diff --git a/lib/puppet/transaction.rb b/lib/puppet/transaction.rb
index 53118755e..77bceec88 100644
--- a/lib/puppet/transaction.rb
+++ b/lib/puppet/transaction.rb
@@ -1,369 +1,370 @@
 require 'puppet'
 require 'puppet/util/tagging'
 require 'puppet/application'
 require 'digest/sha1'
 require 'set'
 
 # the class that actually walks our resource/property tree, collects the changes,
 # and performs them
 #
 # @api private
 class Puppet::Transaction
   require 'puppet/transaction/additional_resource_generator'
   require 'puppet/transaction/event'
   require 'puppet/transaction/event_manager'
   require 'puppet/transaction/resource_harness'
   require 'puppet/resource/status'
 
   attr_accessor :catalog, :ignoreschedules, :for_network_device
 
   # The report, once generated.
   attr_reader :report
 
   # Routes and stores any events and subscriptions.
   attr_reader :event_manager
 
   # Handles most of the actual interacting with resources
   attr_reader :resource_harness
 
   attr_reader :prefetched_providers
 
   include Puppet::Util
   include Puppet::Util::Tagging
 
   def initialize(catalog, report, prioritizer)
     @catalog = catalog
 
     @report = report || Puppet::Transaction::Report.new("apply", catalog.version, catalog.environment)
 
     @prioritizer = prioritizer
 
     @report.add_times(:config_retrieval, @catalog.retrieval_duration || 0)
 
     @event_manager = Puppet::Transaction::EventManager.new(self)
 
     @resource_harness = Puppet::Transaction::ResourceHarness.new(self)
 
     @prefetched_providers = Hash.new { |h,k| h[k] = {} }
   end
 
   # Invoke the pre_run_check hook in every resource in the catalog.
   # This should (only) be called by Transaction#evaluate before applying
   # the catalog.
   #
   # @see Puppet::Transaction#evaluate
   # @see Puppet::Type#pre_run_check
   # @raise [Puppet::Error] If any pre-run checks failed.
   # @return [void]
   def perform_pre_run_checks
     prerun_errors = {}
 
     @catalog.vertices.each do |res|
       begin
         res.pre_run_check
       rescue Puppet::Error => detail
         prerun_errors[res] = detail
       end
     end
 
     unless prerun_errors.empty?
       prerun_errors.each do |res, detail|
         res.log_exception(detail)
       end
       raise Puppet::Error, "Some pre-run checks failed"
     end
   end
 
   # This method does all the actual work of running a transaction.  It
   # collects all of the changes, executes them, and responds to any
   # necessary events.
   def evaluate(&block)
     block ||= method(:eval_resource)
     generator = AdditionalResourceGenerator.new(@catalog, relationship_graph, @prioritizer)
     @catalog.vertices.each { |resource| generator.generate_additional_resources(resource) }
 
     perform_pre_run_checks
 
     Puppet.info "Applying configuration version '#{catalog.version}'" if catalog.version
 
     continue_while = lambda { !stop_processing? }
 
     post_evalable_providers = Set.new
     pre_process = lambda do |resource|
       prov_class = resource.provider.class
       post_evalable_providers << prov_class if prov_class.respond_to?(:post_resource_eval)
 
       prefetch_if_necessary(resource)
 
       # If we generated resources, we don't know what they are now
       # blocking, so we opt to recompute it, rather than try to track every
       # change that would affect the number.
       relationship_graph.clear_blockers if generator.eval_generate(resource)
     end
 
     providerless_types = []
     overly_deferred_resource_handler = lambda do |resource|
       # We don't automatically assign unsuitable providers, so if there
       # is one, it must have been selected by the user.
+      return if missing_tags?(resource)
       if resource.provider
         resource.err "Provider #{resource.provider.class.name} is not functional on this host"
       else
         providerless_types << resource.type
       end
 
       resource_status(resource).failed = true
     end
 
     canceled_resource_handler = lambda do |resource|
       resource_status(resource).skipped = true
       resource.debug "Transaction canceled, skipping"
     end
 
