This document maintains a list of major changes to Superstruct with each new release.
Some changes in Superstruct v2.0 are potentially breaking if you were using the library in unusual and/or undocumented ways. Since it has been almost 2 years since the last significant release, we want to make sure that we don't ruin someone's day by surprising them with a fix that changes buggy behavior they were relying on. For this reason, some changes that would ordinarily be considered a fix are marked as breaking.
🚀 For the absolute majority of users, this should be a smooth upgrade that will not require any changes.
- Validation now correctly fails when arrays are passed to
object(),type(), andrecord()structs. - When coercing an
object()(viamask(),create()orvalidate()with thecoerce: trueoption), arrays will no longer be automatically converted to objects with indexes as keys. See this PR comment.
- The library and its TypeScript typings are now compatible with NodeNext/Node16 module resolution. This means that if you are using the library with TypeScript and ECMAScript modules at the same time, you should no longer run into issues!
- Using
mask()withunion()now correctly masks union members instead of incorrectly failing validation.
- Use with Node.js v14 is now deprecated. Due to incompatible tooling, we are no longer able to test Superstruct on this version of Node.js. Use at your own risk.
Project maintenance has moved to a new volunteer team: @arturmuller and @yeoffrey. Hello there! 👋 We are currently going through the existing issues and PRs, trying to resolve or close the backlog. This might take a little while so please bear with us.
If you are interested in contributing — or helping us process the backlog — we would love your help. Don't hesitate to help us triage, open an issue, or submit a PR. You can also join our Superstruct maintainers Discord: https://discord.gg/pdHrQBjQ96.
If you have questions, suggestions, or are just not sure about something related to Superstruct, head over to GitHub Discussions! We have recently enabled this feature to help us differentiate between actual issues/bugs and everything else. We hope this will be a great new place where Superstruct users can get quick help from us — the maintainers — but also from the community as a whole.
See you at the next release! ✌️
Added an optional message argument to override error messages. You can now pass in a message argument to all of the error checking functions which will override any error message with your own message. If you do, Superstruct's original descriptive message will still be accessible via error.cause.
assert(data, User, 'The user is invalid!')
// StructError: The user is invalid!No breaking changes. This just marks Superstruct being stable for a long time, so better to move to a non 0.* numbering scheme.
Refinement functions are now called with valid, but potentially unrefined values. Previously the functions passed in to refine would always be called with sub-elements (eg. when using objects or arrays) that were completely valid and refined. However, this prevented collecting all the refinement errors from subelements in a validation in one go, which is common when validating forms. Note: this should not have any affect on almost all use cases, so you're safe to upgrade.
Unions can now be coerced. Previously unions created a barrier in coercion such that structs nested inside unions would not have their coercion logic triggered, but this has been fixed.
Assigning preserves type structs. Previously using the assign helper would implicitly convert type structs into object structs which wasn't expected and confusing, but this has been fixed.
The mask helper now works for nested objects. Previously it would only mask the properties at the top-level of a struct, however now it acts deeply. You can use it to define object structs once, but use them either strictly or loosely.
The masked coercion has been removed. This previously allowed you to mix in masking to a specific struct, but the mask helper is a more robust way to do this, and it doesn't force you to maintain two separate structs.
Structs can now define an entries iterator for nested values. Previously iterating through nested values was defined in a one-off manner inside certain structs, but this led to non-uniform support. Now, any struct can define an entries iterator that will cause nested values to be automatically coerced and validated.
Coercion receives context objects and supports nested values. Previously context objects were only passed to the validations and refinements. But now the same context is passed to coercions too so you can implement more complex logic. And coercions are automatically applied to nested values thanks to the addition of entries.
Iteration logic has gotten simpler, and more performant. The addition of the entries logic has enabled us to only ever iterate through a tree of values one time for coercion and validation, instead of once each. This should speed up most standard use cases.
The ctx.fail() function has been removed. Previously you'd use it to return more information about a failure inside a struct. Now you can simply return a partial failure object.
The ctx.check() function has been removed. Previously you'd use it to validate nested objects in more complex struct shapes. Now you can use the new entries property for this instead.
The context.struct and context.value properties have been removed. These properties were previously available, but unnecessary since anywhere you have the context object you will also know the value and the specific struct that is being validated. Keeping them around required extra unnecessary plumbing in the library that made composing structs much more difficult so they were removed.
