bogeeee / restfuncs

MIT License
49 stars 1 forks source link

Restfuncs - HTTP API done proper

What is Restfuncs: Coming from tRPC Restfuncs is also an RPC (Remote Procedure Call) library. It also has a client and sever part and also gives the user end2end type safety. The key differences are: - Restfuncs takes [tRPCs claim: "It's just functions"](https://trpc.io/docs/concepts#its-just-functions) for real also for the server side 😎, resulting in a much simpler usage and way less boilerplate. - Restfuncs uses (native) typescript for validation, instead of ZOD or other type generators. - It has websocket support as first class citizen, enabled by default. Therefore push events can be achieved just via plain callback functions πŸ‘ which you can pass anywhere in the arguments. - Cookie sessions / JWT, CSRF protections and CORS are tightly integrated into Restfuncs instead of beeing 3rd party plugins or having to be coded manually. _This was especially necessary for supporting / syncing with websockets._ - Concepts are simplified, aiming for less total boilerplate and a more shallow learning curve. - Also see the additional features in the [list](#intro--features). Here's a mapping of the tRPC's conceptual items to Restfuncs: - `Procedure`/`Query`/`Mutation` -> No distinctions between them in restfuncs. It's all just @remote methods. Following GET/POST semantics is done by a client (if needed) and Restfuncs [serves just both styles](#rest-interface) instead of needing configuration. - `Router` -> `ServerSession` - `Context` -> `ServerSession` - _you can also store fields there._ - Middlewares -> You just overwrite the ServerSession#doCall method for this (use ctrl+space / intellisense in your IDE). Needs no further explanation or docs, if you know some basic OOP concepts. - `Subscriptions` -> You can simply use callback functions anywhere in your @remote methods. _When called, the events automatically get send to the client via websocket push events. No need to set up that channel or synchronize that context manually πŸ‘_. - Inferring Types -> Not needed. Can just be achieved by Typescript.
What is Restfuncs: Coming from Nest or other "big frameworks" Restfuncs is not a framework for organizing your code (and does not want to be such). It is just a small layer above express, to improve communication needs in a one-tool-for-one-purpose manner. Similiar to those frameworks, it makes coding API endpoints easier and offers you a rich set of features around those (but, as said, just what tightly belongs to http communication. Nothing else). Also, Restfuncs it is made for RPC (Remote Procedure Calls): That means, on the client, you don't code a fetch request by hand, but just call a remote method, as if it was a normal javascript method (but across the wire). It's an old concept in the IT that was a bit forgotten but regains traction again in the JS world, cause it makes especially sense in a scenario with typescript, both, on the client and the server side. Therefore, all your calls can be checked for type safety at compile time (end2end type safety). A similar popular library that offers such a concept is [tRPC](https://trpc.io/). Think of Restfuncs as a more modern alternative of that. A Nest' "Controller" (=Service) corresponds to a "ServerSession" class in Restfuncs. That's the only organization unit, it has. Wiring them to the express routes is done manually. Now you put a few methods (=endpoints) and fields (=session cookie fields) into such a ServerSession class and that's already all the concept ;) There's really nothing more conceptually. Just a lot of configuration options around that.
What is Restfuncs: Coming from Express or new to Javascript Coding Express handlers (let's call them "endpoints" now) is fun, but soon you will notice, that this is pretty low level and you do all that repetitive tasks over and over again, like: - Converting from request/body parameters into local variables - Checking these variables against evil input - Doing error handling and reporting The same thing on the client where, each time, you code a fetch request by hand and do conversion, status checking/error handling,... Now, instead of a request handler, in Restfuncs you code your endpoint as a plain javascript function (with typescript types). And it also offers a client, where you can just virtually call that same function from the client (but it goes over the wire via http or websockets). This It's an old concept in the IT that's called RPC (Remote Procedure Call) and it was a bit forgotten but regains traction again in the JS world, cause it makes especially sense in a scenario with typescript, both, on the client and the server side. Therefore, all your calls can be checked for type safety at compile time (end2end type safety). A similar popular library that offers such a concept is [tRPC](https://trpc.io/). Think of Restfuncs as a more modern alternative of that. But besides RPC, Restfuncs deals with much more aspects around http-communication, that play together and make just sense to be tightly integrated here into this communication library. But see the features for yourself.