     teardown = lambda do
       # Just once per type. No need to punish the user.
       providerless_types.uniq.each do |type|
         Puppet.err "Could not find a suitable provider for #{type}"
       end
 
       post_evalable_providers.each do |provider|
         begin
           provider.post_resource_eval
         rescue => detail
           Puppet.log_exception(detail, "post_resource_eval failed for provider #{provider}")
         end
       end
     end
 
     relationship_graph.traverse(:while => continue_while,
                                 :pre_process => pre_process,
                                 :overly_deferred_resource_handler => overly_deferred_resource_handler,
                                 :canceled_resource_handler => canceled_resource_handler,
                                 :teardown => teardown) do |resource|
       if resource.is_a?(Puppet::Type::Component)
         Puppet.warning "Somehow left a component in the relationship graph"
       else
         resource.info "Starting to evaluate the resource" if Puppet[:evaltrace] and @catalog.host_config?
         seconds = thinmark { block.call(resource) }
         resource.info "Evaluated in %0.2f seconds" % seconds if Puppet[:evaltrace] and @catalog.host_config?
       end
     end
 
     Puppet.debug "Finishing transaction #{object_id}"
   end
 
   # Wraps application run state check to flag need to interrupt processing
   def stop_processing?
     Puppet::Application.stop_requested? && catalog.host_config?
   end
 
   # Are there any failed resources in this transaction?
   def any_failed?
     report.resource_statuses.values.detect { |status| status.failed? }
   end
 
   # Find all of the changed resources.
   def changed?
     report.resource_statuses.values.find_all { |status| status.changed }.collect { |status| catalog.resource(status.resource) }
   end
 
   def relationship_graph
     catalog.relationship_graph(@prioritizer)
   end
 
   def resource_status(resource)
     report.resource_statuses[resource.to_s] || add_resource_status(Puppet::Resource::Status.new(resource))
   end
 
   # The tags we should be checking.
   def tags
     self.tags = Puppet[:tags] unless defined?(@tags)
 
     super
   end
 
   def prefetch_if_necessary(resource)
     provider_class = resource.provider.class
     return unless provider_class.respond_to?(:prefetch) and !prefetched_providers[resource.type][provider_class.name]
 
     resources = resources_by_provider(resource.type, provider_class.name)
 
     if provider_class == resource.class.defaultprovider
       providerless_resources = resources_by_provider(resource.type, nil)
       providerless_resources.values.each {|res| res.provider = provider_class.name}
       resources.merge! providerless_resources
     end
 
     prefetch(provider_class, resources)
   end
 
   private
 
   # Apply all changes for a resource
   def apply(resource, ancestor = nil)
     status = resource_harness.evaluate(resource)
     add_resource_status(status)
     event_manager.queue_events(ancestor || resource, status.events) unless status.failed?
   rescue => detail
     resource.err "Could not evaluate: #{detail}"
   end
 
   # Evaluate a single resource.
   def eval_resource(resource, ancestor = nil)
     if skip?(resource)
       resource_status(resource).skipped = true
     else
       resource_status(resource).scheduled = true
       apply(resource, ancestor)
     end
 
     # Check to see if there are any events queued for this resource
     event_manager.process_events(resource)
   end
 
   def failed?(resource)
     s = resource_status(resource) and s.failed?
   end
 
   # Does this resource have any failed dependencies?
   def failed_dependencies?(resource)
     # First make sure there are no failed dependencies.  To do this,
     # we check for failures in any of the vertexes above us.  It's not
     # enough to check the immediate dependencies, which is why we use
     # a tree from the reversed graph.
     found_failed = false
 
 
     # When we introduced the :whit into the graph, to reduce the combinatorial
     # explosion of edges, we also ended up reporting failures for containers
     # like class and stage.  This is undesirable; while just skipping the
     # output isn't perfect, it is RC-safe. --daniel 2011-06-07
     suppress_report = (resource.class == Puppet::Type.type(:whit))
 
     relationship_graph.dependencies(resource).each do |dep|
       next unless failed?(dep)
       found_failed = true
 
