familiar with the Law of Demeter, search around
for it and give it some thought.

I don’t think I have any LoD violations in my long setups. It’s just
the sheer number of things that some pages need to load. It’s
actually a while since I looked over the code I am referring to. It
may actually just be view specs I am thinking of, where bulky setups
are just a sign of the number of different things being displayed.
But either way, I’ll be vigilant when I return to them.

In spite of what I said earlier, this is true. For specs, coming from
TDD, “the way” is to get from red to green and then refactor. The
refactoring step should be considered every time you get to green.
Where is duplication? How can I improve the design to eliminate it?
You won’t always actually do anything, but the thought process is very
important. It’s like a 24/7 design review.

I just watched this today: <Presentations > Page #1
applying-agile-to-ruby>

Nothing revolutionary if you’ve been doing BDD in Ruby a while, but
one of the things he drives home is the idea that in Ruby (and
dynamic languages in general), you must have the shortest possible
cycles, and must always work to make sure quality of the code is
maintained as high as you can manage.

It’s puzzling looking at developers who don’t do BDD, who think that
refactoring is something you do when adding a new feature would cause
the collapse of half the application. I don’t know how they live
with (a) the stress of leaving a mess behind them and (b) the
impending nervous breakdown when they actually have to fix it (with
nothing to tell them it still works!) I dread to look at some of the
code that is developed with none of the principles you describe
above. Hell, mine is bad enough sometimes and I make the effort.

Many people in the BDD community (and at a
guess I am assuming David is among them) would frown on using
database access in your model code, because persistence is a
separate issue to business logic.

If I were writing the framework, this would be true. But Rails
provides efficiencies in exchange for blending these concepts in to
one. It’s a trade off, but I’ve come to appreciate the efficiencies
enough where I live with it.

You’ve become unusually forgiving lately :slight_smile:

What you want is a set of specs for each of model (ideally, database
persistence and business logic separately

Do you mean like this?

spec/models-persistence
spec/models-business_logic

Or do you just mean separating the concerns within a single spec
document?

I’d never thought about splitting them into directories like that.
Currently I just make sure that associations have a separate spec
block to the others, something like

describe MyModel do

…

end

describe MyModel, “associations” do

…

end

There’s probably not so much to do with ActiveRecord, because your
models HAVE to map directly to tables. But I still test separately
the persistence of any custom methods that build object graphs.

Absolutely agree. You have to go some distance to do any real unit
testing in Rails. But, on the flip side, you get so much for free that
you get from other more highly-decoupled framework that it might be
worth the trade-off. There’s a balance in there somewhere … let me
know when you find it :slight_smile:

I still believe that Ruby is powerful enough that we can build highly-
decoupled frameworks that collapse down to the simple cases by
default. Something that has Rails’ beauty on the inside as well as
the outside. I’ll let you know when I find that!

Ashley

–
blog @ http://aviewfromafar.net/
linked-in @ http://www.linkedin.com/in/ashleymoran
currently @ home

8

On 10/26/07, Tarsoly András [email protected] wrote:

One other thing - you aren’t stubbing CalcumOption.find in
As for the actual problem you asked about… sorry but I am lost.
The test.log sample in the pastie shows unfinished requests in all
The problem is, I’m rather new to rspec / ruby / rails, so I can’t
just start to track this down, however I went into the rspec source
already, but got lost pretty much :slight_smile:

wrong number of arguments (0 for 1)

That’s coming from ActiveRecord::RecordInvalid, which takes an
argument in its constructor. I think I left this out in my example to
you earlier. If you pass raise_error a class, it instantiates in order
to throw an instance of it. To solve your problem, you just need to
create an instance and pass it the mock_model:

and_raise(ActiveRecord::RecordInvalid.new(@option))

Sorry if I led you astray on that one.

Cheers,
David