Intro + features

With restfuncs, you write your http API endpoints just as plain typescript functions. Or better say: methods.
Per-endpoint boilerplate is no more than, i.e.:

@remote greet(name: string) {
    return `Hello ${name}` 
}

See, it uses natural parameters and natural return (and throw) flow, instead of you having to deal with req and res on a lower level. And Restfuncs will take care about a lot more of your daily, low-level communication aspects.
That is (features):

Smaller features:

Negative features (downside):

Getting started

⚠⚠⚠ Security must-read - When using **client certificates**, you must also read the [CSRF protection chapter](#csrf-protection). - When making **raw calls** via the [REST interface](#rest-interface), see security note there. _Apart from that, **all security**, which this communication layer (=this Resfuncs library) is responsible for, **is already enabled by default πŸ‘**. I.e., CRSF protection, CORS, param validation, concealing error stacks,... There will be explicit hints if you need to take further action._

Here's how to set up a server that serves a remote method, named myRemoteMethod, and a client that makes a call to that method:

MyServerSession.ts

import {ServerSession, ServerSessionOptions, remote, UploadFile, ClientCallback, ClientCallbackSet, ClientCallbackSetPerItem, free, withTrim} from "restfuncs-server";
import { tags } from "typia";

export class MyServerSession extends ServerSession {

  static options: ServerSessionOptions = {/* ServerSessionOptions */}

  /**
   * This JSDoc also gets outputted in the public API browser and OpenAPI spec. Write only nice things here ;)
   * @param myUserName Some string param.
   * @param someOptionalComplexParam Your parameters can be of any complex typescript type. They are automatically security-validated at runtime to fit into that schema.
   * Also you can use [Typia's special type tags](https://typia.io/docs/validators/tags/#type-tags) like `string & MaxLength<255>`
   */
  @remote({/* RemoteMethodOptions */})
  myRemoteMethod(myUserName: string, someOptionalComplexParam?: { id?: number, city: string & tags.MaxLength<255> }) {
    return `Hello ${myUserName}`
  }
}


server.ts

import {restfuncsExpress} from "restfuncs-server";
import {MyServerSession} from "./MyServerSession.js"; // For node, you must use use ".js" instead of ".ts" in your imports (yes, strange) !

const app = restfuncsExpress({/* ServerOptions */}) // Drop in replacement for express (enhances the original). It installs a jwt session cookie middleware and the websockets listener. Recommended.
app.use("/myAPI", MyServerSession.createExpressHandler())

// Optional: app.use(helmet(), express.static('dist/web')) // Serve pre-built web pages / i.e. by a bundler like vite, parcel or turbopack. See examples. It's recommended to use the helmet() middleware for additional protection.
// Optional: app.use(...) //<-- Serve *other / 3rd party* express routes here. SECURITY: These are not covered by restfuncs CSRF protection. Don't do write/state-changing operations in here ! Instead do them by MyServerSession.

app.listen(3000); // Listen on Port 3000


client.ts
Use a bundler like vite, parcel or turbopack to deliver this file to the browser

import {RestfuncsClient} from "restfuncs-client";
import {MyServerSession} from "../path/to/server/code/or/its/packagename/MyServerSession.js" // Import the type for full **end2end type support**. A good bundler like vite will see, that in the code below, only types are used (via generic parameter `RestfuncsClient<MyServerSession>`) but it looks for the bundler like no actual server code is called, so it will not try to follow and include server-side code πŸ‘. Note: Despite some rumors, you don't need a monorepo for this cross-referencing. Just packages next to each other. Or client + server files can live even in the same package. See the examples which show both options.

const myRemoteSession = new RestfuncsClient<MyServerSession>("/myAPI", {/* RestfuncsClientOptions */}).proxy;
console.log( await myRemoteSession.myRemoteMethod("Hans") ); // Call your remote method over the wire 😎😎😎

Setting up the build (here, it gets a bit nasty 😈)

tsconfig.json

"compilerOptions": {
    "moduleResolution": "node",
    "experimentalDecorators": true,
    "strictNullChecks": true,
    "sourceMap": true, //optional, recommended
    "plugins": [
        { "transform": "restfuncs-transformer",  "transformProgram": true},
        { "transform": "typia/lib/transform" },
        { "transform": "typescript-rtti/dist/transformer" } ],
},
"exclude": ["dist", "client", "web"], // Make sure, to not accidentially transform your client files.

package.json

"scripts": {
    "dev": "cross-env NODE_ENV=development <use your favourite tsx / bun / jest / vitest / ... to run stuff>"
    "clean": "tspc --build --clean",
    "build": "tspc --build --force",
    "start": "cross-env NODE_ENV=production node --enable-source-maps server.js"
},
"dependencies": {
  "restfuncs-server": "^3.0.0",
  "restfuncs-client": "^2.0.0"
},
"devDependencies": {
  "ts-patch": "^3.0.2",
  "restfuncs-transformer": "^1.0.0",
  "cross-env": "^7.0.3"
},

The build script compiles for production with tspc (instead of tsc) from the ts-patch package, which allows for our 3 transformer plugins in tsconfig.json (No worries: Despite the name "ts-patch", it runs in "live mode" so nothing will be patched here). See, how the transformer chain works.