New Describe utility type. This new utility lets you define a struct from an existing TypeScript type and ensure that the struct's validation matches it, otherwise TypeScript's compiler will error. For example:
type User = {
id: number
name: string
}
const User: Describe<User> = object({
id: string(), // This mistake will fail to pass type checking!
name: string(),
})The coerce helper has changed to be more type-safe! Previously coerce functions were called with value: unknown because they ran before all validation. However, now they take a new second argument that is another struct to narrow the cases where coercions occurs. This means the value for coercion will now be type-safe.
// Previously
const MyNumber = coerce(number(), (value) => {
return typeof value === 'string' ? parseFloat(value) : value
})
// Now
const MyNumber = coerce(number(), string(), (value) => {
return parseFloat(value)
})New assign, pick, and omit object utilities. These utilities make composing object structs together possible, which should make re-using structs in your codebase easier.
// Combine two structs with `assign`:
const a = object({ id: number() })
const b = object({ name: string() })
const c = assign([a, b])
// Pick out specific properties with `pick`:
const a2 = pick(c, ['id'])
// Omit specific properties with `omit`:
const a3 = omit(c, ['name'])New unknown struct. This is the same as the existing any struct, but it will ensure that in TypeScript the value is of the more restrictive unknown type so it encourages better type safety.
const Shape = type({
id: number(),
name: string(),
other: unknown(),
})New integer, regexp, and func structs. These are just simple additions for common use cases of ensuring a value is an integer, a regular expression object (not a string!), or a function.
const Shape = type({
id: integer(),
matches: regexp(),
send: func(),
})New max/min refinements. For refining number (or integer) or date structs to ensure they are greater than or less than a specific threshold. The third argument can indicate whether to make the threshold exclusive (instead of the default inclusive).
const Index = min(number(), 0)
const PastOrPresent = max(date(), new Date())
const Past = max(date(), new Date(), { exclusive: true })Even more information on errors. Errors now expose the error.refinement property when the failure originated in a refinement validation. And they also now have an error.key property which is the key for the failure in the case of complex values like arrays/objects. (Previously the key was retrievable by checking error.path, but this will make the 90% case easier.)
The coerce helper has been renamed to create. This will hopefully make it more clear that it's fully coercing and validating a value against a struct, throwing errors if the value was invalid. This has caused confusion for people who though it would just coerce the value and return the unvalidated-but-coerced version.
// Previously
const user = coerce(data, User)
// Now
const user = create(data, User)The struct, refinement and coercion factories have been renamed. This renaming is purely for keeping things slightly cleaner and easier to understand. The new names are define, refine, and coerce. Separating them slightly from the noun-based names used for the types themselves.
// Previously
const Email = struct('email', isEmail)
const Positive = refinement('positive', number(), n => n > 0)
const Trimmed = coercion(string(), s => s.trim()
// Now
const Email = define('email', isEmail)
const Positive = refine(number(), 'positive', n => n > 0)
const Trimmed = coerce(string(), s => s.trim())Note that the order of refine arguments has changed to be slightly more natural, and encourage scoped refinement names.
The length refinement has been renamed to size. This is to match with the expansion of it's abilities from purely strings and arrays to also now include numbers, maps, and sets. In addition you can also omit the max argument to specify an exact size:
// Previously
const Name = length(string(), 1, 100)
const MyArray = length(array(string()), 3, 3)
// Now
const Name = size(string(), 1, 100)
const MyArray = size(array(string()), 3)
const Id = size(integer(), 1, Infinity)
const MySet = size(set(), 1, 9)The StructType inferring helper has been renamed to Infer. This just makes it slightly easier to read what's going on when you're inferring a type.
// Previously
type User = StructType<typeof User>
// Now
type User = Infer<typeof User>The error.type property has been standardized. Previously it was a human-readable description that sort of incorporated the schema. Now it is simple the plain lowercase name of the struct in question, making it something you can use programmatically when formatting errors.
// Previously
'Array<string>'
'[string,number]'
'Map<string,number>'
// Now
'array'
'tuple'
'map'The 0.10 version is a complete overhaul with the goal of making Superstruct much simpler and easier to understand, and with complete support for runtime type signatures TypeScript.
This makes it much more powerful, however the core architecture has had to change to make it happen. It will still look very similar, but migrating between the versions will be more work than usual. There's no requirement to upgrade, although if you're using Superstruct in concert with TypeScript you will have a much better experience.
