graphql 2.0.27 → 2.2.5

data/lib/graphql/dataloader/async_dataloader.rb

@@ -0,0 +1,85 @@
+# frozen_string_literal: true
+module GraphQL
+  class Dataloader
+    class AsyncDataloader < Dataloader
+      def yield
+        if (condition = Thread.current[:graphql_dataloader_next_tick])
+          condition.wait
+        else
+          Fiber.yield
+        end
+        nil
+      end
+
+      def run
+        job_fibers = []
+        next_job_fibers = []
+        source_tasks = []
+        next_source_tasks = []
+        first_pass = true
+        sources_condition = Async::Condition.new
+        manager = spawn_fiber do
+          while first_pass || job_fibers.any?
+            first_pass = false
+
+            while (f = (job_fibers.shift || spawn_job_fiber))
+              if f.alive?
+                finished = run_fiber(f)
+                if !finished
+                  next_job_fibers << f
+                end
+              end
+            end
+            job_fibers.concat(next_job_fibers)
+            next_job_fibers.clear
+
+            Sync do |root_task|
+              while source_tasks.any? || @source_cache.each_value.any? { |group_sources| group_sources.each_value.any?(&:pending?) }
+                while (task = source_tasks.shift || spawn_source_task(root_task, sources_condition))
+                  if task.alive?
+                    root_task.yield # give the source task a chance to run
+                    next_source_tasks << task
+                  end
+                end
+                sources_condition.signal
+                source_tasks.concat(next_source_tasks)
+                next_source_tasks.clear
+              end
+            end
+          end
+        end
+
+        manager.resume
+        if manager.alive?
+          raise "Invariant: Manager didn't terminate successfully: #{manager}"
+        end
+
+      rescue UncaughtThrowError => e
+        throw e.tag, e.value
+      end
+
+      private
+
+      def spawn_source_task(parent_task, condition)
+        pending_sources = nil
+        @source_cache.each_value do |source_by_batch_params|
+          source_by_batch_params.each_value do |source|
+            if source.pending?
+              pending_sources ||= []
+              pending_sources << source
+            end
+          end
+        end
+
+        if pending_sources
+          fiber_vars = get_fiber_variables
+          parent_task.async do
+            set_fiber_variables(fiber_vars)
+            Thread.current[:graphql_dataloader_next_tick] = condition
+            pending_sources.each(&:run_pending_keys)
+          end
+        end
+      end
+    end
+  end
+end

data/lib/graphql/dataloader/source.rb

@@ -88,6 +88,7 @@ module GraphQL
         raise "Implement `#{self.class}#fetch(#{keys.inspect}) to return a record for each of the keys"
       end
 
+      MAX_ITERATIONS = 1000
       # Wait for a batch, if there's anything to batch.
       # Then run the batch and update the cache.
       # @return [void]
@@ -96,8 +97,8 @@ module GraphQL
         iterations = 0
         while pending_result_keys.any? { |key| !@results.key?(key) }
           iterations += 1
-          if iterations > 1000
-            raise "#{self.class}#sync tried 1000 times to load pending keys (#{pending_result_keys}), but they still weren't loaded. There is likely a circular dependency."
+          if iterations > MAX_ITERATIONS
+            raise "#{self.class}#sync tried #{MAX_ITERATIONS} times to load pending keys (#{pending_result_keys}), but they still weren't loaded. There is likely a circular dependency."
           end
           @dataloader.yield
         end
@@ -161,7 +162,14 @@ module GraphQL
         [*batch_args, **batch_kwargs]
       end
 
-      attr_reader :pending
+      # Clear any already-loaded objects for this source
+      # @return [void]
+      def clear_cache
+        @results.clear
+        nil
+      end
+
+      attr_reader :pending, :results
 
       private
 

data/lib/graphql/dataloader.rb

@@ -27,11 +27,12 @@ module GraphQL
       attr_accessor :default_nonblocking
     end
 
-    AsyncDataloader = Class.new(self) { self.default_nonblocking = true }
+    NonblockingDataloader = Class.new(self) { self.default_nonblocking = true }
 
     def self.use(schema, nonblocking: nil)
       schema.dataloader_class = if nonblocking
-        AsyncDataloader
+        warn("`nonblocking: true` is deprecated from `GraphQL::Dataloader`, please use `GraphQL::Dataloader::AsyncDataloader` instead. Docs: https://graphql-ruby.org/dataloader/async_dataloader.")
+        NonblockingDataloader
       else
         self
       end
@@ -61,6 +62,32 @@ module GraphQL
       @nonblocking
     end
 
