On 19 Jul 2010, at 11:38, Wincent Colaiuta wrote:
> El 19/07/2010, a las 10:58, Matt Wynne escribió:
>
>> On 18 Jul 2010, at 00:10, David Chelimsky wrote:
>>
>>> On Jul 17, 2010, at 1:18 PM, Costa Shapiro wrote:
>>>
>>>> Hello,
>>>>
>>>> I've been thinking of how to express my idea in code, but since I've never
>>>> been involved in RSpec development, I'd better have some feedback here
>>>> first.
>>>> The feature suggestion below applies to any controller-like code under
>>>> spec, i.e. a stateless module producing output or just altering its data
>>>> store (it doesn't necessarily have to be a C of the MVC, but I suppose
>>>> merb/rails developers will particularly appreciate it).
>>>>
>>>> Here is a skimmed sample to illustrate the pain:
>>>>
>>>> describe BookController do
>>>>
>>>> context "registering a book" do
>>>>
>>>> specify "from a new author on a new subject" do
>>>> auth = mock(:name => 'John Doe')
>>>> Author.should_receive(:find_
>>>> by_name).and_return(nil)
>>>> Author.should_receive(:new).and_return(auth)
>>>> auth.should_receive(:save).and_return(true)
>>>>
>>>> subj = mock(:short => 'Mockery')
>>>> Subject.should_receive(:find_by_short).and_return(nil)
>>>> Subject.should_receive(:new).and_return(subj)
>>>> subj.should_receive(:save).and_return(true)
>>>>
>>>> title = 'Specs on Steroids'
>>>>
>>>> book = mock
>>>> Book.should_receive(:new).and_return(book)
>>>> book.should_receive(:save).and_return(true)
>>>>
>>>> post :register :author => auth.name, :title => title, :subject =>
>>>> subj.short
>>>> response.should be_success
>>>> end
>>>>
>>>> specify "from a known author on a new subject" do
>>>> auth = mock(:name => 'Joe Dohn')
>>>> Author.should_receive(:find_by_name).and_return(auth)
>>>>
>>>> subj = mock(:short => 'Mockery')
>>>> Subject.should_receive(:find_by_short).and_return(nil)
>>>> Subject.should_receive(:new).and_return(subj)
>>>> subj.should_receive(:save).and_return(true)
>>>>
>>>> title = 'Specs on Steroids II'
>>>>
>>>> book = mock
>>>> Book.should_receive(:new).and_return(book)
>>>> book.should_receive(:save).and_return(true)
>>>>
>>>> post :register :author => auth.name, :title => title, :subject =>
>>>> subj.short
>>>> response.should be_success
>>>> end
>>>>
>>>> specify "from a known author on a known subject" do
>>>> auth = mock(:name => 'Joe Dohn')
>>>> Author.should_receive(:find_by_name).and_return(auth)
>>>>
>>>> subj = mock(:short => 'Forgery')
>>>> Subject.should_receive(:find_by_short).and_return(subj)
>>>>
>>>> #...
>>>> end
>>>>
>>>> specify "from a new author on a known subject" do
>>>> #...
>>>> end
>>>> end
>>>> end
>>>>
>>>>
>>>> And this is what I have in mind for doing exactly the same job:
>>>>
>>>> describe BookController do
>>>>
>>>> context "registering a book" do
>>>>
>>>> before :any, "from a new author", :author do
>>>> @auth = mock(:name => 'John Doe')
>>>> Author.should_receive(:find_by_name).and_return(nil)
>>>> Author.should_receive(:new).and_return(@auth)
>>>> @auth.should_receive(:save).and_return(true)
>>>> end
>>>>
>>>> before :any, "from a known author", :author do
>>>> @auth = mock(:name => 'Joe Dohn')
>>>> Author.should_receive(:find_by_name).and_return(@auth)
>>>> end
>>>>
>>>> before :any, "on a new subject", :subject do
>>>> @subj = mock(:short => 'Mockery')
>>>> Subject.should_receive(:find_by_short).and_return(nil)
>>>> Subject.should_receive(:new).and_return(@subj)
>>>> @subj.should_receive(:save).and_return(true)
>>>> end
>>>>
>>>> before :any, "on a known subject", :subject do
>>>> @subj = mock(:name => 'Joe Dohn')
>>>> Subject.should_receive(:find_by_name).and_return(@subj)
>>>> end
>>>>
>>>> it "should succeed", :with => [:author, :subject] do
>>>> title = 'Specs on Steroids X'
>>>>
>>>> post :register :author => @auth.name, :title => title, :subject =>
>>>> @subj.short
>>>> response.should be_success
>>>> end
>>>> end
>>>> end
>>>>
>>>> A run of such specs will effectively multiply the tests — automatically —
>>>> choosing before and after blocks as specified.
