Runtime Verification Library for browsers and Node.js.
This library is isomorphic: It runs equally well in modern browsers and on the server with Node.js.
The latest versions of major browsers, and Node LTS releases, are supported.
Give it a test drive with RunKit!
❗️ This library is ESM/CJS in Node, but ESM-only in the browser.
npm install rtvjsThe package's ./dist directory contains various types of builds:
./node/rtv[.dev].cjs: CJS (not minified, primarily for use in Node)./node/rtv[.slim][.dev].mjs: ESM (not minified, primarily for use in Node)./browser/rtv.esm[.slim][.dev].js: ESM (not minified, for use by bundlers; non-slim/self-contained Dev version is good for browser debugging as a counterpart to the minified Prod version)./browser/rtv.esm.min.js: ESM (minified, primarily for use directly in the browser in Prod)
The .dev CJS and ESM builds include dev-only features such as deprecation warnings and more helpful console logs.
These builds are smaller in size to optimize on download time and bundling efficiency.
The .slim CJS and ESM builds depend on @babel/runtime and lodash-es external dependencies, both of which should automatically get installed when you install rtvjs.
The CJS build can be used like this in a Node.js script:
const rtv = require('rtvjs');The ESM build can be used like this either in Node.js, a bundler, or the browser (note a default export is not provided):
import * as rtv from 'rtvjs'; // import all into an `rtv` namespace
import { verify, STRING, ... } from 'rtvjs'; // selective imports onlyWhile a UMD build isn't provided, it's still possible to load the library as a global directly in the browser as follows:
❗️ It's important to specify the full path to the bundle you want to use. Neither one of the CDNs suggested below seem to understand
package.json:exportsin order to find the right file.
<script type="module">
import * as rtv from 'https://esm.sh/rtvjs@5/dist/browser/rtv.esm.min.js';
// OR: import * as rtv from 'https://unpkg.com/rtvjs@5/dist/browser/rtv.esm.min.js';
window.rtv = rtv;
</script>The non-slim builds are self-contained and optimized for browsers.
This README, as well as the API, are hosted at rtvjs.stefcameron.com.
To provide an easy, intuitive way to perform validations at runtime on values whenever they cross the boundaries of an API or a function call.
Tools like TypeScript and Flow are useful for static analysis (i.e. as code is being written and then transpiled to regular JavaScript), but they come at a price and they don't work at runtime.
For example, they can't signal when there are integration issues between frontend and backend systems that are being co-developed. In one conversation, an API may be designed to return an object with certain properties. Later on, an on-the-fly decision to alter the implementation (yes, it happens in spite of the best intentions and processes), or simply a bug in the implementation, may result in an object that is missing an expected property, or has a property with an unexpected value.
Let's consider a case where a "state" property, which is really an enumeration of string values, ends-up set to an unexpected state. What should a client do with an unexpected state when there's no implementation to back it up? Ignoring it could be an option, but perhaps not the best course of action. Even worse, the unexpected state somehow could trickle deep down into code before it finally causes an exception, making it really difficult to find the true source of the problem.
RTV.js can help signal the unexpected state by failing early, right at the API boundary:
async function getTodoList() {
const response = await fetch('/api/todos');
const json = await response.json();
// verify (require) that the response be a list of TODO items: this function
// will throw if `json` doesn't meet the specified typeset (requirement)
rtv.verify(json, [[{ // list of objects (could be empty)
// non-empty string
title: rtv.STRING,
// 'YYYY-MM-DD', or null
due: [rtv.EXPECTED, rtv.STRING, {exp: '\\d{4}-\\d{2}-\\d{2}'}],
// string (could be empty), null, or not even defined
note: [rtv.OPTIONAL, rtv.STRING]
}]]);
return json;
}There may also be a need to ensure that a critical function call is being given the parameters it expects. Rather than write a series of if (!state) { throw new Error('state is required'); } (which don't tell us much about what "state" is expected to be, other than it's required), it would be more helpful to have an easy way to express that "state" should be a non-empty string with a value in a given list (i.e. a value found in an enumeration).
RTV.js can help signal the unexpected state immediately when execution enters the function:
function applyState(state) {
rtv.verify(state, [rtv.STRING, {oneOf: ['on', 'off']}]);
if (state === 'on') {
// turn the lights on
} else {
// turn the lights off
}
}
applyState('on'); // ok
applyState('dimmed'); // ERRORWhile tools like TypeScript and Flow have their merits, they come at a price. Typings or not, integration issues will remain. RTV.js allows you to check for types at runtime, when it really matters, and has a simple API so it's easy to learn.