+    # This is called before the fiber is spawned, from the parent context (i.e. from
+    # the thread or fiber that it is scheduled from).
+    #
+    # @return [Hash<Symbol, Object>] Current fiber-local variables
+    def get_fiber_variables
+      fiber_vars = {}
+      Thread.current.keys.each do |fiber_var_key|
+        # This variable should be fresh in each new fiber
+        if fiber_var_key != :__graphql_runtime_info
+          fiber_vars[fiber_var_key] = Thread.current[fiber_var_key]
+        end
+      end
+      fiber_vars
+    end
+
+    # Set up the fiber variables in a new fiber.
+    #
+    # This is called within the fiber, right after it is spawned.
+    #
+    # @param vars [Hash<Symbol, Object>] Fiber-local variables from {get_fiber_variables}
+    # @return [void]
+    def set_fiber_variables(vars)
+      vars.each { |k, v| Thread.current[k] = v }
+      nil
+    end
+
     # Get a Source instance from this dataloader, for calling `.load(...)` or `.request(...)` on.
     #
     # @param source_class [Class<GraphQL::Dataloader::Source]
@@ -104,6 +131,15 @@ module GraphQL
       nil
     end
 
+    # Clear any already-loaded objects from {Source} caches
+    # @return [void]
+    def clear_cache
+      @source_cache.each do |_source_class, batched_sources|
+        batched_sources.each_value(&:clear_cache)
+      end
+      nil
+    end
+
     # Use a self-contained queue for the work in the block.
     def run_isolated
       prev_queue = @pending_jobs
@@ -128,124 +164,98 @@ module GraphQL
     ensure
       @pending_jobs = prev_queue
       prev_pending_keys.each do |source_instance, pending|
-        source_instance.pending.merge!(pending)
+        pending.each do |key, value|
+          if !source_instance.results.key?(key)
+            source_instance.pending[key] = value
+          end
+        end
       end
     end
 
-    # @api private Move along, move along
     def run
-      if @nonblocking && !Fiber.scheduler
-        raise "`nonblocking: true` requires `Fiber.scheduler`, assign one with `Fiber.set_scheduler(...)` before executing GraphQL."
-      end
-      # At a high level, the algorithm is:
-      #
-      #  A) Inside Fibers, run jobs from the queue one-by-one
-      #    - When one of the jobs yields to the dataloader (`Fiber.yield`), then that fiber will pause
-      #    - In that case, if there are still pending jobs, a new Fiber will be created to run jobs
-      #    - Continue until all jobs have been _started_ by a Fiber. (Any number of those Fibers may be waiting to be resumed, after their data is loaded)
-      #  B) Once all known jobs have been run until they are complete or paused for data, run all pending data sources.
-      #    - Similarly, create a Fiber to consume pending sources and tell them to load their data.
-      #    - If one of those Fibers pauses, then create a new Fiber to continue working through remaining pending sources.
-      #    - When a source causes another source to become pending, run the newly-pending source _first_, since it's a dependency of the previous one.
-      #  C) After all pending sources have been completely loaded (there are no more pending sources), resume any Fibers that were waiting for data.
-      #    - Those Fibers assume that source caches will have been populated with the data they were waiting for.
-      #    - Those Fibers may request data from a source again, in which case they will yeilded and be added to a new pending fiber list.
-      #  D) Once all pending fibers have been resumed once, return to `A` above.
-      #
-      # For whatever reason, the best implementation I could find was to order the steps `[D, A, B, C]`, with a special case for skipping `D`
-      # on the first pass. I just couldn't find a better way to write the loops in a way that was DRY and easy to read.
-      #
-      pending_fibers = []
-      next_fibers = []
-      pending_source_fibers = []
+      job_fibers = []
+      next_job_fibers = []
+      source_fibers = []
       next_source_fibers = []
       first_pass = true
-
-      while first_pass || (f = pending_fibers.shift)
-        if first_pass
+      manager = spawn_fiber do
+        while first_pass || job_fibers.any?
           first_pass = false
-        else
-          # These fibers were previously waiting for sources to load data,
-          # resume them. (They might wait again, in which case, re-enqueue them.)
-          resume(f)
-          if f.alive?
-            next_fibers << f
-          end
-        end
 