All types are created from factories. Previously depending on whether the type was a complex type or a scalar type they'd be defined different. Complex types used factories, whereas scalars used strings. Now all types are exposed as factories.
For example, previously:
import { struct } from 'superstruct'
const User = struct.object({
name: 'string',
age: 'number',
})Now becomes:
import { object, string, number } from 'superstruct'
const User = object({
name: string(),
age: number(),
})Custom scalars are no longer pre-defined as strings. Previously, you would define all of your "custom" types in a single place in your codebase and then refer to them in structs later on with a string value. This worked, but added a layer of unnecessary indirection, and made it impossible to accomodate runtime type signatures.
In the new version, custom types are defined extremely similarly to non-custom types. And this has the added benefit that you can easily trace the custom type definitions by just following import statements.
Here's how it used to work:
import { superstruct } from 'superstruct'
import isEmail from 'is-email'
const struct = superstruct({
types: {
email: isEmail,
},
})
const Email = struct('email')And here's what it would look like now:
import { struct } from 'superstruct'
import isEmail from 'is-email'
const Email = struct('email', isEmail)Validation logic has been moved to helper functions. Previously the assert and is helpers lived on the struct objects themselves. Now, these functions have been extracted into separate helpers. This was unfortunately necessary to work around limitations in TypeScript's asserts keyword.
For example, before:
User.assert(data)Now would be:
import { assert } from 'superstruct'
assert(data, User)Coercion is now separate from validation. Previously there was native logic for handling default values for structs when validating them. This has been abstracted into the ability to define any custom coercion logic for structs, and it has been separate from validation to make it very clear when data can change and when it cannot.
For example, previously:
const output = User.assert(input)Would now be:
const input = coerce(input, User)The coerce step is the only time that data will be transformed at all by coercion logic, and the assert step no longer needs to return any values. This makes it easy to do things like:
if (is(input, User)) {
// ...
}Validation context is now a dictionary of properties. Previously when performing complex validation logic that was dependent on other properties on the root object, you could use the second branch argument to the validation function. This argument has been changed to be a context dictionary with more information. The same branch argument can now be accessed as context.branch, along with the new information.
Unknown properties of objects now have a 'never' type. Previously unknown properties would throw errors with type === null, however the newly introduced 'never' type is now used instead.
Defaults are now defined with a separate coercion helper. Previously all structs took a second argument that defined the default value to use if an undefined value was present. This has been pulled out into a separate helper now to clearly distinguish coercion logic.
For example, previously you'd do:
const Article = struct.object(
{
title: 'string',
},
{
title: 'Untitled',
}
)Whereas now you'd do:
const Article = defaulted(
object({
title: string(),
}),
{
title: 'Untitled',
}
)Optional arguments are now defined with a seperate factory. Similarly to defaults, there is a new optional factory for defined values that can also be undefined.
Previously you'd do:
const Flag = struct('string?')Now you'd do:
const Flag = optional(string())Several structs have been renamed. This was necessary because structs are now exposed directly as variables, which runs afoul of reserved words. So the following renames have been applied:
interface->typeenum->enumsfunction->func
Several structs have been renamed! Superstruct tries to mimic established naming schemes whenever possible for its API, and TypeScript is one of our main comparisons. To make things easier for people, we've renamed a few structs to more closely match their TypeScript counterparts:
- The
liststruct is now calledarray. - The
partialstruct is now calledpick. - The
dictstruct is now calledrecord.
Hopefully this will make them easier to understand at a glance!
The enums struct has been removed! This was special-cased in the API previously, but you can get the exact same behavior by creating an using the array and enum structs:
struct.array(struct.enum(['red', 'blue', 'green']))The any struct has been removed! (Not the scalar though.) Previously struct.any() was exposed that did the same thing as struct(), allowing you to use shorthands for common structs. But this was confusingly named because it has nothing to do with the 'any' scalar type. And since it was redundant it has been removed.
The interface struct now returns the original, unaltered value! In an effort to make things more familiar, the interface struct now always returns the object that it is called with when it passes validation. So if the object was a function, a function will be returned. This makes it match more closely with the idea of "structural typing" that TypeScript and other typing systems are based on. _If you want the old behavior, use the pick struct.
