[rspec-users] New RSpec methods to Object proposal: should_all and should_none

Rodrigo Rosenfeld Rosas lbocseg at yahoo.com.br
Wed Dec 9 17:56:42 EST 2009


Paul Hinze escreveu:
> Wincent Colaiuta <win at wincent.com> on 2009-12-09 at 13:39:
>   
>> El 09/12/2009, a las 19:15, David Chelimsky escribió:
>>
>>     
>>> On Wed, Dec 9, 2009 at 11:55 AM, Pat Maddox <mailinglists at patmaddox.com 
>>>       
>>>> wrote:
>>>> [@admin, @allowed_user].should all(be_allowed_to_visit(url))
>>>> [@admin, @allowed_user].should all_be_allowed_to_visit(url)
>>>>         
>>>> On Dec 9, 2009, at 5:27 AM, David Chelimsky wrote:
>>>>
>>>> On Wed, Dec 9, 2009 at 5:41 AM, Rodrigo Rosenfeld Rosas
>>>> <lbocseg at yahoo.com.br> wrote:
>>>>         
>>>>> I was thinking that it would be great to add 2 additional methods to
>>>>> Object: should_all and should_none.
>>>>>
>>>>> The idea is that we would be able to write tests like:
>>>>>
>>>>> [@admin, @allowed_user].should_all be_allowed_to_visit(url)
>>>>>
>>>>> [@unprivileged, @non_welcome].should_none be_allowed_to_visit(url)
>>>>>
>>>>> Implementation is trivial, but I think that tests would become much
>>>>> cleaner than:
>>>>>
>>>>> [@admin, @allowed_user].each{|u| u.should be_allowed_to_visit(url)}
>>>>>
>>>>> Any thoughts on that?
>>>>>           
>>>> How about:
>>>> each_of(@admin, @allowed_user).should be_allowed_to_visit(url)
>>>> none_of(@admin, @allowed_user).should be_allowed_to_visit(url)
>>>> This gets the cleanliness without adding to Object.
>>>>         
>> I'm puzzled as to why people are so focussed on making specs read like  
>> plain text English when they are still developer-facing Ruby code.
>>
>> Especially suprised in this case of wanting to avoid the "each + block" 
>> enumeration idiom, which is about as "bread and butter" Ruby as you can 
>> get, readable to anybody who's ever read the first chapter of a Ruby book.
>>     
>
> For me the benefit would be clearer error messages on failures, because
> of the potential to continue execution through the entirety of the
> collection of inputs, rather than bailing on the first failed 'should'.
>
> This allowing a developer to diagnose classes of problems with the
> collection rather than seeing failures one-by-one.
>
> Compare the following two options: 
>
> [:foo, :bar, :baz, :qux].each { |x| x.should ==(:qux) }
>   <run 01> "Expected :foo to equal :qux"
>            (...fix :foo...)
>   <run 02> "Expected :bar to equal :qux"
>            (...fix :bar...)
>   <run 03> "Expected :baz to equal :qux"
>            (...realize the _real_ problem is in dependent code that
>             affects all of foo bar and baz, and fix that...)
>   <run 04> PASS
>
> each_of(:foo, :bar, :baz, :qux).should ==(:qux)
>   <run 01> "Expected each of :foo, :bar, :baz, :qux to equal :qux, but
>            :foo, :bar, :baz did not."
>            (...fix dependent code that affects all of foo bar and
>            baz...)
>   <run 02> PASS
>   
I've read all the proposed syntax and I really like them all.
I don't think the syntax below is as clear as the others proposed but I 
understand that it may help improving the failure message:

[@admin, @allowed_user].should all(be_allowed_to_visit(url))

The syntax each_of/none_of is the clearer to me.

The point for the tests to read like English sentences is that it easies 
test reading. Of course programmers will understand the statement using 
Ruby. But when you have tons of tests to read, this kind of improvement 
can make tests easier to be read, even by non-technical users when using 
XP techniques for instance if you try to pair-reading the tests to see 
if both programmer and client are expecting the same results...

Also, the Java community is experiencing the use of JRuby and RSpec to 
test their Java applications and maybe they would prefer the 
"should_all/should all" syntax.

I don't like Cucumber, for instance, because I would take much more time 
to write the stories from what I take writing the tests directly in 
RSpec. And I personally find specs as simple to read as Cucumber 
stories. But let's not get on this discussion, please...

The point that the spec doesn't need to stop in the first error from the 
collection items is also a good one.

There is also another point. Some failure messages don't mention the 
object being tested. In that case, the error message may be unclear 
because one won't know which of the tested items caused the error when 
using an "each" constructor.

Anyway, it is not a serious limitation of Rspec since it is trivial to 
write such solutions if we want in our test suite. I can certainly leave 
without these extensions, but I think that if they get into Rspec core, 
it would be standardized among different projects and probably better 
documented, which makes it easier to integrate a new team member.

Thank you for your proposals.

Best regards,

Rodrigo.


__________________________________________________
Faça ligações para outros computadores com o novo Yahoo! Messenger 
http://br.beta.messenger.yahoo.com/ 



More information about the rspec-users mailing list