The dev script sets NODE_ENV=development to tell Restfuncs that it can ignore all security validations and therefore there's no need for all that transformer chain. So you're fast and unrestricted during developmentπŸ‘. But be warned: You should still check very often, that the production run (build + start) is still working !!! Otherwise, things can break unnoticed, i.e., if you import ".ts" instead of ".js", if there are some tsc-only compile errors, or if the bundler's build tries to follow into server-side symbols. Especially for the last one, it can be hard to track down the reason, if you face a lot of code changes at once. So check often !!!

Congrats, you've got the concept!
Now use your IDE's intellisense and have a quick browse through the /* XxxOptions */. That JSDoc is considered the official documentation and it won't be repeated here. In some cases where more configuration needs to be decided for, contextual error messages will guide you. So don't be scared of them and read them and see them as part of the concept.







Example projects

They use vite, which is a very minimalistic/ (zero conf) web bundler with full support for React/JSX, Typescript, hot module reloading. Hope you'll like this as a starter stack for your webapp.

Store values in the browser-session:

With Restfuncs, session values are stored in a typesafe way as fields in your ServerSession class.

See the following example:

import {ServerSession} from "restfuncs-server";

export class MyServerSession extends ServerSession {

  myLogonUserId?: string // This value gets stored in the session under the key "myLogonUserId".

  @remote whoIsLoggedIn() {
    const user = getUser(this.myLogonUserId); // Read a session field
  }
}

When you modify the fields and after successful completion of the remote method, the fields get serialized into the session and a http cookie is sent accordingly and also the changes are updated to all your (web-)socket connections :). Concurrency behaviour with (web-)socket connections is exactly like with http calls: There won't be a swapping of session values (from another call) right in the middle of a call but before the next call. An exception is: ServerSessionOptions#inMemorySessions.

Note, that Restfuncs will enforce some rules:

File uploads

COMING SOON!, but the docs are already here ;)

There's no big extra way you have to go for file uploads. You just pass a browsers File object anywhere in your remote method's parameters to the server. On the server side, you can then grab that and just suck on the stream and restfuncs will automatically send that file in an extra http request in the background/automatically. There's just one thing. Because the browser API's File class is not type compatible with the server API, you have to cast it to the UploadFile type. See example/client.ts. Example:

import {ServerSession, ServerSessionOptions, UploadFile, remote} from "restfuncs-server";

export class MyServerSession extends ServerSession {

  /**
   * @param someotherField
   * @param myUploadFile You can pass UploadFile objects anywhere/deeply and also as ...rest arguments. As soon as you read from the the stream, the restfuncs client will send that file in an extra http request in the background/automatically.
   */
  @remote myRemoteMethodWithUploadFile(someotherField: string, myUploadFile: UploadFile) {
    // TODO
    return "Your file was uploaded"
  }
}

client.ts

import {UploadFile} from "restfuncs-common";
import {RestfuncsClient} from "restfuncs-client";
import {MyServerSession} from "../path/to/server/code/or/its/packagename/MyServerSession.js" // Gives us the full end2end type support

const myRemoteSession = new RestfuncsClient<MyServerSession>("/myAPI", {/* RestfuncsClientOptions */}).proxy; // Create a client, as usual
const myBrowserFile = document.querySelector("#myFileInput").files[0]; // Retrieve your File object(s) from an <input type="file" /> (here), or from a DragEvent.dataTransfer.files
await myRemoteSession.myRemoteMethodWithUploadFile("someContext", myBrowserFile as UploadFile) // And send myBrowserFile to the server. Note: You must cast myBrowserFile to the `UploadFile` type.

You can also call myRemoteMethodWithUploadFile via REST interface

Server events via callback functions

Tl;dr: Have you tried, sending a callback function to the server? This works😎😎😎...imagine the opportunities! If you store references for longer, mind cleaning them up on client disconnect. The ClientCallbackSet and ClientCallbackSetPerItem util classes will help you with that. All args + results are safely type checked at runtime πŸ›‘πŸ›‘πŸ›‘.