The following statement verifies that the variable "state" is a non-empty string whose value is found in a list of permitted values:
rtv.verify(state, [rtv.STRING, {oneOf: ['on', 'off']}]);The [rtv.STRING, {oneOf: ['on', 'off']}] portion of the example above is called a typeset. It expresses the expectation for the value of the "state" variable.
Typesets must be:
- Easy to express, using rich types and qualifiers.
- Composable, whereby complex typesets can be built by combining multiple typesets into larger ones.
- Easy to customize, using custom validators when the existing types and arguments don't provide the exact verification needed on a value.
- Intuitive, using simple native JavaScript language constructs like strings (for types), inline Arrays
[]for lists and complex typesets, and inline objects{}for shapes (i.e. interfaces). - Serializable to JSON via
JSON.stringify()so they can be easily transferred between systems.- Backend and frontend systems in JavaScript stacks could dynamically inform one another of expectations by sharing typesets.
- Similar to the
@contextproperty of a JavaScript Object for JSON-LD, an object's expected shape could be transferred along with the object itself. - With the exceptions of custom validator functions and the
ctorproperty of shape object arguments.
Tutorials and example uses of the RTV.js library.
To make it clear, in this tutorial, which properties and functions from RTV.js, we'll start by importing everything into an rtv object:
import * as rtv from 'rtvjs';You could also drop the object and import individual names, such as:
import { check, verify, STRING, ... } from 'rtvjs';RTV.js provides two functions for verifying values against typesets. A typeset is simply a set of one or more types that form an expectation about the value:
rtv.verify(value, typeset); // will throw an error if verification fails
rtv.check(value, typeset); // returns the error instead of throwing itTypesets can be strings, objects (shapes), functions (custom validators), or Arrays (multiple possibilities).
At their simplest, typesets are strings that represent type names like STRING, INT, DATE, etc. See the full list of types here.
rtv.verify('Hello world!', rtv.STRING); // ok
rtv.verify('', rtv.STRING); // ERROR: a required string cannot be emptyThe first verification succeeds because the value is a non-empty string. The second one fails because the typeset uses the default qualifier, which is REQUIRED. A required string cannot be empty (nor can it be null or undefined).
In some implementations, an empty string is considered a bad value because it's a falsy value in JavaScript, just like null, undefined, false, 0, and NaN.
There are 3 other qualifiers, EXPECTED, OPTIONAL, and TRUTHY. A typeset may only have one qualifier, and it must be specified before any types.
The only way to specify an alternate qualifier is to use an Array to describe the typeset: [<qualifier>, types...]
If we wanted to accept an empty string (or null) as the value, we could use the EXPECTED qualifier:
rtv.verify('Hello world!', [rtv.EXPECTED, rtv.STRING]); // ok
rtv.verify('', [rtv.EXPECTED, rtv.STRING]); // ok
rtv.verify(null, [rtv.EXPECTED, rtv.STRING]); // okIf we had a variable which we expect to be an object whenever it's value is truthy, we could use the TRUTHY qualifier, which would permit any falsy value, but require the value to be of a specified type otherwise:
let objectOrFalsy = false;
rtv.verify(objectOrFalsy, [rtv.TRUTHY, rtv.PLAIN_OBJECT]); // ok
objectOrFalsy = {hello: 'world!'};
rtv.verify(objectOrFalsy, [rtv.TRUTHY, rtv.PLAIN_OBJECT]); // ok
rtv.verify(objectOrFalsy, [rtv.TRUTHY, rtv.ARRAY]); // ERROR: value is not an array
// similar to how the following code would either print "world!" or not get executed
// depending on the truthiness of `objectOrFalsy`
if (objectOrFalsy) {
console.log(objectOrFalsy.hello);
}Some types accept arguments. Arguments are simple objects that map argument names to values, and immediately follow a type in a typeset. Once again, an Array must be used to describe the typeset. Type arguments are optional, unless otherwise stated; some types don't accept arguments.
