Fix handling of cached values in `Rack::Request`. (#2054)

* Per-class cache keys for cached query/body parameters.

* Use the query parser class as the default cache key.

2 files changed, 171 insertions, 49 deletions

diff --git a/lib/rack/request.rb b/lib/rack/request.rb
index e6969645..86e133cd 100644
--- a/lib/rack/request.rb
+++ b/lib/rack/request.rb
@@ -480,25 +480,114 @@ module Rack
         PARSEABLE_DATA_MEDIA_TYPES.include?(media_type)
       end
 
-      # Returns the data received in the query string.
-      def GET
-        if get_header(RACK_REQUEST_QUERY_STRING) == query_string
-          if query_hash = get_header(RACK_REQUEST_QUERY_HASH)
-            return query_hash
+      # Given a current input value, and a validity key, check if the cache
+      # is valid, and if so, return the cached value. If not, yield the
+      # current value to the block, and set the cache to the result.
+      #
+      # This method does not use cache_key, so it is shared between all
+      # instance of Rack::Request and it's sub-classes.
+      private def cache_for(key, validity_key, current_value)
+        # Get the current value of the validity key and compare it with the input value:
+        if get_header(validity_key).equal?(current_value)
+          # If the values are the same, then the cache is valid, so return the cached value.
+          if has_header?(key)
+            value = get_header(key)
+            # If the cached value is an exception, then re-raise it.
+            if value.is_a?(Exception)
+              raise value.class, value.message, cause: value.cause
+            else
+              # Otherwise, return the cached value.
+              return value
+            end
           end
         end
 
-        set_header(RACK_REQUEST_QUERY_HASH, expand_params(query_param_list))
+        # If the cache is not valid, then yield the current value to the block:
+        value = yield(current_value)
+
+        # Set the validity key to the current value so that we can detect changes:
+        set_header(validity_key, current_value)
+
+        # Set the cache to the result of the block, and return the result:
+        set_header(key, value)
+      rescue => error
+        # If an exception is raised, then set the cache to the exception, and re-raise it:
+        set_header(validity_key, current_value)
+        set_header(key, error)
+        raise
+      end
+
+      # This cache key is used by cached values generated by class_cache_for,
+      # specfically GET and POST. This is to ensure that the cache is not
+      # shared between instances of different classes which have different
+      # behaviour. This includes sub-classes that override query_parser or
+      # expand_params.
+      def cache_key
+        query_parser.class
+      end
+
+      # Given a current input value, and a validity key, check if the cache
+      # is valid, and if so, return the cached value. If not, yield the
+      # current value to the block, and set the cache to the result.
+      #
+      # This method uses cache_key to ensure that the cache is not shared
+      # between instances of different classes which have different
+      # behaviour of the cached operations.
+      private def class_cache_for(key, validity_key, current_value)
+        # The cache is organised in the env as:
+        # env[key][cache_key] = value
+        # and is valid as long as env[validity_key].equal?(current_value)
+
+        cache_key = self.cache_key
+
+        # Get the current value of the validity key and compare it with the input value:
+        if get_header(validity_key).equal?(current_value)
+          # Lookup the cache for the current cache key:
+          if cache = get_header(key)
+            if cache.key?(cache_key)
+              # If the cache is valid, then return the cached value.
+              value = cache[cache_key]
+              if value.is_a?(Exception)
+                # If the cached value is an exception, then re-raise it.
+                raise value.class, value.message, cause: value.cause
+              else
+                # Otherwise, return the cached value.
+                return value
+              end
+            end
+          end
+        end
+
+        # If the cache was not defined for this cache key, then create a new cache:
+        unless cache
+          set_header(key, cache = {})
+        end
+
+        begin
+          # Yield the current value to the block to generate an updated value:
+          value = yield(current_value)
+
+          # Only set this after generating the value, so that if an error or other cache depending on the same key, it will be invalidated correctly:
+          set_header(validity_key, current_value)
+          return cache[cache_key] = value
+        rescue => error
+          set_header(validity_key, current_value)
+          cache[cache_key] = error
+          raise
+        end
+      end
+
+      # Returns the data received in the query string.
+      def GET
+        class_cache_for(RACK_REQUEST_QUERY_HASH, RACK_REQUEST_QUERY_STRING, query_string) do
+          expand_params(query_param_list)
+        end
       end
 
       def query_param_list
-        if get_header(RACK_REQUEST_QUERY_STRING) == query_string
-          get_header(RACK_REQUEST_QUERY_PAIRS)
-        else
-          query_pairs = split_query(query_string, '&')
-          set_header RACK_REQUEST_QUERY_STRING, query_string
-          set_header RACK_REQUEST_QUERY_HASH, nil
-          set_header(RACK_REQUEST_QUERY_PAIRS, query_pairs)
+        cache_for(RACK_REQUEST_QUERY_PAIRS, RACK_REQUEST_QUERY_STRING, query_string) do
+          set_header(RACK_REQUEST_QUERY_HASH, nil)
+          split_query(query_string, '&')
         end
       end
 