-        while @pending_jobs.any?
-          # Create a Fiber to consume jobs until one of the jobs yields
-          # or jobs run out
-          f = spawn_fiber {
-            while (job = @pending_jobs.shift)
-              job.call
+          while (f = (job_fibers.shift || spawn_job_fiber))
+            if f.alive?
+              finished = run_fiber(f)
+              if !finished
+                next_job_fibers << f
+              end
             end
-          }
-          resume(f)
-          # In this case, the job yielded. Queue it up to run again after
-          # we load whatever it's waiting for.
-          if f.alive?
-            next_fibers << f
-          end
-        end
-
-        if pending_fibers.empty?
-          # Now, run all Sources which have become pending _before_ resuming GraphQL execution.
-          # Sources might queue up other Sources, which is fine -- those will also run before resuming execution.
-          #
-          # This is where an evented approach would be even better -- can we tell which
-          # fibers are ready to continue, and continue execution there?
-          #
-          if (first_source_fiber = create_source_fiber)
-            pending_source_fibers << first_source_fiber
           end
+          join_queues(job_fibers, next_job_fibers)
 
-          while pending_source_fibers.any?
-            while (outer_source_fiber = pending_source_fibers.pop)
-              resume(outer_source_fiber)
-              if outer_source_fiber.alive?
-                next_source_fibers << outer_source_fiber
-              end
-              if (next_source_fiber = create_source_fiber)
-                pending_source_fibers << next_source_fiber
+          while source_fibers.any? || @source_cache.each_value.any? { |group_sources| group_sources.each_value.any?(&:pending?) }
+            while (f = source_fibers.shift || spawn_source_fiber)
+              if f.alive?
+                finished = run_fiber(f)
+                if !finished
+                  next_source_fibers << f
+                end
               end
             end
-            join_queues(pending_source_fibers, next_source_fibers)
-            next_source_fibers.clear
+            join_queues(source_fibers, next_source_fibers)
           end
-          # Move newly-enqueued Fibers on to the list to be resumed.
-          # Clear out the list of next-round Fibers, so that
-          # any Fibers that pause can be put on it.
-          join_queues(pending_fibers, next_fibers)
-          next_fibers.clear
         end
       end
 
-      if @pending_jobs.any?
-        raise "Invariant: #{@pending_jobs.size} pending jobs"
-      elsif pending_fibers.any?
-        raise "Invariant: #{pending_fibers.size} pending fibers"
-      elsif next_fibers.any?
-        raise "Invariant: #{next_fibers.size} next fibers"
+      run_fiber(manager)
+
+      if manager.alive?
+        raise "Invariant: Manager fiber didn't terminate properly."
       end
-      nil
-    end
 
-    def join_queues(previous_queue, next_queue)
-      if @nonblocking
-        Fiber.scheduler.run
-        next_queue.select!(&:alive?)
+      if job_fibers.any?
+        raise "Invariant: job fibers should have exited but #{job_fibers.size} remained"
+      end
+      if source_fibers.any?
+        raise "Invariant: source fibers should have exited but #{source_fibers.size} remained"
       end
-      previous_queue.concat(next_queue)
+    rescue UncaughtThrowError => e
+      throw e.tag, e.value
+    end
+
+    def run_fiber(f)
+      f.resume
+    end
+
+    def spawn_fiber
+      fiber_vars = get_fiber_variables
+      Fiber.new(blocking: !@nonblocking) {
+        set_fiber_variables(fiber_vars)
+        yield
+        # With `.transfer`, you have to explicitly pass back to the parent --
+        # if the fiber is allowed to terminate normally, control is passed to the main fiber instead.
+        true
+      }
     end
 
     private
 
-    # If there are pending sources, return a fiber for running them.
-    # Otherwise, return `nil`.
-    #
-    # @return [Fiber, nil]
-    def create_source_fiber
+    def join_queues(prev_queue, new_queue)
+      @nonblocking && Fiber.scheduler.run
+      prev_queue.concat(new_queue)
+      new_queue.clear
+    end
+
+    def spawn_job_fiber
+      if @pending_jobs.any?
+        spawn_fiber do
+          while job = @pending_jobs.shift
+            job.call
+          end
+        end
+      end
+    end
+
+    def spawn_source_fiber
       pending_sources = nil
       @source_cache.each_value do |source_by_batch_params|
         source_by_batch_params.each_value do |source|
@@ -257,55 +267,12 @@ module GraphQL
       end
 
