web-data-fetching-trpc（网络数据获取 trpc）

tRPC Type-Safe API Patterns

Quick Guide: tRPC provides end-to-end type safety by sharing TypeScript types directly from server to client -- no code generation, no schema files. Export AppRouter type from your router (this is the key bridge). Use Zod for input validation, TRPCError with proper codes for errors, and middleware for auth. v11 is the current stable version: transformer goes inside httpBatchLink(), subscriptions use async generators (not observable()), and @trpc/tanstack-react-query is the recommended React integration.

<critical_requirements>

CRITICAL: Before Using This Skill

(You MUST export AppRouter type from your tRPC router for client-side type inference)

(You MUST use TRPCError with appropriate error codes -- never throw raw Error objects)

(You MUST use Zod for input validation on ALL procedures accepting user input)

(You MUST place transformer inside httpBatchLink() in v11 -- NOT at client level)

</critical_requirements>

Auto-detection: tRPC router, initTRPC, createTRPCClient, createTRPCContext, @trpc/server, @trpc/client, @trpc/react-query, @trpc/tanstack-react-query, TRPCError, procedure, publicProcedure, protectedProcedure, query, mutation, subscription, httpBatchLink, queryOptions, mutationOptions, useTRPC

When to use:

• Building APIs in TypeScript monorepos with shared types
• End-to-end type safety without code generation
• Full-stack TypeScript applications where both client and server are TypeScript
• Projects where types should flow automatically from backend to frontend

When NOT to use:

• Public APIs consumed by third parties (use OpenAPI/REST)
• Non-TypeScript clients (mobile apps, other languages)
• Need HTTP caching at CDN level (tRPC uses POST by default)
• GraphQL requirements with partial queries

Key patterns covered:

• Router and procedure definition (initTRPC, router, procedure)
• Input validation with Zod schemas
• Context and middleware for authentication
• Error handling with TRPCError codes
• React integration via @trpc/tanstack-react-query (recommended) or @trpc/react-query (classic)
• Optimistic updates, infinite queries, subscriptions

Detailed Resources:

• examples/core.md - Router setup, CRUD, provider, type inference, queryOptions
• examples/middleware.md - Logging, rate limiting, org-scoped access
• examples/infinite-queries.md - Cursor pagination, infinite scroll
• examples/optimistic-updates.md - Optimistic updates with rollback
• examples/subscriptions.md - Async generator subscriptions, SSE
• examples/file-uploads.md - FormData file uploads (v11+)
• reference.md - Decision frameworks, error codes, anti-patterns, v11 migration

Philosophy

tRPC eliminates API layer friction by sharing types directly between server and client. No schemas to write, no code to generate -- export your router type and import it client-side for full autocompletion and type safety.

Core principles:

• Zero schema duplication: Types flow from backend to frontend automatically
• TypeScript-native: Leverages TypeScript's type inference, not code generation
• Procedure-based: Queries read data, mutations write data -- clear separation
• Composable middleware: Build reusable authentication and validation layers
• Built on TanStack Query: Full caching, invalidation, and optimistic updates via React Query

Trade-offs:

• Requires TypeScript on both ends (no polyglot support)
• Best in monorepos where types can be shared directly
• Not suitable for public APIs needing OpenAPI documentation
• Uses POST by default -- no HTTP caching without configuration

Core Patterns

Pattern 1: tRPC Initialization and Router Setup

Initialize tRPC once per application. Export the router and procedure factories.

import { initTRPC, TRPCError } from "@trpc/server";
import { ZodError } from "zod";
import type { Context } from "./context";

const t = initTRPC.context<Context>().create({
  errorFormatter({ shape, error }) {
    return {
      ...shape,
      data: {
        ...shape.data,
        zodError:
          error.cause instanceof ZodError ? error.cause.flatten() : null,
      },
    };
  },
});

export const router = t.router;
export const publicProcedure = t.procedure;
export const middleware = t.middleware;

Why good: Single initialization point, error formatter provides structured Zod errors to client, exported factories enable composition across router files

See examples/core.md Pattern 1 for complete router and context factory.

Pattern 2: Procedures with Zod Input Validation

Zod schemas provide runtime validation AND TypeScript inference from a single source.

const createUserSchema = z.object({
  email: z.string().email(),
  name: z.string().min(1).max(100),
});

export const userRouter = router({
  create: protectedProcedure
    .input(createUserSchema)
    .mutation(async ({ input, ctx }) => {
      // input is typed: { email: string; name: string }
      return ctx.db.user.create({ data: input });
    }),
});

// BAD: No input validation -- input is 'unknown'
publicProcedure.mutation(async ({ input }) => {
  return ctx.db.user.create({ data: input as any }); // Dangerous!
});

Why bad: Without Zod validation, input is unknown type, no runtime validation, injection risks, as any defeats TypeScript

See examples/core.md Pattern 2 for complete CRUD router.

Pattern 3: Authentication Middleware

Middleware narrows context types -- ctx.user becomes non-nullable after auth middleware.

const isAuthenticated = middleware(async ({ ctx, next }) => {
  if (!ctx.session || !ctx.user) {
    throw new TRPCError({ code: "UNAUTHORIZED" });
  }
  return next({ ctx: { ...ctx, session: ctx.session, user: ctx.user } });
});

export const protectedProcedure = publicProcedure.use(isAuthenticated);

Why good: Auth enforced at procedure definition, TypeScript narrows ctx.user to non-nullable, eliminates duplicated if-checks in every handler

See examples/middleware.md for logging, rate limiting, and org-scoped access patterns.

