Warden::Manager#initialize clobbers default_strategies

[ginjo]

When passing a hash of options, or a Warden::Config instance to Warden::Manager#initialize (where the 'use' method eventually hits in a middleware instance), any passed :default_strategies are clobbered - or rather corrupted - and end up breaking the authenticate! method. This also happens when passing :scope_defaults into Warden::Manager#initialize. The only workable way to send config into Warden::Manager#initialize is with the block in the 'use' method. The expected behavior would seem to be that a hash, a Warden::Config instance, or a block passed to Warden::Manager#initialize would result in a manager instance with identical configuration.

Breaking example:

new_manager = Warden::Manager.new(Proc.new{}, :default_strategies=>{:default=>[:password]})
puts new_manager.to_yaml

--- !ruby/object:Warden::Manager
config: !ruby/hash:Warden::Config
  :default_scope: :default
  :scope_defaults: {}
  :default_strategies:
    :_all:
    - :default
    - :password
  :intercept_401: true
app: !ruby/object:Proc {}

Notice that default_strategies now has a key :_all, and worse, is now looking for a strategy called :default, which will raise an exception (unless skip_missing_strategies is always set, or unless a strategy named :default is explicitly defined).

Another breaking example:

new_manager = Warden::Manager.new(Proc.new{}, :scope_defaults=>{:default => {:strategies=>[:password]}})
puts new_manager.to_yaml

--- !ruby/object:Warden::Manager
config: !ruby/hash:Warden::Config
  :scope_defaults:
    :default:
      :strategies:
      - :password
  :default_scope: :default
  :default_strategies: {}
  :intercept_401: true
app: !ruby/object:Proc {}

In this case, the default_strategies are not being constructed by Warden::Config, and Warden::Manager does not correct the missing data. Is this intended behavior, or is this an issue with Warden::Config?