Now to the content:

You want your client to react to a certain event that happens on the server ? Restfuncs solves this very elegantly: You can have functions in the remote method parameters. Even nested. Just think of callback functions in your daily javascript life (but not about callback-hell from the old days before async/await/Promise was introduced πŸ˜†πŸ˜†πŸ˜†). You call them on the server (any time and as often as you want) and they get executed on the client ;) All via (web-)socket push events of course. Example:

// On the server:
@remote notifyMeWhenSomeoneEntersTheChatroom(chatroom: string, onUserEntersChat (user: User) => void) {
    // ...later, somewhere in your code:
    onUserEntersChat(someUser) // Call the callback
}

// on the client:
myRemoteSession.notifyMeWhenSomeoneEntersTheChatroom("#restfuncs_is_great", (user) => { console.log(`${user.name} entered the chatroom`) })

Callbacks can also return some result (via Promise) and the server can await it. The callack's arguments and the result are validated at runtime, just like with a normal remote method. Except the automatic trimming is not enabled by default. In the RemoteMethodOptions you'll find some options for changing the default behaviour.

Worry free instance remembering

It's propably not worth mentioning, but: Same function instances on the client result in same function instances on the server. The server remembers them. This allows for worry free use of addEventListener(somefunction) + removeEventListener(somefunction) style code

Trim off extra properties

This is not done by default. If you want to trim off extra properties, similar to RemoteMethodOptions#trimArguments, you have to use the withTrim function:

import {withTrim} from "restfuncs-server";
// ...later, somewhere in your code:
withTrim(onUserEntersChat)(someUser) // Call the callback

for more info, see the JSDoc of withTrim.

Limitations

Callbacks can only be declared 'inline' in the line of your remote method's declaration. Like: @remote myRemoteMethod(...the arrow(s) => must be somewhere inside here...). Deeply nested in some (inline) object structure is also fine. But not in an external type and then refer to it. This is for the restfuncs-transformer to be able to scan them and generate the security validation code. At that transformation stage, it only sees the syntax tree and can't follow into types.

Prevent resource exhaustion on the server

In the world wide web. Clients may not always be so friendly to call removeEventListener but instead just disconnect. To help you, clean those up to now grow your precious memory, you can either listen for disconnect events via:

import {ClientCallback} from "restfuncs-server";

(myCallbackFn as ClientCallback).socketConnection.onClose(() => {
    // <- unregister myCallbackFn here
})

...or use the utility classes ClientCallbackSet and ClientCallbackSetPerItem (import {...} from "restfuncs-server"). You can add listeners to them and they get removed automatically on disconnect. See JSDoc there.

Advanced: Resource cleanup on the client

Just in case you have a very heavy and rich client, you may at some point wonder, how and when the references to the callback functions are cleaned up:

Advanced

Access the raw request/response/websocket fields

There's a virtual runtime field named call where you can access your runtime context. Use intellisense for docs. Example:

@remote myRemoteMethod() {
  const req = this.call.req; // Express's req object
  const res = this.call.res; // Express's res object 
  const conn = this.call.socketConnection; // Restfuncs's connection. Wraps the engine.io socket.

  res!.header("Content-Type", "text/plain")

  return `your ip address is: ${req!.ip}`
}

Intercept calls

On the server

It's possible to intercept each call to i.e. check for auth (see example project) or handle errors. In your ServerSession class, override the doCall method Use your IDE's intellisense (ctrl+space) to override it. It would look like this:

export class MyServerSession extends ServerSession {

  protected async doCall(funcName: string, args: any[]) {
    try {
      // ... Intercept before
      const result = await super.doCall(funcName, args);
      // ... Intercept after
      return result;
    } catch (e) { // Intercept on error (if really needed). Also see the #error-handling chapter
      throw e; // When doing so, it is adviced to throw an error again. Restfuncs will do the proper reporting to the client.  
    } finally {
      //... Intercept finally
    }
  }
}

On the client

Same same like on the server: You override the RestfuncsClient#doCall method. To get to the goal, you must subclass the RestfuncsClient. Example:

//Create your own subclass of RestfuncsClient...
class MyRestfuncsClient<S> extends RestfuncsClient<S> {

  async doCall(funcName: string, args: any[]) {
    try {
      // ... Intercept before
      let result = await super.doCall(funcName, args);
      // ... Intercept after
      return result;
    } catch (e) { // Intercept on error
      throw e;
    } finally {
      //... Intercept finally
    }
  }
}

const myRemoteSession = new MyRestfuncsClient<MyServerSession>(...).proxy // ... and use your subclass in place of RestfuncsClient

Working example

Hello world Web app with authentication Open in StackBlitz

This example intercepts calls on the server and on the client.

