Agenda

This was a brainstorming session amongst the core team with the goal of cataloging pain points and outstanding feature requests for product developers using Relay. We also discussed the high-level value proposition of Relay.

Product Requirements

This list covers pain points, work-in-progress features, and new features.

• Performance
  • fast startup & time to interaction (TTI)
  • managing resource consumption (including cache eviction/GC)
  • maintaining app responsiveness
• Control over Consistency / Staleness
  • when to force fetch?
  • controlling staleness → leads to force fetching
  • force fetching → can lead to missing fields on records & missing records in lists
• Mutations
  • consistency with non-Relay portions of an application
  • range/other configs are confusing, easy to mess up w/o good feedback
  • performance proportional to what is visible, not entire history of what has been fetched
  • know when optimistic updates are missing data relative to the fat query and warn developer to update
• Scalability
  • isolating different sections of an app w.r.t. to perf/consistency
  • different scheduling policies (sync/async)
  • different caching/staleness policies
• Understanding variables
  • how to access route params in containers?
  • prepareVariables - undocumented, executed unpredictably, makes persisting queries difficult
  • confusion over the need to pass variables in the query and in props (note: it's to differentiate the same child being embedded twice in the same parent but with different props).
• Non-Facebook GraphQL
  • more pagination options (e.g. windowed pagination or non-cursor based)
  • remove Facebook GraphQL-isms such as node, viewer, etc.
• Developer experience
  • derive flow types from fragments
  • solution for null-checks everywhere
• New Features
  • observing data outside of containers
  • manually updating the store (ex: appending to a list w/o a mutation)
  • local schema & mutations and/or redux integration
  • server rendering in OSS
  • work offline in OSS
  • subscriptions in OSS
  • GraphQL-less Relay (or, optionally, Relay w/o a GraphQL server)
  • preloading data for a predicted subsequent view

Value Proposition

We discussed the high-level value proposition of Relay and what we think are the most and least important aspects to maintain going forward.

Core Values

• Fast by default; performance is a feature.
• Control of consistency and updates.
• No manual data-fetching.
• Co-location of data dependencies and view logic.
• Aligns with the React mentality: component composition and local reasoning.
• Global efficiency (wins for one app benefit all).

Non-Goals

• Absolute consistency. This has never been an explicit guarantee, and providing controlled means to relax consistency could open up new possibilities in the design space.
• Declarative connections (i.e. first(n) and Relay diffs for you). We've found this can be useful for deeply nested views and may be less useful for top-level list views.
• Fat/tracked queries: implementation detail.
• Diff queries: implementation detail.
• Disk cache outside of native apps (though likely desired in OSS).