Computed values function signatures have changed! Previously a computed value would be called with a signature of (value, root) in some cases and (value, parent) in others. This was confusing, and the cause for the inconsistency was complex. This logic has been simplified, and now computed values are called with (value, branch, path) in all cases.
struct.dynamic((value, branch, path) => {
value === branch[branch.length - 1] // you can get the value...
const parent = branch[branch.length - 2] // ...and the parent...
const key = path[path.length - 1] // ...and the key...
value === parent[key]
const root = branch[0] // ...and the root!
})The path is an array of keys representing the nested value's location in the root value. And the branch is an array of all of the sub values along the path to get to the current one. This allows you to always be able to receive both the parent and the root values from any location—as well as any value in between.
The error.errors property has been renamed error.failures, and isn't cyclical. It being cyclical caused lots of issues whenever an StructError object was attempted to be serialized. And the errors property was slightly confusing because the elements of the array weren't full error objects. The new structure is easier to understand and work with.
The error.reason property is no longer special-cased. Previously you could return a "reason" string from validator functions and it would be added to error objects. However, now you must return an error properties object (with a reason property if you'd like), and all of the properties will be added to the error object. This makes Superstruct even more flexible as far as custom error details go.
The type property of structs have been rewritten to be more clear. This is an implementation mostly, but the struct.type string which shows up in error messages have been tweaked to be slightly more clear exactly what type they are checking for.
Superstruct is now written in TypeScript. It was rewritten from the ground up to make use of types, and to have better inline documented if you use a TypeScript-compatible IDE. There are probably improvements that can be made, so if you'd like to contribute please do!
A new partial struct mimics TypeScript's Partial utility. The new struct validates that its input partially matches an object defined as a set of properties with associated types. All of the properties of the object are optional.
A new size struct allows validating array and string lengths. The new struct validates that its input has a certain size, by checking its length property. This works strings or arrays.
You can now provide a custom Error setting. By passing in your own constructor when configuring Superstruct you can have complete control over the exact errors that are generated by structs that fail validation.
- The build process now outputs ES5 code. Previously it was outputting ES6 code, which posed problems for some builders. This change shouldn't really affect anyone negatively, but it's being released as a breaking version just in case.
- Invalid
Dateobjects are now considered invalid. Previously using the built-in'date'validator would only check that the object was aDateinstance, and not that it was a valid one. This has been fixed, and although it is technically a breaking change, most everyone would have expected this behavior to begin with.
-
Validators must now return
true,falseor an error reason string. Previously any truthy value would be considered valid. Now you can provide more information for the thrown errors by providing a string which will be attached aserror.reason. However, this means that truthy string values now equate to invalid, not valid. -
Property validators now receive
dataas their second argument. Previously you only had access to the propertyvalue, but now you also have access to the entire object'sdata.
- Errors can now contain reason information. Validator functions can now return string instead of a boolean, denoting the reason a value was invalid. This can then be used to create more helpful error messages.
-
objectstructs are no longer optional-ish. Previously object struct types would not throw ifundefinedwas passed and no properties were required. This was not only confusing, but complex to maintain. Now if you want an object struct to be optional, use thestruct.optional(...)helper. -
Removed the
Struct.defaultmethod. If you need to get the default value, use theStruct.validateorStruct.assertmethods's return value instead.
- Added the
dict,enum,intersection,unionandtuplestructs. These are all available asstruct.dict,struct.enum, etc.
-
The
validate()method now returns[ error, result ]. Previously it only had a single return value, which necessitated extra type checking to see if the value was an error or a result. Now you can just destructure the array to get either return value, for easier coding. -
Errors have been simplified, removing "codes". Previously there were multiple types of errors that were thrown and you could differentiate between them with the
error.codeproperty. But the other properties of the error already let you infer the code, so having multiple types of errors made for a larger API surface without much benefit.
- Structs are now functions again. 😄 They are built on the same underlying schema classes underneath though, since that helps the code structure. But to allow for the
struct = Struct({ ... })syntax the structs themselves have changed to be function.
- The basic case is now
Struct(data). Previously you had to useStruct.assert(data). Although theassertmethod (and others) are still there, the basic case is a bit terser and more similar to the struct-initializing APIs in other languages.
- Structs are now classes instead of functions. This is better in terms of the API being a bit less magic-y. It's also useful so that we can add other helpful methods to structs besides the
assertmethod. What was previouslystruct(data)is nowstruct.assert(data).
🎉