user_mailer.rb call to uninitialized AppConfig.first causes RSpec to fatally fail to start

[shippy]

When RSpec is run with the default ... line uncommented, RSpec fails to start. The reason is that there is no AppConfig.first record while in test mode. Therefore, the call to

class UserMailer < ActionMailer::Base
  default :from => AppConfig.first.admin_email, :cc => AppConfig.first.admin_email
  # (snip)

results in the error

jane-ribeira (Simon): ~/Coding/reservations [396_model_tests|✚ 2] 
05:41 PM $ rspec
/Users/Simon/Coding/reservations/app/mailers/user_mailer.rb:6:in `<class:UserMailer>': undefined method `admin_email' for nil:NilClass (NoMethodError)
    from /Users/Simon/Coding/reservations/app/mailers/user_mailer.rb:2:in `<top (required)>'
(snip)

This if-statement fixes the problem temporarily:

class UserMailer < ActionMailer::Base
  # Workaround so that RSpec start-up doesn't fail. TODO: Have RSpec initialize AppConfig with configuration.
  unless (AppConfig.first.nil?)
    default :from => AppConfig.first.admin_email, :cc => AppConfig.first.admin_email
  end
# (snip)

The workaround prevents the line from being executed and ensures that RSpec can run, but also means that we cannot test the proper setup of user_mailer.rb. It is not clear, however, at what stage of test-environment loading this error happens and how it can be prevented.

That may prove difficult, as reversing the condition and debugging in the following manner caused the first Ruby segmentation fault I have ever seen:

class UserMailer < ActionMailer::Base
  if (AppConfig.first.nil?)
    binding.pry
    default :from => AppConfig.first.admin_email, :cc => AppConfig.first.admin_email
  end

jane-ribeira (Simon): ~/Coding/reservations [396_model_tests|✚ 1] 
05:45 PM $ rspec

Frame number: 0/25
Frame type: class

From: /Users/Simon/Coding/reservations/app/mailers/user_mailer.rb @ line 7 :

/Users/Simon/.rvm/gems/ruby-1.9.3-p327/gems/coderay-1.0.8/lib/coderay.rb:170: [BUG] Segmentation fault
ruby 1.9.3p327 (2012-11-10 revision 37606) [x86_64-darwin11.4.2]

-- Control frame information -----------------------------------------------
c:0081 p:0013 s:0315 b:0315 l:000314 d:000314 METHOD /Users/Simon/.rvm/gems/ruby-1.9.3-p327/gems/coderay-1.0.8/lib/coderay.rb:170
c:0080 p:0031 s:0308 b:0305 l:000296 d:000304 BLOCK  /Users/Simon/.rvm/gems/ruby-1.9.3-p327/gems/pry-0.9.10/lib/pry/code.rb:312
c:0079 p:---- s:0302 b:0302 l:000301 d:000301 FINISH
c:0078 p:---- s:0300 b:0300 l:000299 d:000299 CFUNC  :each
c:0077 p:0044 s:0297 b:0297 l:000296 d:000296 METHOD /Users/Simon/.rvm/gems/ruby-1.9.3-p327/gems/pry-0.9.10/lib/pry/code.rb:311
c:0076 p:0011 s:0292 b:0292 l:000291 d:000291 METHOD /Users/Simon/.rvm/gems/ruby-1.9.3-p327/gems/pry-0.9.10/lib/pry/code.rb:391
(snip)

[mnquintana]

@shippy Is this still an issue?

[orenyk]

I'm having a related issue; RSpec is running but all of the UserMailer tests are failing since there's no valid admin e-mail to serve as a "from" address for the tested e-mails. I'm sure there must be a way to initialize the test db but I'm not sure what the best solution would be.

[shippy]

Confirmed: If I revert the workaround from 4a7163e, the issue still persists. (It's worth mentioning that I spent some time trying to replicate with guard/spork running before I undid the commit, and the tests passed because the Rails environment had already been loaded.)

One way of avoiding this is to replicate AdminMailer and simply make the default from: "no-reply@reservations.app". But IIRC some of the abuses in Bass were uncovered because a senior admin went through all the e-mails that were sent out and that she was cc:ed on, so that's clearly a valuable feature.

As for pre-populating the test database with data, we like to start each spec with a clean slate. (For one, the AppConfig-related specs actually test app reactions to an empty database.) But maybe there's a way to let RSpec create an app-minimum setup in the spork.prefork? But that would require spork to run all the time, which is not practical...

[orenyk]

Also, weren't we switching away from Spork to use spring instead?

[shippy]

Eh, I'm dumb. This is an easy solution that feeds the wolf and leaves the goats whole:

class UserMailer < ActionMailer::Base
  # Workaround so that RSpec start-up doesn't fail. TODO: Have RSpec initialize AppConfig with configuration.
  unless (AppConfig.first.nil?)
    default from: AppConfig.first.admin_email, cc: AppConfig.first.admin_email
  else
    default from: 'no-reply@reservations.app'
  end

[orenyk]

Yea, that should do it :-)

[shippy]

See PR #643.

[shippy]

With #643 merged, this is no longer an issue. It's worth noting that we've implemented a workaround, not an actual solution -- and we might want to figure out a way to set up the testing environment with an AppConfig prior to any loading at a later date. That said, closing.