       if pending_sources
-        # By passing the whole array into this Fiber, it's possible that we set ourselves up for a bunch of no-ops.
-        # For example, if you have sources `[a, b, c]`, and `a` is loaded, then `b` yields to wait for `d`, then
-        # the next fiber would be dispatched with `[c, d]`. It would fulfill `c`, then `d`, then eventually
-        # the previous fiber would start up again. `c` would no longer be pending, but it would still receive `.run_pending_keys`.
-        # That method is short-circuited since it isn't pending any more, but it's still a waste.
-        #
-        # This design could probably be improved by maintaining a `@pending_sources` queue which is shared by the fibers,
-        # similar to `@pending_jobs`. That way, when a fiber is resumed, it would never pick up work that was finished by a different fiber.
-        source_fiber = spawn_fiber do
+        spawn_fiber do
           pending_sources.each(&:run_pending_keys)
         end
       end
-
-      source_fiber
-    end
-
-    def resume(fiber)
-      fiber.resume
-    rescue UncaughtThrowError => e
-      throw e.tag, e.value
-    end
-
-    # Copies the thread local vars into the fiber thread local vars. Many
-    # gems (such as RequestStore, MiniRacer, etc.) rely on thread local vars
-    # to keep track of execution context, and without this they do not
-    # behave as expected.
-    #
-    # @see https://github.com/rmosolgo/graphql-ruby/issues/3449
-    def spawn_fiber
-      fiber_locals = {}
-
-      Thread.current.keys.each do |fiber_var_key|
-        # This variable should be fresh in each new fiber
-        if fiber_var_key != :__graphql_runtime_info
-          fiber_locals[fiber_var_key] = Thread.current[fiber_var_key]
-        end
-      end
-
-      if @nonblocking
-        Fiber.new(blocking: false) do
-          fiber_locals.each { |k, v| Thread.current[k] = v }
-          yield
-        end
-      else
-        Fiber.new do
-          fiber_locals.each { |k, v| Thread.current[k] = v }
-          yield
-        end
-      end
     end
   end
 end
+
+require "graphql/dataloader/async_dataloader"

data/lib/graphql/duration_encoding_error.rb

@@ -0,0 +1,16 @@
+# frozen_string_literal: true
+module GraphQL
+  # This error is raised when `Types::ISO8601Duration` is asked to return a value
+  # that cannot be parsed as an ISO8601-formatted duration by ActiveSupport::Duration.
+  #
+  # @see GraphQL::Types::ISO8601Duration which raises this error
+  class DurationEncodingError < GraphQL::RuntimeTypeError
+    # The value which couldn't be encoded
+    attr_reader :duration_value
+
+    def initialize(value)
+      @duration_value = value
+      super("Duration cannot be parsed: #{value}. \nDuration must be an ISO8601-formatted duration.")
+    end
+  end
+end