>>>> I'm sorry, I haven't thought the DSL through, but I hope the main idea can
>>>> be seen: contexts do not have to be hierarchical.
>>>> In my opinion, adding some sort of context selection+combination
>>>> capabilities (AOP-style) will contribute greatly to the expressiveness of
>>>> the spec language.
>>>
>>> I think the idea of mixing/matching sub-contexts is very interesting, but
>>> it definitely needs from fleshing out. It would have to be easy to
>>> read/understand in the spec file as well as the output.
>>>
>>> Also, this only works if every combination should behave the same way. I
>>> think we'd need a means of saying "given these combinations of data, expect
>>> these outcomes".
>>>
>>> Anybody else have thoughts on this?
>>
>> It's a nice idea.
>>
>> I'm not sure whether I'd use it though. I think this idea comes from the
>> desire to write specs that are *complete*, which I can perfectly understand
>> but I don't think I subscribe to anymore. I prefer to really craft the
>> examples so there's 'just enough' tests but no more than that. I'd be
>> worried this might offer a temptation to think less about why you're writing
>> each example, and I'd be worried how that would help me to do TDD.
>>
>> It should be possible to do something like this using macros now, right? Can
>> I suggest that the OP has a go at refactoring his code using macros and we
>> can see how it looks?
>
> I know that the posted code may be a contrived toy example for the purposes
> of illustration, but when I see a spec like that alarm bells start to ring.
> So much mocking, so many assertions in each example block etc. And it's not
> at all clear what the pertinent behavior is that you want to test here,
> because each example looks exactly like a one-to-one rewrite of the original
> implementation that uses mocks instead of real objects.
>
> And when the alarm bells start to ring, before I think about changing my
> testing framework to make things easier, I look at the code under test to see
> if it could be changed to be more testable.
>
> So we basically have a controller action that accepts three parameters
> (author, title, and subject), and it has a conditional code path for two of
> those:
>
> if thing.exists
> great
> else
> create it
> end
>
> And in your spec you're wanting to test for all the different permutations of
> new author/existing author and new subject/existing subject.
>
> First thing you could do to eliminate a lot of mocking is use something like
> "find_or_create_by_name" and "find_or_create_by_short". Then you only have to
> mock three calls (one for each parameter) and forget about the permutations
> entirely.
>
> This is an example of pushing logic down into the model in order to make
> controllers simpler and more testable. If "find_or_create_by_*" doesn't do
> what you need it to, then create a model method which does.
>
> You could go even further and create a "register" method on your Book class
> which accepted the three parameters of author, title, and subject, and did
> everything which you are currently doing in your controller in the model
> layer instead. Then your controller spec becomes ridiculously simple, can be
> tested with a single mock, and the rest of the logic now resides in a model,
> which is easily testable.
>
> So whether or not the example was a toy example, the need for the any
> automatic permutation and spec generation in RSpec has disappeared. Let's
> imagine, however, that the need was still there. Would adding this kind of
> code to RSpec itself be a good idea?
>
> I don't necessarily think so. Matt says you can probably do this right now by
> using macros. I don't actually know what he means by that, but I do know that
> there are cases where I sometimes want a bunch of nearly identical specs, and
> I generate them in code using enumeration or some other means; ie. dumb
> example:
>
> [:foo, :bar, :baz].do |thing|
> describe "#{thing} dimensions" do
> it 'has length' do
> thing.to_s.length.should > 0
> end
> end
> end
That's the kind of thing I meant by macro, yes. Is that a well-understood use
of the word?
>
> So like I said, if your tests are painful, I think the first port of call
> should be to look at how the code under test could be change. Good code isn't
> just code that works. It's also code that is readable, maintainable, any
> among many other things, testable. Adding support to RSpec to make it easier
> to test bad code doesn't seem the right thing to do.
>
> Maybe you have another example that could illustrate how what you propose
> would be useful, but right now I don't really see the need for this kind of
> thing.
>
> Cheers,
> Wincent
>
>
>
> _______________________________________________
> rspec-users mailing list
> rspec-users@rubyforge.org
> http://rubyforge.org/mailman/listinfo/rspec-users
cheers,
Matt
http://blog.mattwynne.net
+44(0)7974 430184
_______________________________________________
rspec-users mailing list
rspec-users@rubyforge.org
http://rubyforge.org/mailman/listinfo/rspec-users
