Awesome

znv

Parse your environment with Zod.

Pass in a schema and your process.env. Get back a validated, type-safe, read-only environment object that you can export for use in your app. You can optionally provide defaults (which can be matched against NODE_ENV values like production or development), as well as help strings that will be included in the error thrown when an env var is missing.

Status

Unstable: znv has not yet hit v1.0.0, and per semver there may be breaking changes in minor versions before the v1.0.0 release. Any (known) breaking changes will be documented in release notes. znv is used in production in several services at the primary author's workplace. Feedback and suggestions about final API design are welcome.

Contents

• Quickstart
• Motivation
• Usage
  • parseEnv
  • Extra schemas
• Coercion rules
• Comparison to other libraries
• Complementary tooling
• How do I pronounce znv?

Quickstart

npm i znv zod
# or
yarn add znv zod

Create a file named something like env.ts:

import { parseEnv } from "znv";
import { z } from "zod";

export const { NICKNAME, LLAMA_COUNT, COLOR, SHINY } = parseEnv(process.env, {
  NICKNAME: z.string().min(1),
  LLAMA_COUNT: z.number().int().positive(),
  COLOR: z.enum(["red", "blue"]),
  SHINY: z.boolean().default(true),
});

console.log([NICKNAME, LLAMA_COUNT, COLOR, SHINY].join(", "));

Let's run this with ts-node:

$ LLAMA_COUNT=huge COLOR=cyan ts-node env.ts

Oops! Let's fix those issues:

$ LLAMA_COUNT=24 COLOR=red NICKNAME=coolguy ts-node env.ts

Now we see the expected output:

coolguy, 24, red, true

Since parseEnv didn't throw, our exported values are guaranteed to be defined. Their TypeScript types will be inferred based on the schemas we used — COLOR will be even be typed to the union of literal strings 'red' | 'blue' rather than just string.

A more elaborate example:

// znv re-exports zod as 'z' to save a few keystrokes.
import { parseEnv, z, port } from "znv";

export const { API_SERVER, HOST, PORT, EDITORS, POST_LIMIT, AUTH_SERVER } =
  parseEnv(process.env, {
    // you can provide defaults with `.default()`. these will be validated
    // against the schema.
    API_SERVER: z.string().url().default("https://api.llamafy.biz"),

    // specs can also be more detailed.
    HOST: {
      schema: z.string().min(1),

      // the description is handy as in-code documentation, but is also printed
      // to the console if validation for this env var fails.
      description: "The hostname for this service.",

      // instead of specifying defaults as part of the zod schema, you can pass
      // them in the `defaults` object. a default will be matched based on the
      // value of `NODE_ENV`.
      defaults: {
        production: "my-cool-llama.website",
        test: "cool-llama-staging.cloud-provider.zone",

        // "_" is a special token that can be used in `defaults`. its value will
        // be used if `NODE_ENV` doesn't match any other provided key.
        _: "localhost",
      },
    },

    // znv provides helpers for a few very common environment var types not
    // covered by zod. these can have further refinements chained to them:
    PORT: port().default(8080),

    // using a zod `array()` or `object()` as a spec will make znv attempt to
    // `JSON.parse` the env var if it's present.
    EDITORS: z.array(z.string().min(1)),

    // optional values are also supported and provide a way to benefit from the
    // validation and static typing provided by zod even if you don't want to
    // error out on a missing value.
    POST_LIMIT: z.number().optional(),

    // use all of the expressiveness of zod, including enums and post-processing.
    AUTH_SERVER: z
      .enum(["prod", "staging"])
      .optional()
      .transform((prefix) =>
        prefix ? `http://auth-${prefix}.cool-llama.app` : "http://localhost:91"
      ),
  });

If any env var fails validation, parseEnv() will throw. All failing specs will be aggregated in the error message, with each showing the received value, the reason for the failure, and a hint about the var's purpose (if description was provided in the spec).