Html / images / binary as a result

To serve a non API result, the remote method must explicitly set the content type. Return the result via string, Buffer or Readable. Example:

    @remote({isSafe: true /* Lessen restrictions and allow this method to be called by GET ... */}) 
    getAvatarImage(name: string) {
        // ... therefore (SECURITY) code in `isSafe` methods must perform read operations only !
        this.call.res!.contentType("image/x-png")
        return fs.createReadStream("/someImage.png") // Returns a Readable which is streamed to client. You can also return Buffer, String, File(TODO)
    }

⚠⚠⚠ Security note: When serving html with rich content or with scripts, you might want to add the helmet middleware in front of your ServerSession for additional protection via app.use("/myAPI", helmet(), MyServerSession.createExpressHandler())

REST interface

⚠⚠⚠ Security must-read For handcrafted calls from inside a **browser**, **you** (the clients) need to care about [protecting **your** session from CSRF attacks from other browser windows](#csrf-protection).

Tl;dr: Just form the http call from your imagination, and its likely a way that works, or Restfuncs will tell you exactly, what's wrong with the params.

Now to the content:
Like the name Restfuncs suggests, there's also a REST/http interface for the case that you don't use the neat RestfuncsClient, or you want to call these from non-js languages, etc.
Restfuncs follows a zero conf / gracefully accepting / non-strict approach (a client is still free to implement strictness to the REST paradigm):
The following example remote method...

    async getBook(name: string, authorFilter?: string) {

    }

...can be called in almost every imaginable way through http like:

Method Url Body Description
GET /getBook/1984/George%20Orwell List arguments in the path
GET /getBook?1984,George%20Orwell List arguments in the query
GET /getBook?name=1984&authorFilter=George%20Orwell Name arguments in the query
GET _/getBook?_<custom implementation> Override the parseQuery method in your ServerSession subclass. See JSDoc. Here's a discussion about different url serializers
GET /book ... Read "GET book" like getBook. Applies to other http verbs also. Additionally "PUT book" will try to call updateBook or setBook cause this sounds more common in programming languages.
POST /getBook {"name": "1984", "authorFilter":"George Orwell"} Name arguments inside JSON body
POST /getBook ["1984", "George Orwell"] List arguments inside JSON body
POST /getBook/1984 "George Orwell" Single JSON primitive
POST /getBook/1984 George Orwell Plain string. For this you must explicitly set the Content-Type header to text/plain
POST /getBook name=1984&authorFilter=George%20Orwell Classic Html <form> with Content-Type = application/x-www-form-urlencoded. Still remember these ? They can be used here as well ;)
POST /getBook/1984 <Any binary data> Binary Data. Your function parameter (i.e. here the 2nd one) must be of type Buffer.

You are free to mix these styles ;) The styles are parsed in the order as listed, so arguments from a lower line in the table will -override named- or -append to listed- ones from above.

Also it's possible to have Readable and Buffers as parameters ...

    async uploadAvatarImage(userName: string, image: Readable) {

    }

...can be called through http like:

Method Url Body Description
POST /uploadAvatarImage/Donald%20Duck <> Binary data directly in the body (TODO)

Content types

To specify what you send and how it should be interpreted, set the Content-Type header to

Auto value conversion

Parameter values will be reasonably auto converted to the actual declared type.

Restfuncs won't try to convert to ambiguous types like string|bool cause that would be too much magic and could cause unwanted behaviour flipping in your app (i.e., someone evil enters 'true' as username and this makes its way to a query param).

Note for the security cautious of you: After all this "wild" parameter collection and auto conversion, the actual call-ready parameters will be security-checked again in a second stage.

Receiving content (json-like result)

To specify what you want to receive in the response, Set the Accept header to

Using websockets

Tl;dr: The RestfuncsClient already uses Websockets by default. There's usually nothing to configure.

It is required that you use the restfuncsExpress() instead of express() server, like shown in the getting started chapter/server.ts. Make sure, the path /engine.io_restfuncs is reachable and not blocked by your proxy. See ServerOptions#engineIoOptions.path. Also see the RestfuncsClientOptions#useSocket and RestfuncsClientOptions#shareSocketConnections options.

