I'm having an issue with testing my controllers and using Warden.
All examples point at stubbing request.env['warden']. This causes issues in my controllers when I call env['warden'], which then returns nil.
For a crude example, using this:
request.env['warden'] = double(Warden, :authenticate => nil,
:authenticate! => nil,
:authenticated? => false)
And a simple before filter like this:
before_filter do
redirect_to new_user_session_url unless env['warden'].authenticated?
end
I get a nil.
I just managed to get it working using controller.env['warden'] = ... and it works.
This makes sense, since it's sitting right at the controller level, so I guess my question is what wouldn't it work in the I've seen all examples.
I have this in my spec_helper:
config.include Warden::Test::Helpers
Any help would be great!
I wrote controller test helpers for Warden.
http://kentaroimai.com/articles/1-controller-test-helpers-for-warden
Despite many examples telling you to implement Warden through env['warden'] in your Rails app. It seems the correct way to access it through request.env['warden'].
It found this out by raising env in my controllers during tests, and this always came out nil.
It seems in Warden, https://github.com/hassox/warden/blob/master/lib/warden/proxy.rb#L13
There is an accessor for the rack environment, which won't exist in test mode due to the absence of Rack in controller tests. Please someone check this.
So when stubbing request.env in RSpec, your implementation needs to point at request.env.
It seems a necessary evil in my mind. But if there is anyone with a good explanation or work around, I'd love to continue this discussion.
Related
I am trying to define some helper methods to be used in the app's controller, but it seems that rails don't even call the controller. just for the test I have the following controller in my app/controllers/my_engine/application_controller.rb and as the documents say rails should find it first and an error should raise because THIS_SHOULD_PRODUCE_ERROR is unknown, but the rspec happily executing without any errors!
class ApplicationController < ActionController::Base
THIS_SHOULD_PRODUCE_ERROR
end
I even tried to mimic the devise's way but the results are the same!
The guide section on the app directory suggests that the application_controller in an engine "will provide any common functionality for the controllers of the engine".
So I wouldn't expect that any additions to that controller will be available to all controllers in an application.
That also means that your application_controller is, I suspect, not getting called when you're running your test. Which would explain why you're not seeing an error.
In terms of how devise does it I think you need to be looking at how define_helpers works. The code you've linked to in your question is the application controller in the test app for the devise gem.
I noticed that I have got things wrong, and the application_controller in the engine does not get applied to application_controller in the app! Also, I couldn't figure out how the devise did it, but I have come up with the simple workaround for this:
require_relative 'controllers/helpers'
module Acu
module Injectors
class << self
ActiveSupport::Notifications.subscribe "start_processing.action_controller" do |**args|
eval((Acu::Configs.get :base_controller).to_s).class_eval do
include Acu::Controllers::Helpers
end
end
end
end
end
This will inject controller helpers to the user's base controller (which I get from the user, default: :ApplicationController) at the end of the class, which is perfect for me (but don't know how to add it to begging of the class if anyone needs it)
Does Ruby on Rails 3 (3.0.7) offer a mechanism to reload the class cache on demand? I am running some tests where I overwrite some methods in ApplicationController and in some of my models to stub out authentication. For example, I call a method (I'm using Cucumber for testing) in some of my tags (Before('#tag')) that does this:
ApplicationController.send(:define_method, :current_user) do
#current_user ||= User.where(:id => cookies[:user_id]).first
end
...where the actual current_user method in ApplicationController has a different behavior.
This works great until I need to run some tests in which I do not want to stub out those methods. With config.cache_classes = true set in my test environment, ApplicationController is not reinitialized without my monkey patch, making the tests I don't want to stub out fail. With config.cache_classes = false, my monkey patch is forgotten on the next request, causing all of the tests that I need to stub to fail.
Ideally, I would like to be able to erase the class cache in an After('#tag') method, if this is possible. How can I accomplish that? Or is there a better way of stubbing out certain methods in certain scenarios that I am overlooking?
You could take inspiration from this great SO answer, and make good use of ActionDispatch::Callbacks.
ActionDispatch::Reloader.cleanup!
ActionDispatch::Reloader.prepare!
I posted the rationale behind this over here: Why does code need to be reloaded in Rails 3?
If its bad to cross post the same answer, kindly let me know how its preferred to post an answer thats relevant to two questions...I'm happy to oblige.
As of newer Rails (> 4.0), I was able to reload class definitions, in console, for instance, with reload!.
I've tried reading through various blog posts that attempt to explain alias_method_chain and the reasons to use it and not use it. In particular, I took heed to:
http://weblog.rubyonrails.org/2006/4/26/new-in-rails-module-alias_method_chain
and
http://yehudakatz.com/2009/03/06/alias_method_chain-in-models/
I still do not see any practical use for alias_method_chain. Would anyone be able to explain a few things.
1 - is it still used at all?
2 - when would you use alias_method_chain and why?
1 - is it still used at all?
Apparently yes, alias_method_chain() is still used in Rails (as of version 3.0.0).
2 - when would you use
alias_method_chain and why?
(Note: the following is largely based on the discussion of alias_method_chain() in Metaprogramming Ruby by Paolo Perrotta, which is an excellent book that you should get your hands on.)
Let's start with a basic example:
class Klass
def salute
puts "Aloha!"
end
end
Klass.new.salute # => Aloha!
Now suppose that we want to surround Klass#salute() with logging behavior. We can do that what Perrotta calls an around alias:
class Klass
def salute_with_log
puts "Calling method..."
salute_without_log
puts "...Method called"
end
alias_method :salute_without_log, :salute
alias_method :salute, :salute_with_log
end
Klass.new.salute
# Prints the following:
# Calling method...
# Aloha!
# ...Method called
We defined a new method called salute_with_log() and aliased it to salute(). The code that used to call salute() still works, but it gets the new logging behavior as well. We also defined an alias to the original salute(), so we can still salute without logging:
Klass.new.salute_without_log # => Aloha!