The STRING type accepts arguments, one of which is min. It lets us specify the minimum length of the string. By default, when the qualifier is REQUIRED, min defaults to 1, but we can override that:
rtv.verify('Hello world!', [rtv.STRING, {min: 0}]); // ok
rtv.verify('', [rtv.STRING, {min: 0}]); // ok
rtv.verify(null, [rtv.STRING, {min: 0}]); // ERRORThis verifies the value cannot be null or undefined because of the (implied) REQUIRED qualifier. However, it could be empty because the min argument allows a zero-length string as the value.
So far, we've seen simple typesets: Either just a string as the type name, or the type name and some arguments, and an optional qualifier that precedes it. There may be cases where a value could be one of multiple types. To verify against additional types, an Array is used to state all the possibilities: [<qualifier>, <type1>, <type1-args>, <type2>, <type2-args>, ...]. This is called an "Array typeset", which we've already seen in the two previous sections.
Since a value can only be of a single type at any given time, Array typesets are evaluated using a short-circuit OR conjunction, which means the verification will pass as long as at least one type verifies the value (and verification will stop evaluating any other types against the value once a match is made).
For example, we could verify that a value is either a boolean, or a string that looks like a boolean:
const typeset = [rtv.BOOLEAN, rtv.STRING, {
exp: '^(?:true|false)$',
expFlags: 'i'
}];
rtv.verify(true, typeset); // ok
rtv.verify('true', typeset); // ok
rtv.verify('True', typeset); // ok
rtv.verify('TRUE', typeset); // ok
rtv.verify(false, typeset); // ok
rtv.verify('false', typeset); // okSince the check for the
BOOLEANtype is faster than evaluating a regular expression against a string, we list theBOOLEANtype first in the typeset.
It's worth pointing out here that:
The same type can appear multiple times in the same typeset.
This is very useful when a value could be of one or another set of values.
For example, we could verify that a value is a finite number in two different ranges:
const typeset = [
rtv.FINITE, {min: 0, max: 9},
rtv.FINITE, {min: 100, max: 199}
];
rtv.verify(1, typeset); // ok
rtv.verify(50, typeset); // ERROR
rtv.verify(150, typeset); // ok
rtv.verify(-1, typeset); // ERROR
rtv.verify(200, typeset); // ERRORThis is also useful for composition where you need to combine multiple smaller typesets into a larger one:
const lowerRangeTs = [rtv.FINITE, {min: 0, max: 9}];
const upperRangeTs = [rtv.FINITE, {min: 100, max: 199}];
const typeset = [...lowerRangeTs, ...upperRangeTs];
rtv.verify(1, typeset); // ok
rtv.verify(50, typeset); // ERRORtypeset here will yield the same results as in the previous example.
Most of the time, especially when integrating with an API, you'll want to verify what you receive against an expected shape. A shape describes the interface an object is expected to have. As the term implies, an interface describes the properties, and types thereof, expected on an object while ignoring any other properties that the object may have (since the code using this object shouldn't care about them anyway).
Plain JavaScript objects are used to describe shapes, where expected property names are own-enumerable properties mapped to typesets. For example, we could describe a simple TODO item like this:
{
title: rtv.STRING, // non-empty string
created: rtv.DATE, // Date instance
priority: rtv.INT // some whole number
}Since typesets are fully nestable/composable, we can get a bit more sophisticated by using Array typesets so we can provide arguments and different qualifiers:
{
title: rtv.STRING, // required (non-empty) title
created: [rtv.OPTIONAL, rtv.DATE], // either a TODO or just a note
priority: [rtv.INT, {oneOf: [0, 1, 2]}] // 0=none, 1=low, 2=high
}Since shapes also represent objects, they have an implied (default) type of OBJECT. When fully-qualified (which means not using any implied typeset elements like the qualifier and type), the shape would move into the special $ argument of the OBJECT type:
[rtv.REQUIRED, rtv.OBJECT, {$: {
title: rtv.STRING,
created: [rtv.OPTIONAL, rtv.DATE],
priority: [rtv.INT, {oneOf: [0, 1, 2]}]
}
}]When the default object type is sufficient, it's really easy to nest shapes. Let's say our TODO item also had a note, which is an object with "text" and "updated" properties:
const {STRING, DATE, INT} = rtv;
const {EXPECTED, OPTIONAL} = rtv;
const item = {
title: 'Make Christmas Oatmeal',
due: new Date('12/25/2018'),
priority: 1,
note: {
text: 'Make 4 cups to have enough to share!',
updated: new Date('09/21/2018')
}
};
rtv.verify(item, {
title: STRING,
created: [OPTIONAL, DATE],
priority: [INT, {oneOf: [0, 1, 2]}],
note: { // <- nested shape
text: STRING, // <- required contents
updated: DATE // <- required Date
}
}); // okThe typeset above would require a TODO item to have a "note" with a non-empty string value for "text", and a Date instance for "updated". We could make the entire note optional, however, by expecting it to be either null if a note wasn't provided, or the shape if one was:
const {STRING, DATE, INT} = rtv;
const {EXPECTED, OPTIONAL} = rtv;
const item = {
title: 'Make Christmas Oatmeal',
due: new Date('12/25/2018'),
priority: 1,
note: null
};
rtv.verify(item, {
title: STRING,
created: [OPTIONAL, DATE],
priority: [INT, {oneOf: [0, 1, 2]}],
note: [EXPECTED, { // <- null, or note object
text: STRING,
updated: DATE
}]
}); // okWhen the default object type is implied, this is called the shorthand syntax. For shapes, it may be used when the typeset is the shape itself, or in an Array typeset that is not fully-qualified, when a qualifier immediately precedes the shape (as we've done above for the "note" property).