Error handling

Tl;dr: usually no need to implement it, because Restfuncs already brings a comprehensive reporting to the client.

Besides that, there are options to tweak: See ServerSessionOptions#logErrors and ServerSessionOptions#exposeErrors.

Errors with http status code or special errors

If, for some specific occurrence, you need to specify the http status code or have the error always (in production) be sent to the client or have custom error properties which should (always) be sent to the client, then throw a new CommunicationError(...message..., {httpStatusCode: ...}) or a custom subclass of it. Here's the import: import {CommunicationError} from "restfuncs-server"

Security

CSRF protection

Tl;dr: In a normal situation (= no basic auth, no client-certs and using the RestfuncsClient), Restfuncs already has a strong CSRF protection by default (corsReadToken, enforced by the RestfuncsClient). For other situations, read the following:

Restfuncs has the following 3 protection levels (weakest to hardest) to protect against CSRF attacks. See list below. You can enforce it by the ServerSessionOptions#csrfProtectionMode setting.
By default/ undefined, the client can decide the protection mode. "wait a minute, how can this be secure ?" See explanation. This way, all sorts of clients can be served. Think of non-browser clients where CSRF does not have relevance, so their devs are not bugged with implementing token fetches.
Explanation: The clients indicate, which csrfProtection mode they want to "play" in a header proactively on every request. Restfuncs will raise an error, if another browser client (or i.e an attacker from another browser tab) wants to play a different mode, at the moment it tries to access the (same) session. Meaning, once the (cookie-) session is created, the protection mode is stored in there. Note: "proactively" means: no header = defaulting to preflight is still allowed, as long as it's consistent.

The above policy (let the clients decide) only covers sessions. So when using client-certificates or basic auth, you must explicitly decide for a setting, and you should use at least set it to corsReadToken when dealing with browser clients.

Here are the modes. ServerSessionOptions#csrfProtectionMode / RestfuncsClient#csrfProtectionMode can be set to:

Notes:

Hardening security for the paranoid

Performance

Writes to the session fields have some overhead

It costs an additional http roundtrip + 1 websocket roundtrip + (auto.) resend of unprocessed websocket calls. This is to ensure fail-safe commits to the http cookie and to ensure security. So keep that in mind.

Use your own cookie handler

The build-in cookie handler goes like this, as already mentioned in the set up: const app = restfuncsExpress({/* ServerOptions */}) // Drop in replacement for express (enhances the original). Alternatively (not recommended), you can still set up your own cookie handler, after using the classic express: app = express(), or by setting ServerOptions#installSessionHandler to false. Restfuncs will still synchronize the content to/from (Web-) Socket connections.

Multi server environment

When using a load balancer in front of your servers, you have to configure it for sticky sessions, because the underlying engine.io uses http long polling as a first, failsafe approach. You might try to also change that.

Tips & tricks

Using typescript to automatically trim the output into the desired form

By default, restfuncs trims off all extra properties in the result of your remote methods to match the exact declared typescript type. You can make use of this in combination with these two handy typescript utility types: [Pick<Type, Keys>](using Pick and Omit) and Omit<Type, Keys> Example:

type IUser=  {
  name: string,
  age: number,
  password: string,
}

@remote returnsPublicUser(): Pick<IUser, "name" | "age"> { // This will return the user without password
    const user = {name: "Franz", age: 45, password: "geheim!"} // got it from the db somewhere
    return user;
}

@remote returnsPublicUser(): Omit<IUser, "password">{  // Also this will return the user without password
   ...
}

or you could also create a new type and go with returnsSafeUser(): SanitizedUser {...}. Etc. etc. you've got all the world of typescript here ;)

Validate stuff on the inside

Now that you've gone all the long way of setting up the build, you have Typia at hand and can use it to validate your objects, i.e. before they get stored the db. Example:

import typia, { tags } from "typia"

type User = {
  name: string & tags.MaxLength<255>
}

if(process.env.NODE_ENV === 'production') { // cause in dev, you usually run without transformed code
  typia.assertEquals<User>(myUser) // Validate myUser before storing it in the db
}
db.store(myUser)

Also you can inspect all your types at runtime

Migration from 2.x

As the 2.x release was announced to be non production-ready, here is how to migrate to the production-ready 3.x version, where those issues were fixed

That's it !

Comparison to other RPC libraries

Comparison table

Contribution

See DEVELOPMENT.md

lPlaces where your help would be needed