Functional tests : rescue_action-rewrite causing problems (autotest)

rubyonrails-talk
1

Functional tests all come with a handy line that gives more sensible exception reporting, looking something like :

class FooController; def rescue_action(e) raise e end; end

This can cause a few problems. For example, assert_response :missing isn't going to work : the 404 status is no longer set on getting an ActiveRecord::RecordNotFound. Within functional tests, this isn't too bad : you'd probably be content with something like assert_raises(ActiveRecord::RecordNotFound){get 'show', :id => 40000}

However, I'm struggling to deal with this and autotest. AFAICT, autotest doesn't reload the controllers when moving between functional tests and integration tests, so controllers still have the rewritten rescue_action method in integration. So the exception gets caught at dispatcher.rb:42, which always returns a status of 500. The end result is that in autotest's integration testing, you might get 404 or you might get 500, depending on whether or not your functional tests got loaded first on that particular run.

Has anyone found a workaround for this? At the moment I'm having to remove the "def rescue_action(e) raise e end" line from functional tests, and it's making functional test debugging a bit painful...

Jon

2

No takers?

Jonathan del Strother wrote:

3

This is causing problems even outside autotest.

rake test

will run functionals, and in the same Ruby process, run the integration tests. Most if not all controllers will have the redefined rescue_action method...

The thing that should be done is to do it in setup, and replace that in teardown.

A bit like the mocks do. They replace a method, and on teardown, put the method back into place.

Bye !

4

Francois Beausoleil wrote:

> However, I'm struggling to deal with this and autotest. AFAICT, > autotest doesn't reload the controllers when moving between > functional tests and integration tests, so controllers still have the > rewritten rescue_action method in integration. So the exception gets > caught at dispatcher.rb:42, which always returns a status of 500. > The end result is that in autotest's integration testing, you might > get 404 or you might get 500, depending on whether or not your > functional tests got loaded first on that particular run.

This is causing problems even outside autotest.

rake test

will run functionals, and in the same Ruby process, run the integration tests. Most if not all controllers will have the redefined rescue_action method...

The thing that should be done is to do it in setup, and replace that in teardown.

Just to see if anyone can find a cleaner way, this is what I'm using at the moment in my test_helper :

unless ActionController::Base.method_defined? :rescue_action_without_optional_raise   class ActionController::Base     attr_accessor :reraise_exceptions

    def rescue_action_with_optional_raise(e)       if reraise_exceptions         raise e       else         rescue_action_without_optional_raise(e)       end     end     alias_method_chain :rescue_action, :optional_raise

    def initialize(opts={})       self.reraise_exceptions = opts[:reraise_exceptions]     end   end end

Then, in the functional tests' setup method, I can do @controller = PostsController.new(:reraise_exceptions => true), which will give me a controller that gives a nice stack trace, without polluting rescue_action's behaviour for integration testing.

Anyone spot a better way?

Jon