@@ -507,33 +596,13 @@ module Rack
       # This method support both application/x-www-form-urlencoded and
       # multipart/form-data.
       def POST
-        if get_header(RACK_REQUEST_FORM_INPUT).equal?(get_header(RACK_INPUT))
-          if form_hash = get_header(RACK_REQUEST_FORM_HASH)
-            return form_hash
-          end
+        class_cache_for(RACK_REQUEST_FORM_HASH, RACK_REQUEST_FORM_INPUT, get_header(RACK_INPUT)) do
+          expand_params(body_param_list)
         end
-
-        set_header(RACK_REQUEST_FORM_HASH, expand_params(body_param_list))
       end
 
       def body_param_list
-        if error = get_header(RACK_REQUEST_FORM_ERROR)
-          raise error.class, error.message, cause: error.cause
-        end
-
-        begin
-          rack_input = get_header(RACK_INPUT)
-
-          form_pairs = nil
-
-          # If the form data has already been memoized from the same
-          # input:
-          if get_header(RACK_REQUEST_FORM_INPUT).equal?(rack_input)
-            if form_pairs = get_header(RACK_REQUEST_FORM_PAIRS)
-              return form_pairs
-            end
-          end
-
+        cache_for(RACK_REQUEST_FORM_PAIRS, RACK_REQUEST_FORM_INPUT, get_header(RACK_INPUT)) do |rack_input|
           if rack_input.nil?
             form_pairs = []
           elsif form_data? || parseable_data?
@@ -544,19 +613,16 @@ module Rack
               # form_vars.sub!(/\0\z/, '') # performance replacement:
               form_vars.slice!(-1) if form_vars.end_with?("\0")
 
-              set_header RACK_REQUEST_FORM_VARS, form_vars
+              # Removing this line breaks Rail test "test_filters_rack_request_form_vars"!
+              set_header(RACK_REQUEST_FORM_VARS, form_vars)
+
               form_pairs = split_query(form_vars, '&')
             end
           else
             form_pairs = []
           end
-
-          set_header RACK_REQUEST_FORM_INPUT, rack_input
-          set_header RACK_REQUEST_FORM_HASH, nil
-          set_header(RACK_REQUEST_FORM_PAIRS, form_pairs)
-        rescue => error
-          set_header(RACK_REQUEST_FORM_ERROR, error)
-          raise
+          
+          form_pairs
         end
       end
 
diff --git a/test/spec_request.rb b/test/spec_request.rb
index 2a3f792a..e525621a 100644
--- a/test/spec_request.rb
+++ b/test/spec_request.rb
@@ -1554,12 +1554,19 @@ EOF
     rack_input.write(input)
     rack_input.rewind
 
-    req = make_request Rack::MockRequest.env_for("/",
-                      "rack.request.form_hash" => { 'foo' => 'bar' },
-                      "rack.request.form_input" => rack_input,
-                      :input => rack_input)
+    form_hash_cache = {}
+
+    req = make_request Rack::MockRequest.env_for(
+      "/",
+      "rack.request.form_hash" => form_hash_cache,
+      "rack.request.form_input" => rack_input,
+      :input => rack_input
+    )
 
-    req.POST.must_equal req.env['rack.request.form_hash']
+    form_hash = {'foo' => 'bar'}.freeze
+    form_hash_cache[req.cache_key] = form_hash
+
+    req.POST.must_equal form_hash
   end
 
   it "conform to the Rack spec" do
@@ -1957,4 +1964,53 @@ EOF
       DelegateRequest.new super(env)
     end
   end
+
+  class UpperRequest < Rack::Request
+    def expand_params(parameters)
+      parameters.map do |(key, value)|
+        [key.upcase, value]
+      end.to_h
+    end
+
+    # If this is not specified, the behaviour becomes order dependent.
+    def cache_key
+      :my_request
+    end
+  end
+
+  it "correctly expands parameters" do
+    env = {"QUERY_STRING" => "foo=bar"}
+
+    request = Rack::Request.new(env)
+    request.query_param_list.must_equal [["foo", "bar"]]
+    request.GET.must_equal "foo" => "bar"
+
+    upper_request = UpperRequest.new(env)
+    upper_request.query_param_list.must_equal [["foo", "bar"]]
+    upper_request.GET.must_equal "FOO" => "bar"
+
+    env['QUERY_STRING'] = "foo=bar&bar=baz"
+
+    request.GET.must_equal "foo" => "bar", "bar" => "baz"
+    upper_request.GET.must_equal "FOO" => "bar", "BAR" => "baz"
+  end
+
+  class BrokenRequest < Rack::Request
+    def expand_params(parameters)
+      raise "boom"
+    end
+  end
+
+  it "raises an error if expand_params raises an error" do
+    env = {"QUERY_STRING" => "foo=bar"}
+
+    request = Rack::Request.new(env)
+    request.GET.must_equal "foo" => "bar"
+
+    broken_request = BrokenRequest.new(env)
+    lambda { broken_request.GET }.must_raise RuntimeError
+
+    # Subsequnt calls also raise an error:
+    lambda { broken_request.GET }.must_raise RuntimeError
+  end
 end