Many times, an API response or a function's arguments will contain a list of values or objects. At their most basic, lists are simple JavaScript Arrays that contain values of some type. The simplest way to verify a list is homogenous is to use the shorthand syntax for the ARRAY type:
[[rtv.STRING]]This would verify that an Array contains non-empty string values, but the Array could be empty, given the default arguments.
Note the nested Array.
What the example above defines is an Array typeset that has a single implied ARRAY type with an element typeset of STRING that will be applied to all elements found in the Array.
When the full notation is used, the element typeset moves into the $ argument:
[rtv.ARRAY, {$: [rtv.STRING]}] // same as before, but in full notationEither form is acceptable, and either form can show-up anywhere in a typeset. Therefore, we could verify a value is either a boolean, an Array of non-empty strings, or an Array of integers like this:
[rtv.BOOLEAN, [rtv.STRING], [rtv.INT]]A more practical example could be requiring a TODO item to have a non-empty list of notes associated with it, if "notes" isn't null, meaning there are no notes (i.e. either "notes" is null because there are no notes, or "notes" is an Array of note objects containing at least one note):
const {STRING, DATE, INT} = rtv;
const {EXPECTED, OPTIONAL} = rtv;
const item = {
title: 'Make Christmas Oatmeal',
due: new Date('12/25/2018'),
priority: 1,
note: null
};
const shape = {
title: STRING,
created: [OPTIONAL, DATE],
priority: [INT, {oneOf: [0, 1, 2]}],
note: [EXPECTED, ARRAY, { // <- null, or non-empty Array of notes
$: {
text: STRING,
updated: DATE
},
min: 1 // <- require a non-empty Array when not null
}]
};
rtv.verify(item, shape); // ok
item.notes = [];
rtv.verify(item, shape); // ERROR: `notes` cannot be empty
item.notes.push({
text: 'Make 4 cups to have enough to share!',
updated: new Date('09/21/2018')
});
rtv.verify(item, shape); // okWhile built-in validations are convenient, there may be occasions where a type, or even its arguments, aren't sufficient to verify the value. In that case, the typeset can be customized with a custom validator function.
The function on its own is considered a valid typeset, and gets an implied type of ANY, which validates anything, even
undefinedandnull, regardless of the qualifier.
Let's say we wanted to verify that a value is a multiple of two. None of the numeric type arguments will verify that on their own, so we would need a custom validator:
function validator(value) {
const n = parseInt(value);
return (!IsNaN(n) && n % 2 === 0);
}
rtv.verify(2, validator); // ok
rtv.verify(3, validator); // ERRORA custom validator can fail the verification either by returning a falsy value (other than undefined), or throwing an Error. When a falsy value is returned, a default Error will be generated. Throwing an error with a helpful message is the recommended way to fail verification because of a custom validator:
function(value) {
const n = parseInt(value);
if (IsNaN(n) || n % 2 != 0) {
throw new Error('Not a number, or not a multiple of two.');
}
}
rtv.verify(2, validator); // ok
rtv.verify(3, validator); // ERROR (rootCause: 'Not a number...')The error thrown by the custom validator (or the one generated by the library) will be included in the rootCause property of the failed verification results.
