y is a simple web framework for TypeScript. It includes a validation library similar to Zod, and a simple route system similar to express. y is very opinionated and not suitable for all use cases, e.g. since it doesn't provide good support for custom middleware or data formats other than JSON.
Instead, it makes it easy to build well-documented software: y supports automatic generation of OpenAPI documentation for all endpoints. This includes generating JSON schemas for all request and response bodies that are specified using y schemas.
First, install y with your favorite package manager:
npm install @wemakefuture/y
yarn add @wemakefuture/y
bun add @wemakefuture/y
Then, create your first endpoint in src/routes/up.ts
:
import { y } from "@wemakefuture/y";
export default y.endpoint("/up", {
summary: "Get server status.",
method: "GET",
query: y.object({}),
headers: y.object({}),
req: y.object({}),
res: y.object({ message: y.string() }),
do(req) {
return {
body: {
message: "Healthy.",
},
};
},
});
And add this to src/index.ts
:
import { y } from "@wemakefuture/y";
y.app(`${__dirname}/routes`).then((app) => {
app.listen(3000);
});
This starts a server with all endpoints in src/routes
and listens on
port 3000.
The core construct of y is the endpoint
. Endpoints are created like this:
import { y } from "@wemakefuture/y";
export default y.endpoint("/up", {
summary: "Get server status.",
description:
"Returns a 200 status code if the server is running, otherwise a 502.",
method: "GET",
query: y.object({}),
headers: y.object({}),
req: y.object({}),
res: y.object({
message: y.string(),
}),
do(req) {
if (isHealthy()) {
return {
status: 200,
body: {
message: "Healthy.",
},
};
}
return {
status: 502,
body: {
message: "Not healthy.",
},
};
},
});
This is a lot, so let's break it down:
-
/up
is the HTTP endpoint path. -
summary
is a simple summary of the endpoint. This will appear as the title in the OpenAPI documentation. -
description
is an optional longer description that will appear after expanding the OpenAPI documentation for this endpoint. -
method
is the HTTP method used to access the endpoint. Options areGET
,POST
,PUT
, andDELETE
. -
query
is the schema for the query parameters. -
headers
is the schema for the HTTP headers. -
req
is the schema for the request body. This should be empty forGET
andPOST
. -
res
is the schema for the response body. -
do
is the function that will actually be executed. It can be either sync or async, and must return a response body, and optionally status codes and headers.
Schemas are described in more detail here. In more complex endpoints, error handling is also an important topic.
Inside the do
function, the request data can be accessed using the req
parameter.
-
req.body
is the validated request body, matching the schema specified for the endpoint. -
req.headers
is just likereq.query
, but for headers. -
req.http
is a request-specific HTTP client, which is described in detail here. -
req.log
is a request-specific logger. Currently, the logger isn't used for anything and just prints everything to the console, but that will change in the future. -
req.params
are the parameters extracted from the path. Unfortunately, they are not type-checked statically. -
req.query
is the typed object containing all query parameters. If the query parameters do not match the schema specified for the endpoint, an error is raised. -
req.url
is the HTTP path, so it does not include the hostname, and starts with/
.
Endpoints are always part of an entire application. With y, you create an application like this:
const app = await y.app(`${__dirname}/routes`);
This will automatically find all endpoints in the provided directory. Right now,
all .ts
and .js
files are loaded, except for any .test.ts
, .schema.ts
,
.util.ts
, .d.ts
, .test.js
, .schema.js
, or .util.js
files. Only the
default export is considered as an endpoint, so every endpoint will need to be
put into its own file. The routes
path should be absolute. There is currently
no way to add endpoints manually to an application.
Once the application is created, it can be started like this:
const context = app.listen(3000);
This starts an HTTP server and listens on port 3000. The context
that is
returned can be used for stopping the server again:
await context.stop();
This stops the server, but still continues to serve all current connections. Once every connection is closed, the method returns.
Apps can also be used to generate OpenAPI documentation:
const docs = app.docs({
info: {
title: "My title.",
description: "My description.",
version: "1.0.0",
},
servers: [
{
url: "prod.example.com",
description: "The production server.",
},
],
});
console.log(docs);
To make y mostly typesafe, query parameters, headers, and the request and response bodies are specified using schemas. If you have worked with Zod, schemas should be very familiar.
There is a large number of functions that create a schema. Some of them take
subschemas, others can be configured using methods called on the schemas. In all
cases, the schema itself is immutable, and modifying functions like
.optional()
always create a new schema instead of modifying the existing one.