Motivation

Environment variables are one way to pass runtime configuration into your application. As promoted by the Twelve-Factor App methodology, this helps keep config (which can vary by deployment) cleanly separated from code, encouraging maintainable practices and better security hygiene. But passing in configuration via env vars can often turn into an ad-hoc affair, with access and validation scattered across your codebase. At worst, a misconfigured environment will launch and run without apparent error, with issues only making themselves apparent later when a certain code path is hit. A good way to avoid this is to declare and validate environment variables in one place and export the validated result, so that other parts of your code can make their dependencies on these vars explicit.

Env vars represent one of the boundaries of your application, just like file I/O or a server request. In TypeScript, as in many other typed languages, these boundaries present a challenge to maintaining a well-typed app. Zod does an excellent job at parsing and validating poorly-typed data at boundaries into clean, well-typed values. znv facilitates its use for environment validation.

What does znv actually do?

znv is a small module that works hand-in-hand with Zod. Since env vars, when defined, are always strings, Zod schemas like z.number() will fail to parse them out-of-the-box. Zod allows you to use a preprocess schema to handle coercions, but peppering your schemas with preprocessors to this end is verbose, error-prone, and clunky. znv wraps each of the Zod schemas you pass to parseEnv in a preprocessor that tries to coerce a string to a type the schema expects.

These preprocessors don't do any validation of their own — in fact, they try to do as little work as possible and defer to your schema to handle the validation. In practice, this should be pretty much transparent to you, but you can check out the coercion rules if you'd like more info.

Since v3.20, Zod provides z.coerce for primitive coercion, but this is often too naive to be useful. For example, z.coerce.boolean() will parse "false" into true, since the string "false" is truthy in JavaScript. znv will coerce "false" into false, which is probably what you expect.

znv also makes it easy to define defaults for env vars based on your environment. Zod allows you to add a default value for a schema, but making a given default vary by environment or only act as a fallback in certain environments is not straightforward.

Usage

parseEnv(environment, schemas)

Parse the given environment using the given schemas. Returns a read-only object that maps the keys of the schemas object to their respective parsed values.

Throws if any schema fails to parse its respective env var. The error aggregates all parsing failures for the schemas.

environment: Record<string, string | undefined>

You usually want to pass in process.env as the first argument.

It is not recommended to use znv for general-purpose schema validation — just use Zod (with preprocessors to handle coercion, if necessary).

schemas: Record<string, ZodType | DetailedSpec>

Maps env var names to validators. You can either use a Zod schema directly, or pass a DetailedSpec object that has the following fields:

• schema: ZodType

  The Zod validator schema.

• description?: string

  Optional help text that will be displayed when this env var is missing or fails to validate.

• defaults?: Record<string, SchemaInput | undefined>

  An object that maps from NODE_ENV values to values that will be passed as input to the schema if this var isn't present in the environment. For example:

  const schemas = {
    FRUIT: {
      schema: z.string().min(1),
      defaults: {
        production: "orange",
        development: "banana",
      },
    },
  };
  
  // FRUIT wll have value "banana".
  const { FRUIT } = parseEnv({ NODE_ENV: "development" }, schemas);
  
  // FRUIT wll have value "orange".
  const { FRUIT } = parseEnv({ NODE_ENV: "production" }, schemas);
  
  // FRUIT wll have value "fig".
  const { FRUIT } = parseEnv({ NODE_ENV: "production", FRUIT: "fig" }, schemas);
  
  // FRUIT wll have value "apple".
  const { FRUIT } = parseEnv({ FRUIT: "apple" }, schemas);
  
  // this will throw, since NODE_ENV doesn't match "production" or "development".
  const { FRUIT } = parseEnv({}, schemas);
  

  defaults accepts a special token as a key: _. This is like the default clause in a switch case — its value will be used if NODE_ENV doesn't match any other key in defaults.

  (As an aside, it is not recommended to use staging as a possible value for NODE_ENV. Your staging environment should be as similar to your production environment as possible, and NODE_ENV=production has special meaning for several tools and libraries. For example, npm install and yarn install by default won't install devDependencies if NODE_ENV=production; Express and React will also behave differently depending on whether NODE_ENV is production or not. Instead, your staging environment should also set NODE_ENV=production, and you should define your own env var(s) for any special configuration that's necessary for your staging environment.)