Custom validators are intended to be used as compliments to existing types rather than complete replacements. For example, rather than worry about parsing the value as an integer and checking to see if it's not a number, we could let RTV.js first verify the value is an integer by using an Array typeset:
const typeset = [rtv.INT, (v) => v % 2 === 0];
rtv.verify(2, typeset); // ok
rtv.verify(3, typeset); // ERROR (rootCause: 'Verification failed...')An Array typeset may have at most one custom validator, and it must be the last element. Each sub-typeset may have its own validator. When one or more types are in the typeset, the validator is immediately invoked if one of the types matches (i.e. verifies) the value (any remaining types are ignored):
const typeset = [rtv.INT, rtv.STRING, (v) => v % 2 === 0];
// in both cases, STRING verification is skipped because INT matches first
rtv.verify(2, typeset); // ok
rtv.verify(3, typeset); // ERROR (rootCause: 'Verification failed...')Finally, we could enhance our TODO item verification with a custom validator that verifies the created Date is not in the past:
const {STRING, DATE, INT} = rtv;
const {EXPECTED, OPTIONAL} = rtv;
const item = {
title: 'Make Christmas Oatmeal',
due: new Date(Date.now() + 24 * 60 * 60 * 1000), // tomorrow
priority: 1,
note: null
};
const shape = {
title: STRING,
created: [
OPTIONAL,
DATE,
(v) => !v || v.getTime() >= Date.now()) // <- validator
],
priority: [INT, {oneOf: [0, 1, 2]}],
note: [EXPECTED, ARRAY, {
$: {
text: STRING,
updated: DATE
},
min: 1
}]
};
rtv.verify(item, shape); // ok
item.due = new Date(Date.now() - 12 * 60 * 1000); // 12 hours ago
rtv.verify(item, shape); // okNotice how the validator must handle
nullandundefinedvalues because of the OPTIONAL qualifier, and is careful to return a truthy result so that the property remains optional.
One of the main goals of this library is to help validate API payloads at runtime, which is where the backend/frontend contract tends to get broken for various reasons.
Sometimes, the payloads returned from the API are much larger than your client cares for. That's why RTV.js makes it easy to define shapes that verify only the subset of properties your app really cares about.
To go even further, while it verifies a given value, RTV.js also generates a Minimum Viable Value (MVV) that represents the "smallest" dataset that would still pass the same validation.
To put this in perspective, if you're dealing with, say, a Kubernetes API and you get a massive payload consisting of a long list of cluster objects, each further described by a very large Kubernetes cluster JSON object from which you only care to use 10 properties, you'll only write a shape that describes those 10 properties. RTV.js will then generate a new object containing only those 10 properties (no matter how deeply nested they are) based on the shape you used for verification.
Discarding the original value and using the MVV instead could potentially save your app from heavy memory use depending on how many closures in your code end-up imprisoning those large objects.
To illustrate this, let's say you had retrieved a list of tasks from an API endpoint:
const tasks = [
{
title: 'Implement the feature',
description: 'A very long description...',
due: new Date(),
tags: [
{ id: 1, name: 'tag1' },
{ id: 2, name: 'tag2' },
{ id: 3, name: 'tag3' },
],
notes: [
{
text: 'Note 1',
author: 'Sam',
date: new Date(),
tags: [
{ id: 4, name: 'tag4' },
{ id: 5, name: 'tag5' },
{ id: 6, name: 'tag6' },
],
},
{
text: 'Note 2',
author: 'Susie',
date: new Date(),
tags: [
{ id: 7, name: 'tag7' },
{ id: 8, name: 'tag8' },
{ id: 9, name: 'tag9' },
],
},
],
},
{
title: 'Add the tests',
description: 'Long description...',
due: new Date(),
tags: [
{ id: 1, name: 'tag1' },
{ id: 2, name: 'tag2' },
],
notes: [
{
text: 'Note 1',
author: 'Melissa',
date: new Date(),
tags: [{ id: 4, name: 'tag4' }],
},
{
text: 'Note 2',
author: 'Patrick',
date: new Date(),
tags: [
{ id: 8, name: 'tag8' },
{ id: 9, name: 'tag9' },
],
},
{
text: 'Note 3',
author: 'Melissa',
date: new Date(),
tags: [