The schemas should be very self-explanatory in general, and contain
documentation comments. The best way to learn is to try them! Here is a list of
all schemas:
- y.array
- y.boolean
- y.date
- y.enum
- y.intersection
- y.number
- y.object
- y.record
- y.string
- y.undefined
- y.union
- y.unknown
These are the same names as the schemas in Zod. Some schemas, like z.any
in
Zod, won't be added to y, since using any leads to lots of type unsafety.
Instead, use y.unknown
and use explicit type-checking.
There is also the more general y.BodyType
. All of the above schemas are
subclasses of y.BodyType
, but there are also:
- y.raw, which accepts any raw buffer with the specified content type
- y.either, which works just like y.union, except it may also contain a y.raw
Schemas can be passed to endpoints for validation, but you can
use them to validate things yourself, too. You can do that with the y.parse
function. It takes an unknown value and returns a typed and parsed value, or
throws a y.ValiationError
.
If you need to use the type of a y.BodyType
, you can do this using
y.Typeof<typeof bodyType>
. In this case, typeof bodyType
is the TypeScript
type of the schema itself, and y.Typeof
turns that into the parsed type.
It is generally possible to return any status code from an endpoint, including
status codes that indicate failure. However, it is often simpler to just throw
an error, especially in nested method calls. To reduce boilerplate code
associated with catching these errors, y automatically handles errors derived
from y.HttpError
, and returns their message and status code as an HTTP
response. There are some predefined error classes:
y.BadRequestError
y.UnauthorizedError
y.PaymentRequiredError
y.ForbiddenError
y.NotFoundError
y.ConflictError
If any other error is thrown inside an endpoint and not caught, y will
automatically return a 500 response, with the message Internal Server Error
.
This is supposed to prevent accidental leakage of sensitive information.
y contains a small embedded HTTP client that directly interfaces with the
logger. It is similar to Axios, but based on fetch
. You can access the HTTP
client using req.http
, the methods are fairly self-explanatory.
HTTP paths are used for specifying how the endpoint can be reached. They always
start with /
, and can contain multiple segments, each separated using /
from
the others. Segments can only contain lower-case letters, digits and hyphens.
Segments can be optional, in which case they are followed by ?
. Optional
segments must be at the end of the path, i.e. they may not be followed by
non-optional segments. Segments can also be parameters, in which case they are
preceded by :
. This means they can match any segment.
All of these rules are checked automatically once creating the path. If you want
to check a path programmatically, you can use y.validatePath
, which checks the
path and throws an exception if it is invalid.
The path parameters can be accessed inside endpoints using
req.params
.
y generally tries to follow a semantic versioning model. Right now, y is pre-1.0, so breaking changes can occur on every minor release.
- 0.8.0 - Removed
identity
encoding, changed error field toerrorMessage
- 0.7.3 - Changed
accept-encoding
fory.Http
toidentity
to prevent memory leak - 0.7.2 - Added time and connection count to connection log
- 0.7.1 - Made headers on test service methods optional
- 0.7.0 - Added test service and env parser
- 0.6.7 - Fixed failed 0.6.6 release
- 0.6.6 - Made
request
method ofy.Http
public - 0.6.5 - Fixed crash when validation error occurs
- 0.6.4 - Added changelog and custom category option for endpoints
- 0.6.3 - Updated documentation and fixed NodeJS compatibility
- 0.6.2 - Fixed auto-importing of utility files
- 0.6.1 - Fixed auto-import error
- 0.6.0 - Replaced
y.router
withy.app
, removed CLI completely - 0.5.0 - Removed need for
y fix
- 0.4.1 - Updated
y init
command - 0.4.0 - Added listen context to correctly terminate server
- 0.3.11 - Added basic way to stop server after
y.listen
- 0.3.10 - Improved OpenAPI generation
- 0.3.9 - Fixed parsing of empty JSON body
- 0.3.8 - Fixed incorrect NPM version
- 0.3.7 - Added
y doc
command - 0.3.6 - Added
req.http
client - 0.3.5 - Added log on each request
- 0.3.4 - Fixed endpoint result schema
- 0.3.3 - Fixed
y.either
export status - 0.3.2 - Added
y.either
- 0.3.1 - Fixed request body parsing
- 0.3.0 - Added
y.raw
- 0.2.1 - Fix for
y fix
command - 0.2.0 - Added NodeJS support and CLI interface
- 0.1.2 - Fixed y.number() minimum and maximum checks
- 0.1.1 - Added test cases, documentation and license
- 0.1.0 - Initial release