So, salute() is now called salute_without_log(). If we want logging, we can call either salute_with_log() or salute(), which are aliases of the same method. Confused? Good!
According to Perrotta, this kind of around alias is very common in Rails:
Look at another example of Rails
solving a problem its own way. A few
versions ago, the Rails code contained
many instances of the same idiom: an
Around Alias (155) was used to add a
feature to a method, and the old
version of the method was renamed to
something like
method_without_feature(). Apart from
the method names, which changed every
time, the code that did this was
always the same, duplicated all over
the place. In most languages, you
cannot avoid that kind of duplication.
In Ruby, you can sprinkle some
metaprogramming magic over your
pattern and extract it into its own
method... and thus was born
alias_method_chain().
In other words, you provide the original method, foo(), and the enhanced method, foo_with_feature(), and you end up with three methods: foo(), foo_with_feature(), and foo_without_feature(). The first two include the feature, while the third doesn't. Instead of duplicating these aliases all around, alias_method_chain() provided by ActiveSupport does all the aliasing for you.
alias_method_chain has been deprecated in Rails 5 in favour of Module#prepend.
Pull request: https://github.com/rails/rails/pull/19434
Changelog: https://github.com/rails/rails/blob/b292b76c2dd0f04fb090d49b90716a0e6037b41a/guides/source/5_0_release_notes.md#deprecations-4
I'm not sure if it's gone out of style with Rails 3 or not, but it is still actively used in versions before that.
You use it to inject some functionality before (or after) a method is called, without modifying any place that calls that method. See this example:
module SwitchableSmtp
module InstanceMethods
def deliver_with_switchable_smtp!(mail = #mail)
unless logger.nil?
logger.info "Switching SMTP server to: #{custom_smtp.inspect}"
end
ActionMailer::Base.smtp_settings = custom_smtp unless custom_smtp.nil?
deliver_without_switchable_smtp!(mail = #mail)
end
end
def self.included(receiver)
receiver.send :include, InstanceMethods
receiver.class_eval do
alias_method_chain :deliver!, :switchable_smtp
end
end
end
That's an addition to ActionMailer to allow swapping out of the SMTP settings on each call to deliver!. By calling alias_method_chain you are able to define a method deliver_with_switchable_smtp! in which you do your custom stuff, and call deliver_without_switchable_smtp! from there when you're done.
alias_method_chain aliases the old deliver! to your new custom method, so the rest of your app doesn't even know deliver! now does your custom stuff too.
is it used at all?
Seems so. It's a common practice among Rails developers
when would you use alias_method_chain and why?
Despite the warnings, alias_method_chain is still the main strategy used when injecting functionality to an existing method, at least was in Rails 2.x and is followed by many people extending it. Yehuda ought to remove alias_method_chain from rails 3.0 to say from his posts and comments in Rails tickets. It is still used by many extensions that add custom behavior at certain points of the execution, such as loggers, error reporters, benchmarking, data injection, etc.
IMO, the best alternative is to include a module, thus you have decoration over delegation. (For example, follow example 4 in this post). That way you can alter the objects even individually if you'd like, without polluting the class' methods. The downside to this is that the method lookup chain increases for each module you inject, but this is what modules are for anyway.
Very interesting question, will keep a look on what other people think about it.
I've been writing RSpec tests for some Rails controllers and I've discovered a strong impulse to ensure that the Authlogic authentication is working properly. I also feel like I should be verifying that each action uses the same application-wide layout. However, writing tests for this behavior in every single action seems silly.
What I'd like to see are one-line matchers for filters and layouts, similar to Shoulda's matchers for associations and verifications. Unfortunately, no such matchers seem to be available (except for some Test::Unit macros for filters in this blog post). I'm tempted to just write them myself, but not being able to find anyone who's already done it makes me question whether or not a need for such matchers actually exists.
So my question is, how do you test your controllers' common behavior (if you test it at all), and would one-liner matchers testing filters and layouts be useful? Myself, I'm trying to decide between one-liners in the controller specs combined with speccing the filter explicitly, or just speccing the filter and ignoring the filters and layouts in the controllers (since they're only one line of code anyway).
I don't like the idea of writing specs for filters -- that seems too close to the implementation. If you had used TDD/BDD methods to build your controller from scratch, presumably you'd have written the action first, added some logic (e.g. to handle authentication) and then realized it should go into a filter instead. If your spec is along the lines of "Reject an index request if the current user is not the account user", your spec ought to be able to do something like (aircode):
current_user = Factory.create(:unauthorized)
controller.should_not receive(:index)
get :index
request.should redirect_to(some_safe_path)
And it doesn't matter whether the action is using a filter or not.
You can DRY up controller specs with Rspec macros. So (more hand-waving):
describe MyController do
should_reject_anonymous(self)
...
end
module ControllerMacros
def should_reject_anonymous(test_controller)
describe test_controller, "Authentication" do
it "rejects index" do
test_controller.should_not_receive(:index)
get :index
response.should redirect_to(some_safe_path)
end
end
end
end
I am using Clearance for authentication with my rails app which works. But now all my unit tests fail because they redirect to the sign in page, which makes sense. Clearance seems to have helper functions for fixing that but I can only find them for Shoulda. Are there equivalent helpers for Test::Unit?
I thought there'd be more 'magic' to how the shoulda macros work but it's actually really simple. To add the sign_in behavior just add this to the test_helper.rb
def sign_in_as(user)
#controller.current_user = user
return user
end