Pattern 4: AppRouter Type Export

This is the KEY to tRPC's type safety. Export the router type for client-side inference.

export const appRouter = router({
  user: userRouter,
  post: postRouter,
});

// THIS IS ESSENTIAL -- without it, clients have no type inference
export type AppRouter = typeof appRouter;

Use inferRouterInputs/inferRouterOutputs for extracting procedure types:

import type { inferRouterInputs, inferRouterOutputs } from "@trpc/server";
type RouterInputs = inferRouterInputs<AppRouter>;
type RouterOutputs = inferRouterOutputs<AppRouter>;

// Extract specific type
type User = RouterOutputs["user"]["getById"];

See examples/core.md Pattern 4 for complete type inference utilities.

Pattern 5: React Integration (v11 Recommended)

v11 introduces @trpc/tanstack-react-query with queryOptions/mutationOptions factories that work directly with TanStack Query hooks.

// Setup: createTRPCContext provides typed hooks
import { createTRPCContext } from "@trpc/tanstack-react-query";
export const { TRPCProvider, useTRPC } = createTRPCContext<AppRouter>();

// Usage: standard TanStack Query hooks with tRPC type safety
const trpc = useTRPC();
const { data } = useQuery(trpc.user.getById.queryOptions({ id: userId }));

v11 CRITICAL: Transformer must be inside httpBatchLink(), NOT at createTRPCClient() level.

// BAD: v11 error
createTRPCClient({ transformer: superjson, links: [...] });

// GOOD: transformer inside the link
httpBatchLink({ url: "/api/trpc", transformer: superjson });

See examples/core.md Patterns 3 and 5 for complete provider and component setup.

Pattern 6: Error Handling with TRPCError

Use standardized error codes that map to HTTP status codes.

// Server: throw TRPCError with appropriate code
throw new TRPCError({
  code: "NOT_FOUND",
  message: "User not found",
});

throw new TRPCError({
  code: "INTERNAL_SERVER_ERROR",
  message: "Failed to delete",
  cause: error, // Preserves original stack trace
});

// Client: typed error handling
const trpc = useTRPC();
const deletePost = useMutation({
  ...trpc.post.delete.mutationOptions(),
  onError: (error) => {
    switch (error.data?.code) {
      case "NOT_FOUND":
        toast.error("Not found");
        break;
      case "FORBIDDEN":
        toast.error("Not allowed");
        break;
    }
  },
});

See reference.md for complete error code table with HTTP status mappings.

Pattern 7: Optimistic Updates

Cancel queries, snapshot state, optimistically update, rollback on error, invalidate on settle.

const trpc = useTRPC();
const queryClient = useQueryClient();

const toggleTodo = useMutation({
  ...trpc.todo.toggle.mutationOptions(),
  onMutate: async ({ id }) => {
    await queryClient.cancelQueries({ queryKey: trpc.todo.list.queryKey() });
    const previousTodos = queryClient.getQueryData(trpc.todo.list.queryKey());
    queryClient.setQueryData(trpc.todo.list.queryKey(), (old: any) =>
      old?.map((t: any) =>
        t.id === id ? { ...t, completed: !t.completed } : t,
      ),
    );
    return { previousTodos };
  },
  onError: (err, vars, context) => {
    if (context?.previousTodos)
      queryClient.setQueryData(
        trpc.todo.list.queryKey(),
        context.previousTodos,
      );
  },
  onSettled: () =>
    queryClient.invalidateQueries({ queryKey: trpc.todo.list.queryKey() }),
});

Why good: Immediate UI feedback, automatic rollback on failure, eventual consistency via invalidation

See examples/optimistic-updates.md for complete pattern with like button example.

<red_flags>

RED FLAGS

High Priority Issues:

• Missing export type AppRouter -- clients have no type inference, defeats purpose of tRPC
• Raw throw new Error() -- should use TRPCError with appropriate code for HTTP mapping
• Procedures without .input() validation -- no runtime validation, type is unknown
• Auth checks in procedure body -- should use middleware for protected procedures
• Transformer at client level in v11 -- must be inside httpBatchLink(), not at createTRPCClient() level

Medium Priority Issues:

• Missing SuperJSON transformer -- Date/Map/Set won't serialize correctly
• No error formatter -- Zod errors should be formatted for better client DX
• Optimistic updates without rollback -- must include onError handler to restore previous state
• Using observable() for subscriptions -- v11 uses async generators; observable() is the v10 pattern
• Using rawInput in middleware -- v11 changed to getRawInput() function

Gotchas & Edge Cases:

• httpBatchLink combines requests -- all batched requests share the same HTTP status code
• SuperJSON transformer must be configured on BOTH client and server
• Context is created per-request -- don't store mutable state in context
• Middleware runs in order -- auth middleware should come before rate limiting
• Query keys are auto-generated -- use queryKey() method (v11) or getQueryKey() for manual access
• Subscription reconnection with tracked() requires lastEventId in input schema
• Don't retry mutations (retry: false) -- retrying writes can cause duplicates

</red_flags>

<critical_reminders>

CRITICAL REMINDERS

(You MUST export AppRouter type from your tRPC router for client-side type inference)

(You MUST use TRPCError with appropriate error codes -- never throw raw Error objects)

(You MUST use Zod for input validation on ALL procedures accepting user input)

(You MUST place transformer inside httpBatchLink() in v11 -- NOT at client level)

Failure to follow these rules will break type safety, cause runtime errors, and defeat the purpose of using tRPC.

</critical_reminders>