data/lib/graphql/execution/interpreter/runtime/graphql_result.rb

@@ -0,0 +1,170 @@
+# frozen_string_literal: true
+
+module GraphQL
+  module Execution
+    class Interpreter
+      class Runtime
+        module GraphQLResult
+          def initialize(result_name, parent_result, is_non_null_in_parent)
+            @graphql_parent = parent_result
+            if parent_result && parent_result.graphql_dead
+              @graphql_dead = true
+            end
+            @graphql_result_name = result_name
+            @graphql_is_non_null_in_parent = is_non_null_in_parent
+            # Jump through some hoops to avoid creating this duplicate storage if at all possible.
+            @graphql_metadata = nil
+          end
+
+          def path
+            @path ||= build_path([])
+          end
+
+          def build_path(path_array)
+            graphql_result_name && path_array.unshift(graphql_result_name)
+            @graphql_parent ? @graphql_parent.build_path(path_array) : path_array
+          end
+
+          attr_accessor :graphql_dead
+          attr_reader :graphql_parent, :graphql_result_name, :graphql_is_non_null_in_parent
+
+          # @return [Hash] Plain-Ruby result data (`@graphql_metadata` contains Result wrapper objects)
+          attr_accessor :graphql_result_data
+        end
+
+        class GraphQLResultHash
+          def initialize(_result_name, _parent_result, _is_non_null_in_parent)
+            super
+            @graphql_result_data = {}
+          end
+
+          include GraphQLResult
+
+          attr_accessor :graphql_merged_into
+
+          def set_leaf(key, value)
+            # This is a hack.
+            # Basically, this object is merged into the root-level result at some point.
+            # But the problem is, some lazies are created whose closures retain reference to _this_
+            # object. When those lazies are resolved, they cause an update to this object.
+            #
+            # In order to return a proper top-level result, we have to update that top-level result object.
+            # In order to return a proper partial result (eg, for a directive), we have to update this object, too.
+            # Yowza.
+            if (t = @graphql_merged_into)
+              t.set_leaf(key, value)
+            end
+
+            @graphql_result_data[key] = value
+            # keep this up-to-date if it's been initialized
+            @graphql_metadata && @graphql_metadata[key] = value
+
+            value
+          end
+
+          def set_child_result(key, value)
+            if (t = @graphql_merged_into)
+              t.set_child_result(key, value)
+            end
+            @graphql_result_data[key] = value.graphql_result_data
+            # If we encounter some part of this response that requires metadata tracking,
+            # then create the metadata hash if necessary. It will be kept up-to-date after this.
+            (@graphql_metadata ||= @graphql_result_data.dup)[key] = value
+            value
+          end
+
+          def delete(key)
+            @graphql_metadata && @graphql_metadata.delete(key)
+            @graphql_result_data.delete(key)
+          end
+
+          def each
+            (@graphql_metadata || @graphql_result_data).each { |k, v| yield(k, v) }
+          end
+
+          def values
+            (@graphql_metadata || @graphql_result_data).values
+          end
+
+          def key?(k)
+            @graphql_result_data.key?(k)
+          end
+
+          def [](k)
+            (@graphql_metadata || @graphql_result_data)[k]
+          end
+
+          def merge_into(into_result)
+            self.each do |key, value|
+              case value
+              when GraphQLResultHash
+                next_into = into_result[key]
+                if next_into
+                  value.merge_into(next_into)
+                else
+                  into_result.set_child_result(key, value)
+                end
+              when GraphQLResultArray
+                # There's no special handling of arrays because currently, there's no way to split the execution
+                # of a list over several concurrent flows.
+                into_result.set_child_result(key, value)
+              else
+                # We have to assume that, since this passed the `fields_will_merge` selection,
+                # that the old and new values are the same.
+                into_result.set_leaf(key, value)
+              end
+            end
+            @graphql_merged_into = into_result
+          end
+        end
+
+        class GraphQLResultArray
+          include GraphQLResult
+
+          def initialize(_result_name, _parent_result, _is_non_null_in_parent)
+            super
+            @graphql_result_data = []
+          end
+
+          def graphql_skip_at(index)
+            # Mark this index as dead. It's tricky because some indices may already be storing
+            # `Lazy`s. So the runtime is still holding indexes _before_ skipping,
+            # this object has to coordinate incoming writes to account for any already-skipped indices.
+            @skip_indices ||= []
+            @skip_indices << index
+            offset_by = @skip_indices.count { |skipped_idx| skipped_idx < index}
+            delete_at_index = index - offset_by
+            @graphql_metadata && @graphql_metadata.delete_at(delete_at_index)
+            @graphql_result_data.delete_at(delete_at_index)
+          end
+
+          def set_leaf(idx, value)
+            if @skip_indices
+              offset_by = @skip_indices.count { |skipped_idx| skipped_idx < idx }
+              idx -= offset_by
+            end
+            @graphql_result_data[idx] = value
+            @graphql_metadata && @graphql_metadata[idx] = value
+            value
+          end
+
+          def set_child_result(idx, value)
+            if @skip_indices
+              offset_by = @skip_indices.count { |skipped_idx| skipped_idx < idx }
+              idx -= offset_by
+            end
+            @graphql_result_data[idx] = value.graphql_result_data
+            # If we encounter some part of this response that requires metadata tracking,
+            # then create the metadata hash if necessary. It will be kept up-to-date after this.
+            (@graphql_metadata ||= @graphql_result_data.dup)[idx] = value
+            value
+          end
+
+          def values
+            (@graphql_metadata || @graphql_result_data)
+          end
+        end
+      end
+    end
+  end
+end