{ id: 3, name: 'tag3' },
{ id: 4, name: 'tag4' },
],
},
],
},
];You might verify such a payload like this, because you only care about title, tags (specifically, tag names), and notes for the tasks, and only text and tags (specifically, tag ids):
const result = rtv.verify(tasks, [[{
title: rtv.STRING,
tags: [[{ name: rtv.STRING }]],
notes: [[{
text: rtv.STRING,
tags: [
[
{
id: rtv.SAFE_INT,
},
],
],
}]],
}]]);If the validation is successful, the returned RtvSuccess object would have the MVV stored in its mvv property, and it would look like this:
[
{
title: 'Implement the feature',
tags: [
{ name: 'tag1' },
{ name: 'tag2' },
{ name: 'tag3' },
],
notes: [
{
text: 'Note 1',
tags: [
{ id: 4 },
{ id: 5 },
{ id: 6 },
],
},
{
text: 'Note 2',
tags: [
{ id: 7 },
{ id: 8 },
{ id: 9 },
],
},
],
},
{
title: 'Add the tests',
tags: [
{ name: 'tag1' },
{ name: 'tag2' },
],
notes: [
{
text: 'Note 1',
tags: [{ id: 4 }],
},
{
text: 'Note 2',
tags: [
{ id: 8 },
{ id: 9 },
],
},
{
text: 'Note 3',
tags: [
{ id: 3 },
{ id: 4 },
],
},
],
},
]RTV.js provides a configuration interface which allows checks (rtv.check(value, typeset)) and verifications (rtv.verify(value, typeset)) to be globally enabled or disabled:
rtv.config.enabled = false; // default: true
rtv.verify('foo', rtv.INT); // no-op, always returns RtvSuccess
rtv.check('foo', rtv.INT); // no-op, always returns RtvSuccessBut why even make the call at all?
if (rtv.config.enabled) {
rtv.verify('foo', rtv.INT);
}
// OR more terse:
rtv.config.enabled && rtv.verify('foo', rtv.INT);
// OR with your own global:
DO_TYPE_CHECKS && rtv.verify('foo', rtv.INT);Now, a JavaScript bundler that supports tree shaking (e.g. Webpack or Rollup) can be configured to completely exclude the entire code for a build. This could be handy if you're concerned about script download size over runtime checks, say, in a production build. See the Rollup example for more information.
Let's say we're building a simple TODO app. We might use the following object as representative of a "todo" item in a list:
const item = {
title: 'Make Christmas Oatmeal',
due: new Date('12/25/2018'),
priority: 1,
notes: [
{
text: 'Ingredients: Cranberries, apples, cinnamon, walnuts, raisins, maple syrup.',
updated: new Date('09/20/2018')
},
{
text: 'Make 4 cups to have enough to share!',
updated: new Date('09/21/2018')
}
]
};We can describe this object using two shapes:
const {STRING, DATE, INT} = rtv;
const priorities = [1, 2, 3, 4]; // simple enumeration of priority levels
const shapes = {
get todo() { // 'todo' shape
return {
title: STRING,
due: DATE,
priority: [INT, {oneOf: priorities}], // use 'priorities' enum
notes: [[this.note]] // compose 'note' shape into this 'todo' shape
};
},
get note() { // 'note' shape
return {
text: STRING,
updated: DATE
};
}
};Now we can verify that "todo" is a valid TODO item:
rtv.verify(item, shapes.todo);The above verification will pass because "todo" meets the requirements of the shape.
Now let's change the second note in "todo" such that its "updated" property is a boolean, true (a simple indication that the note was changed at some point -- a change that seems to make sense, but would break code that expects a Date object to use for formatting in the UI, for example):
todo.notes[1].updated = true;Lexically, there's no reason for this assignment to fail, but the boolean value violates what is stated in the spec for a TODO item.
If we were to run the same verification again, an exception would be thrown. The exception would be an RtvError with the following properties:
rtv.verify(item, shapes.todo);
// RtvError exception thrown from the above statement:
{
message: 'Verification failed: path="/notes/1/updated", mismatch=["!","DATE"], typeset={"title":"STRING","due":"DATE","priority":["INT",{"oneOf":[1,2,3,4]}],"notes":[[{"text":"STRING","updated":"DATE"}]]}',
path: ['notes', '1', 'updated'], // path to the property that failed verification
mismatch: ['!', 'DATE'], // fully-qualified typeset that caused the failure
typeset: {...}, // reference to "shapes.todo"
value: {...}, // reference to "todo"
...
