The warden method in the Devise::TestHelpers module adds a Warden proxy
object to the request environment hash under the 'warden' key. Including
this module in your test case registers that method as a callback, which
runs before every test:
https://github.com/plataformatec/devise/blob/v3.4.1/lib/devise/test_helpers.rb#L12
The request object itself is created in a callback added by Rails:
https://github.com/rails/rails/blob/v4.2.0/actionpack/lib/action_controller/test_case.rb#L687
So before each test runs, the Rails callback creates the request object,
and then the Devise callback adds a Warden proxy object to it.
I was using the rspec-retry gem (https://github.com/y310/rspec-retry),
and noticed that my controller specs would always fail whenever they
were retried with this error:
NoMethodError: undefined method `authenticate!' for nil:NilClass
When rspec-retry re-runs a failed test, it runs the setup callbacks
again. The Rails callback creates a new request object, but because of
the memoization that was here before, the Devise callback wouldn't add a
Warden proxy to it, which was causing the error.
With this change, the Warden setup code will still only run once as long
as the request object stays the same, but if it changes a new Warden
proxy will be added to the new request object.
There's no real need to pass 2 variables to the view to figure that out,
we can simply display the message relying on whether or not the
`@minimum_password_length` variable is present.
Allow to ensure valid upon confirming.
We might want to consider confirmation status in validations. For example, maybe we want to require certain fields upon confirmation, but not at registration.
BCrypt has a limit of 72 characters for the password. Anything beyond 72
characters is truncated. This commit updates the validation to limit
passwords to less than 72 characters, keeping within the limitation.
As we are already slicing the Hash, we must be sure that this method will send
a "safe" object down to the other finder methods that will use the Hash entries
to do the querying.
... is enabled.
The old code may cause conflicts when developing an engine (i.e. invalid
route name) and according to @josevalim the reason for it to remain is
unknown.