       # See above. --daniel 2011-06-06
       unless suppress_report then
         resource.notice "Dependency #{dep} has failures: #{resource_status(dep).failed}"
       end
     end
 
     found_failed
   end
 
   # A general method for recursively generating new resources from a
   # resource.
   def generate_additional_resources(resource)
     return unless resource.respond_to?(:generate)
     begin
       made = resource.generate
     rescue => detail
       resource.log_exception(detail, "Failed to generate additional resources using 'generate': #{detail}")
     end
     return unless made
     made = [made] unless made.is_a?(Array)
     made.uniq.each do |res|
       begin
         res.tag(*resource.tags)
         @catalog.add_resource(res)
         res.finish
         add_conditional_directed_dependency(resource, res)
         generate_additional_resources(res)
       rescue Puppet::Resource::Catalog::DuplicateResourceError
         res.info "Duplicate generated resource; skipping"
       end
     end
   end
 
   # Should we ignore tags?
   def ignore_tags?
     ! @catalog.host_config?
   end
 
   def resources_by_provider(type_name, provider_name)
     unless @resources_by_provider
       @resources_by_provider = Hash.new { |h, k| h[k] = Hash.new { |h, k| h[k] = {} } }
 
       @catalog.vertices.each do |resource|
         if resource.class.attrclass(:provider)
           prov = resource.provider && resource.provider.class.name
           @resources_by_provider[resource.type][prov][resource.name] = resource
         end
       end
     end
 
     @resources_by_provider[type_name][provider_name] || {}
   end
 
   # Prefetch any providers that support it, yo.  We don't support prefetching
   # types, just providers.
   def prefetch(provider_class, resources)
     type_name = provider_class.resource_type.name
     return if @prefetched_providers[type_name][provider_class.name]
     Puppet.debug "Prefetching #{provider_class.name} resources for #{type_name}"
     begin
       provider_class.prefetch(resources)
     rescue => detail
       Puppet.log_exception(detail, "Could not prefetch #{type_name} provider '#{provider_class.name}': #{detail}")
     end
     @prefetched_providers[type_name][provider_class.name] = true
   end
 
   def add_resource_status(status)
     report.add_resource_status(status)
   end
 
   # Is the resource currently scheduled?
   def scheduled?(resource)
     self.ignoreschedules or resource_harness.scheduled?(resource)
   end
 
   # Should this resource be skipped?
   def skip?(resource)
     if missing_tags?(resource)
       resource.debug "Not tagged with #{tags.join(", ")}"
     elsif ! scheduled?(resource)
       resource.debug "Not scheduled"
     elsif failed_dependencies?(resource)
       # When we introduced the :whit into the graph, to reduce the combinatorial
       # explosion of edges, we also ended up reporting failures for containers
       # like class and stage.  This is undesirable; while just skipping the
       # output isn't perfect, it is RC-safe. --daniel 2011-06-07
       unless resource.class == Puppet::Type.type(:whit) then
         resource.warning "Skipping because of failed dependencies"
       end
     elsif resource.virtual?
       resource.debug "Skipping because virtual"
     elsif !host_and_device_resource?(resource) && resource.appliable_to_host? && for_network_device
       resource.debug "Skipping host resources because running on a device"
     elsif !host_and_device_resource?(resource) && resource.appliable_to_device? && !for_network_device
       resource.debug "Skipping device resources because running on a posix host"
     else
       return false
     end
     true
   end
 
   def host_and_device_resource?(resource)
     resource.appliable_to_host? && resource.appliable_to_device?
   end
 
   def handle_qualified_tags( qualified )
     # The default behavior of Puppet::Util::Tagging is
     # to split qualified tags into parts. That would cause
     # qualified tags to match too broadly here.
     return
   end
 
   # Is this resource tagged appropriately?
   def missing_tags?(resource)
     return false if ignore_tags?
     return false if tags.empty?
 
     not resource.tagged?(*tags)
   end
 end
 
 require 'puppet/transaction/report'