}The cause property is providing us with the fully-qualified version of the nested typeset that caused the failure. The original typeset simply specified DATE as the nested typeset for the note.updated property.
In reality, all typesets have a qualifier, and the default qualifier is '!' which means the value is required. Required values can neither be undefined nor null. Depending on the type, other restrictions may be imposed, such as the STRING type, which must also not be empty (by default).
For brevity, typesets don't always have to be fully-qualified since the default qualifier is implied when not specified. Note that a typeset must have exactly one qualifier, implied or not, but each nested typeset may have its own qualifier.
For example, some TODO items may not have due dates. However, our shape currently requires them. To handle this requirement, we could alter the nested typeset of the todo.due property to be ['*', DATE] This would state that the due property is expected rather than required, which means its value could be null (but still not undefined). There is a third qualifier, '?', which would indicate the value is optional, in which case it could also be undefined (which, in JavaScript terms, means the property could also not even exist anywhere up the prototype chain of the todo object).
The RtvError object can also be obtained without catching an exception thrown by using the rtv.check() method:
rtv.check(item, shapes.todo); // returns the RtvError objectIf the check was successful, an RtvSuccess would be returned instead. Since both RtvError and RtvSuccess objects have a common valid: boolean property, it's easy to check for success and failure:
if (rtv.check(item, shapes.todo).valid) {
// check passed, "todo" is valid!
} else {
// check failed, ignore the item
}Finally, we can check simple values too:
rtv.verify('1', rtv.INT); // ERROR: not an integer number
rtv.verify('', [rtv.EXPECTED, rtv.STRING]); // ok: expected strings can be null/emptyThis is an advanced use of the RTV.js library. I recommend you read through the Getting Started guide or the Verifications example first.
Let's suppose we have the following shape that describes a simple note:
const {STRING, DATE} = rtv; // some types
const {EXPECTED} = rtv; // some qualifiers
const tags = ['car', 'money', 'reminder', 'grocery'];
const noteShape = {
// required, non-empty string
text: STRING,
// required Array (could be empty) of non-empty tags names from the user's
// list of "tags"
tags: [[STRING, {oneOf: tags}]],
// required Date when the note was created
created: DATE,
// expected date of update (either null, or Date)
updated: [EXPECTED, DATE]
};Based on this shape, we can dynamically define a JavaScript class with getters and setters that ensure they are being set correctly:
const classGenerator = function(shape) {
const ctor = function(initialValues) {
// by definition, a shape descriptor is made-up of its own-enumerable
// properties, so we enumerate them
const props = Object.keys(shape);
const typesets = {}; // prop -> fully-qualified Array typeset
const values = {}; // prop -> value
let initializing = true; // true while we apply "initialValues"
props.forEach((prop) => {
typesets[prop] = rtv.fullyQualify(shape[prop]);
Object.defineProperty(this, prop, {
enumerable: true,
configurable: true, // could be false to lock this down further
get() {
return values[prop];
},
set(newValue) {
const typeset = typesets[prop].concat(); // shallow clone
if (initializing) {
// allow each property to be initially null, or as the typeset specifies
// so we don't end-up with junk data
// NOTE: in a fully-qualified typeset, the qualifier is always the
// first element
typeset[0] = EXPECTED;
}
// we assume there are no interdependencies between nested typesets
// this verification will throw an RtvError if the "newValue"
// violates the property's typeset
rtv.verify(newValue, typeset);
values[prop] = newValue;
}
});
if (initialValues && initialValues.hasOwnProperty(prop)) {
// go through the setter for verification
this[prop] = initialValues[prop];
} else {
// initialize to null
values[prop] = null;
}
});
initializing = false;
};
return ctor;
};Now we can generate a Note class and create an instance:
const Note = classGenerator(noteShape);
const note = new Note({text: 'Hello world!'});
note.text; // "Hello world!", since it was initialized
note.created; // null, since it wasn't initialized
note.text = ''; // ERROR: "text" must be a non-empty stringLet's revisit the Note shape from the Dynamic Classes example, but we'll add one more property, tagCount:
const {STRING, DATE, SAFE_INT} = rtv; // some types
const {EXPECTED} = rtv; // some qualifiers
const tags = ['car', 'money', 'reminder', 'grocery'];
const noteShape = {
// required, non-empty string
text: STRING,
