@tahanabavi/typefetch-query-core

Setup

The client is the whole setup surface. Declare invalidation once, here — no call site downstream ever names a cache key:

import { QueryClient } from '@tahanabavi/typefetch-query-core'
import { api } from './api' // a typefetch ApiClient

const client = new QueryClient({
  relations: { 'user.updateUser': ['user.getUser'] },
})

relations reads "<mutation id>": ["<query id>", …]. The id is the contract's own stable endpointId (typesocket calls it eventId) — the engine resolves either, which is exactly why the same client drives both transports.

Cross-tab seams

Two more options exist for adapters that mirror the cache across tabs or replay it offline. Both are additive — leave them off and the engine runs exactly as above.

import { collectSources } from '@tahanabavi/typefetch-query-core'

const client = new QueryClient({
  // Wrap every fetch and mutation. Run the request, or resolve without it to
  // hand back a value another tab already broadcast.
  gate: (ctx, run) => run(),
  // Turn an endpoint id back into its endpoint, so an inbound mirror can write
  // through the typed `setQueryData` — even for an endpoint this tab has yet to
  // mount. `collectSources` builds the map from a client's `.modules` tree.
  sources: collectSources(api.modules),
})

setQueryData takes a matching { updatedAt } so a mirrored value keeps the origin's timestamp rather than looking freshly fetched. Each of these is a hook a sync or offline adapter attaches to; none is needed to use the engine on its own.