• Eliminated a race condition that would cause RelayGarbageCollector to fatal when, in the middle of a readRelayDiskCache traversal, a container attempts to subscribe to a record not yet registered with the garbage collector.
• The garbage collector now strictly increments references to all subscribed nodes, and strictly decrements references to all previously subscribed nodes, eliminating a class of race condition by ensuring an exact 1:1 correspondence of increment/decrement calls for a given node.
• Replaced GraphQLStoreDataHandler with RelayRecord, and added the RelayRecord#isRecord method.
• Introduced RelayQueryIndexPath which tracks fragment indexes to the nearest parent field during query traversal. This replaces the existing logic used to generate field serialization keys.
• Added RelayContext, a step toward making all Relay state contextual.
• Improved query printing performance via short circuiting and inlining.
• Moved record writing functions out of RelayRecordStore and into a new RelayRecordWriter class.
• The Babel plugin now prints canHaveSubselections metadata on object-like fields that can contain child fields, making it possible to determine if a given field in a query is a true leaf node, or an object-type field having no subselections. This replaces RelayQueryNode#isScalar with RelayQueryNode#canHaveSubselections.
• RANGE_DELETE mutation configs now allow you to specify an array path to a deleted node, rather than just a deletedIDFieldName.
• Renamed Records to RecordMap.
• Record reads from cache managers are now abortable. If the network request finishes before the cache read, for instance, the cache read can be cancelled.
• Added USERS.md; a catalog of products and developers that use Relay.
• You can now interpolate an array of fragments into a Relay.QL query. eg. ${containers.map(c => c.getFragment('foo'))}
• Fixed a bug where invalid queries could be printed due to different variables having the same value; duplicates are now avoided.