// required Array (could be empty) of non-empty tags names from the user's
// list of "tags"
tags: [[STRING, {oneOf: tags}]],
//
tagCount: SAFE_INT, // <- NEW
//
// required Date when the note was created
created: DATE,
// expected date of update (either null, or Date)
updated: [EXPECTED, DATE]
};The tagCount property should always be an integer equal to the length of the tags array. The most basic validation we could do is the above: Mark it as a SAFE_INT. The problem is, it's not a complete validation because the following Note would pass, however it would still be invalid:
const note = {
text: 'Buy potatoes',
tags: ['reminder', 'grocery'],
tagCount: 1, // <- does not match length of `tags` array
created: new Date(Date.now()),
updated: null
}
rtv.verify(note, noteShape); // ok (but not ok...)To address this issue, we can use the context parameter provided to any custom validator since it provides a reference to the originalValue being validated:
const noteShape = {
...,
// tagCount: SAFE_INT, // <- BEFORE
tagCount: [
SAFE_INT,
(value, match, typeset, context) => { // <- AFTER
if (value !== context.originalValue.tags.length) {
throw new Error('tags and tagCount mismatch');
}
}
],
...
};The first parameter, value, is the value of the tagCount property being validated by the typeset in which the custom validator is located. The fourth parameter, context, provides some additional information such as the original value, that being the Node object itself (the note object given to rtv.verify(note, typeset)).
rtv.verify(note, noteShape); // ERROR: 'tags and tagCount mismatch'With this change, we now have a reactive validation, since it reacts (or adjusts) according to some of the data its given.
originalValue, however, may not always be sufficient because it never changes regardless of the hierarchy of objects being validated. Let's say we wanted to validate a list of notes:
rtv.verify([note], [[noteShape]]); // ERROR: 'cannot read property "length" of undefined'
// (because `originalValue` is the array and does not have a `tags` property,
// so `originalValue.tags.length` causes an exception)This is where parent (and parentKey) are handy: parent will always refer to the immediate enclosing Object, Array, Map, or Set, whenever a property or element within it is being validated. parentKey will be the property or index, depending on parent's type, whose value is being validated. See custom validator context for more details on these two properties.
Therefore, we can change our typeset to this:
const noteShape = {
...,
tagCount: [
SAFE_INT,
(value, match, typeset, context) => {
// if (value !== context.originalValue.tags.length) { // <- BEFORE
if (value !== context.parent.tags.length) { // <- AFTER
throw new Error('tags and tagCount mismatch');
}
}
],
...
};And now the validation works again:
rtv.verify([note], [[noteShape]]); // ok!RTV.js is not your only choice for runtime verification of values. Here are some alternatives you should consider. Compare them to what this library offers and choose the best one to fit your needs!
-
Joi offers object schema validation. In the hapi ecosystem, this is commonly paired with Hoek. Note that
Joiis only supported in Node.js environments. -
prop-types is useful if you're building a React app, but you can't get it to fail on purpose (it's React's support for it that causes errors in the console at runtime in a development build).
-
yup is useful if you want to validate an object schema (and can be a viable alternative to
Joiwhich is only supported on Node.js sinceyupis also supported in the browser).While it has similarities to RTV.js, the picture is very different in practice: Yup has a concept of coercions and transformations such that, for instance,
yup.string().required().validateSync(1)would not fail. Even ifstrictmode is enabled (for which there's no global setting; that would be on a per-statement basis), there are cases where it would still pass! You wouldn't be the only one concerned with this (and it's never been resolved, which is fine sinceyupisn't strictly focused on exact types for validations).RTV.js, on the other hand, is specially designed to validate values against exact types, no implicit type coercions or transformations. It's either a string or it's not. That's what's needed for type validation. Otherwise, (IMO) it's useless, in terms of validation, because you can't be certain of what you have (e.g. if
"1"and1were the same, wouldvalue.substr()always work...?).
Contributing, including local development guide.
See the list of proposed enhancements. Up-vote the ones you like to help contributors prioritize them!
Feel free to log an enhancement if you have an idea! You may also file a PR, although it might be best to discuss your idea with the community first by creating an enhancement issue.