Awesome
Preface
I said I want SIMPLE runtypes. Just functions that validate and return data. Combine them into complex types and TypeScript knows their structure. That's how runtypes work.
<!-- toc --> <!-- tocstop -->Install
# npm
npm install simple-runtypes
# yarn
yarn add simple-runtypes
Example
- Define the Runtype:
import * as st from 'simple-runtypes'
const userRuntype = st.record({
id: st.integer(),
name: st.string(),
email: st.optional(st.string()),
})
now, ReturnType<typeof userRuntype>
is equivalent to
interface {
id: number
name: string
email?: string
}
- Use the runtype to validate untrusted data
userRuntype({ id: 1, name: 'matt' })
// => {id: 1, name: 'matt'}
userRuntype({ id: 1, name: 'matt', isAdmin: true })
// throws an st.RuntypeError: "invalid field 'isAdmin' in data"
Invoke a runtype with use
to get a plain value back instead of throwing errors:
st.use(userRuntype, { id: 1, name: 'matt' })
// => {ok: true, result: {id: 1, name: 'matt'}}
st.use(userRuntype, { id: 1, name: 'matt', isAdmin: true })
// => {ok: false, error: FAIL}
st.getFormattedError(FAIL)
// => 'invalid keys in record: ["isAdmin"] at `<value>` in `{"id":1,"name": "matt", ... }`'
Not throwing errors is way more efficient and less obscure.
Throwing errors and catching them outside is more convenient:
try {
... // code that uses runtypes
} catch (e) {
if (st.isRuntypeError(e)) {
console.error(getFormattedError(e))
return
}
throw e
}
Why?
Why should I use this over the plethora of other runtype validation libraries available?
- Strict: by default safe against
__proto__
injection attacks and unwanted properties - Fast: check the benchmark
- Friendly: no use of
eval
, and a small footprint with no dependencies - Flexible: optionally modify the data while it's being checked - trim strings, convert numbers, parse dates
Benchmarks
@moltar has done a great job comparing existing runtime type-checking libraries in moltar/typescript-runtime-type-benchmarks.
@pongo has benchmarked simple-runtypes
against io-ts
in pongo/benchmark-simple-runtypes.
Documentation
Intro
A Runtype
is a function that:
- receives an unknown value
- returns that value or a copy if all validations pass
- throws a
RuntypeError
when validation fails or returnsValidationResult
when passed touse
interface Runtype<T> {
(v: unknown) => T
}
Runtypes are constructed by calling factory functions.
For instance, string
creates and returns a string runtype.
Check the factory functions documentation for more details.
Usage Examples
Nesting
Collection runtypes such as record
, array
, and tuple
take runtypes as their parameters:
const nestedRuntype = st.record({
name: st.string(),
items: st.array(st.record({ id: st.integer, label: st.string() })),
})
nestedRuntype({
name: 'foo',
items: [{ id: 3, label: 'bar' }],
}) // => returns the same data
Strict Property Checks
When using record
, any properties which are not defined in the runtype will cause the runtype to fail:
const strict = st.record({ name: st.string() })
strict({ name: 'foo', other: 123 })
// => RuntypeError: Unknown attribute 'other'
Using record
will keep you safe from any __proto__
injection or overriding attempts.
Ignore Individual Properties
To ignore individual properties, use ignore
, unknown
or any
:
const strict = st.record({ name: st.string(), other: st.ignore() })
strict({ name: 'foo', other: 123 })
// => {name: foo, other: undefined}
Optional Properties
Use the optional
runtype to create optional properties:
const strict = st.record({
color: st.optional(st.string()),
width: st.optional(st.number()),
})
Non-strict Property Checks
Use nonStrict
to only validate known properties and remove everything else:
const nonStrictRecord = st.nonStrict(st.record({ name: st.string() }))
nonStrictRecord({ name: 'foo', other: 123, bar: [] })
// => {name: foo}
Discriminating Unions
simple-runtypes
supports Discriminating Unions via the union
runtype.
The example found in the TypeScript Handbook translated to simple-runtypes
:
const networkLoadingState = st.record({
state: st.literal('loading'),
})
const networkFailedState = st.record({
state: st.literal('failed'),
code: st.number(),
})
const networkSuccessState = st.record({
state: st.literal('success'),
response: st.record({
title: st.string(),
duration: st.number(),
summary: st.string(),
}),
})
const networdStateRuntype = st.union(
networkLoadingState,
networkFailedState,
networkSuccessState,
)
type NetworkState = ReturnType<typeof networkStateRuntype>
Finding the runtype to validate a specific discriminating union with is done efficiently with a Map
.
Custom Runtypes
Write your own runtypes as plain functions, e.g. if you want to turn a string into a BigInt
:
const bigIntStringRuntype = st.string({ match: /^-?[0-9]+n$/ })
const bigIntRuntype = st.runtype((v) => {
const stringCheck = st.use(bigIntStringRuntype, v)
if (!stringCheck.ok) {
return stringCheck.error
}
return BigInt(stringCheck.result.slice(0, -1))
})
bigIntRuntype('123n') // => 123n
bigIntRuntype('2.2') // => error: "expected string to match ..."
Reference
Basic runtypes that match JavaScript/TypeScript types:
Meta runtypes:
Objects and Array Runtypes:
Combinators:
Shortcuts:
Roadmap / Todos
size
- a meta-runtype that imposes a size limit on types, maybe via convert-to-json and .length on the value passed to it- rename
stringLiteralUnion
toliterals
orliteralUnion
and make it work on all types thatliteral
accepts - rename record to object: #69
- improve docs:
- preface: what is a runtype and why is it useful
- why: explain or link to example that shows "strict by default"
- show that
simple-runtypes
is feature complete because it can- express all TypeScript types
- is extendable with custom runtypes (add documentation)
- add small frontend and backend example projects that show how to use
simple-runtypes
in production
- test all types with tsd
- add more combinators: partial, required, get, ...
- separate
Runtype
andInternalRuntype
and type runtype internals (see this comment)
current tasks (metadata) notes
- check that intersection & union tests do properly test the distribution stuff
- make getMetadata public
- maybe make metadata typed and include all options so that you can walk the tree to create testdata orjson-schemas from types
- maybe add a
serialize
function to each runtype too? to use instead of JSON.stringify and to provide a full-service library? - maybe make
any
a forbidden type of a runtype