[rspec-users] customize description and failure messages when writing rspec matcher dsl

Anran Yang 08to09 at gmail.com
Sun Nov 27 08:22:06 EST 2011

On Nov 24, 2011, at 12:48 AM, Anran Yang wrote:

> Dear all,
> I recently tried to write a custom matcher using rspec dsl, which I'd
like to use as the following:
> ==========hpgc_spec.rb===================
> require "spec_helper"
> describe Hpgc do
>   specify do
>     should be_structured_as(
>     { :app => {
>         :has => [
>             :name,
>             :author,
>             :org,
>             :version,
>             :abstract,
>             :uri
>         ],
>         :can_be_a => {
>             :dataset => {
>                 :can_be_a => {
>                     :vector => {},
>                     :raster => {}
>                 }
>             },
>             :program => {
>                 :has => [
>                     :language,
>                     :parallel_programming_model,
>                     :parallel_performance_overview,
>                     :category
>                 ]
>             }
>         },
>         :has_many => {
>             :app_option => {
>                 :has => [
>                     :switch,
>                     :description
>                 ]
>             },
>             :app_io => {
>                 :has => [
>                     :name,
>                     :description,
>                     :is_in
>                 ]
>             }
>         }
>     }})
>   end
> end
> ===========================================
> The purpose is to test the entire model structure(attributes, inheritance
and association).

Then term Behavior Driven Development came into being, in part, to
discourage this approach to testing. We believe testing should be about
behavior, and that testing structures leads to highly coupled tests that
are hard to change.

I'd recommend thinking about the need for each of these attributes, come up
with examples of how they are used and write those instead. That said, see
below for more on how matchers work.

It made me thinking a while. But at last I found that I've thought about
it, though it may only happen unconsciously. So below is why I need this
kind of spec.
I must admit the Spec is against BDD's spirit. However, I think BDD may not
be enough for all kinds of projects in all kinds of circumstance. In my
project, the model itself is also a product, not only the "surface" of the
software. Maybe the best way to accomplish this is to separate this kind of
test and the "real" BDD test. But rspec can support them perfectly so I'm
lazy this time.
Maybe there are some other projects involved in this kind of situation? I
mean, some projects with a concrete domain model, so to some extent the
development is driven by both behavior and knowledge?

> Then I wrote following code to implement it:
> ========be_structured_as.rb===================
> # This is a matcher for test hpgc model structure
> RSpec::Matchers.define :be_structured_as do |structure|
>   def check(namespace, klass, desc)
>     (desc[:has_many] ||= {}).keys.each { |item|
klass.to_s.camelize.constantize.new.should have_many
item.to_s.pluralize.to_sym }
>     if !desc[:can_be_a] || desc[:can_be_a].empty?
>       obj = Factory.create(klass)
>       obj.should have_these_attributes (desc[:has] ||= [])
>     else
>       desc[:can_be_a].keys.map { |sub| sub.to_s }.should ==
>       desc[:can_be_a].each { |key, value| check namespace, key, value }
>     end
>   end
>   match do |namespace|
>     self.class.send :include, namespace
>     structure.each { |key, value| check namespace, key, value }
>   end
> end
> ===============================================
> To simplify the matcher definition,  I used some other machers like
"obj.should have_these_attributes (desc[:has] ||= [])", this works fine,
thanks to the excellent new matcher dsl syntax. Howerver, I got trouble
when running the test using "rspec spec/models/hpgc_spec.rb -fd". The test
passed but print
> <Screenshot.png>
> instead of something like "Hpgc should be structured as ...". I think it
may due to the other matcher I used because the message is from the matcher

That's correct. The default behavior is that the block passed to the
"match" method in the matcher DSL is expected to return a boolean. If it
raises an exception instead, that exception bubbles up and becomes the
source of the failure message.

> I try to add method "description" but seems it doesn't work.

"description" is used for the documentation formatter, not for failure
messages. If you run "rspec spec --format documentation" you'll see names
for all your examples.

Maybe I failed express my problem clearly, but the question is for pass
message, not failure message. Knowing how rspec raise exception help little
to get my spec say "Hpgc should be structured as... " when it passes. The
method "description" failed to present itself even when the spec passed, as
long as there is an inner "should" clause.

> And the failure message is also a problem. I search the web and come
across some solutions, but all of them need manually build the message.
I've not come out a clean way to do this(I must add some invasive clauses
to track the error) and think maybe the best behaviour would be that it
"throw" the failure message of inner matchers or false-value expressions.
But this time it just told me something like "Hpgc is expected to be
structured as ...".

Instead of using the "match" method in the matcher, you can use
"match_unless_raises", which returns false if it captures an exception,
letting the matcher control the failure message (which is what I think
you're looking for):

RSpec::Matchers.define :be_structured_as do |structure|
 def check(namespace, klass, desc)
   # ....

 match_unless_raises do |namespace|
   self.class.send :include, namespace
   structure.each { |key, value| check namespace, key, value }


 Is the method in rspec-expectation? I searched the local doc but didn't
find it. Also I don't quite understand how the method can help track the
failing message. It return false, but I still cannot figure out which line
fails, isn't it?

Thank you for your inspiring reply and tolerance to my poor English :-)


> Any idea?
> Best Regards,
> YANG Anran
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://rubyforge.org/pipermail/rspec-users/attachments/20111127/05996314/attachment-0001.html>

More information about the rspec-users mailing list