  Caveats aside, _ lets you express a few interesting scenarios:

  // one default for production, and one for all other environments, including
  // development and testing.
  { production: "prod default", _: "dev default" }
  
  // default for all non-production environments, but require the var to be
  // passed in for production.
  { production: undefined, _: "dev default" }
  
  // unconditional default. equivalent to adding `.default("some default")`
  // to the zod schema, but this might be more stylistically consistent with
  // your other specs if they use the `defaults` field.
  { _: "unconditional default" }
  

  Some testing tools like Jest set NODE_ENV to test, so you can also use defaults to provide default env vars for testing.

  parseEnv doesn't restrict or validate NODE_ENV to any particular values, but you can add NODE_ENV to your schemas like any other env var. For example, you could use NODE_ENV: z.enum(["production", "development", "test", "ci"]) to enforce that NODE_ENV is always defined and is one of those four expected values.

Extra schemas

znv exports a very small number of extra schemas for common env var types.

port()

port() is an alias for z.number().int().nonnegative().lte(65535).

deprecate()

deprecate() is an alias for z.undefined().transform(() => undefined as never). parseEnv will throw if a var using the deprecate() schema is passed in from the environment.

Coercion rules

znv tries to do as little work as possible to coerce env vars (which are always strings when they're present) to the input types of your schemas. If the env var doesn't look like the input type, znv will pass it to the validator as-is with the assumption that the validator will throw. For example, if your schema is z.number(), znv will test it against a numeric regex first, rather than unconditionally wrap it in Number() or parseFloat() (and thus coerce it to NaN).

By modifying as little as possible, znv tries to get out of Zod's way and let it do the heavy lifting of validation. This also lets us produce less confusing error messages: if you pass the string "banana" to your number schema, it should be able to say "you gave me 'banana' instead of a number!" rather than "you gave me NaN instead of a number!"

Coercions only happen at the top level of a schema. If you define an object with nested schemas, no coercions will be applied to the keys.

Some notable coercion mechanics:

• If your schema's input is a boolean, znv will coerce "true", "yes" and "1" to true, and "false", "no" and "0" to false. All other values will be passed through.

  Some CLI tool conventions dictate that a variable simply being present in the environment (even with no value, eg. setting MY_VALUE= with no right-hand side) should be interpreted as true. However, this convention doesn't seem to be in widespread use in Node, probably because it causes the var to evaluate to the empty string (which is falsy). znv demands a little more specificity by default, while still hedging a bit for some common true/false equivalents. If you want the "any defined value" behaviour, you can use z.string().optional().transform(v => v === undefined ? false : true).

• If your schema's input is an object or array (or record or tuple), znv will attempt to JSON.parse the input value if it's not undefined or the empty string.

  Remember, with great power comes great responsibility! If you're using an object or array schema to pass in dozens or hundreds of kilobytes of data as an env var, you may be doing something wrong. (Certain platforms also impose limits on environment variable length.)

• If your schema's input is a Date, znv will call new Date() with the input value. This has a number of pitfalls, since the Date() constructor is excessively forgiving. The value is passed in as a string, which means trying to pass a Unix epoch will yield unexpected results. (Epochs need to be passed in as number: new Date() with an epoch as a string will either give you invalid date or a completely nonsensical date.) You should only pass in ISO 8601 date strings, such as those returned by Date.prototype.toISOString(). Improved validation for Date schemas could be added in a future version.

• Zod defines "nullable" as distinct from "optional". If your schema is nullable, znv will coerce undefined to null. Generally it's preferred to simply use optional.

Comparison to other libraries

Envalid

Envalid is a nice library that inspired znv's API design. Envalid is written in TypeScript and performs some inference of the return value based on the validator schema you pass in, but won't do things like narrow enumerated types (str({ choices: ['a', 'b'] })) to a union of literals. Expressing defaults is more limited (you can't have different defaults for test and development environments, for example). Defaults are not passed through validators.

Envalid's validators are built-in and express a handful of types with limited options and no ability to perform postprocessing. For other use cases you have to write your own custom validators.

Envalid wraps its return value in a proxy, which can't be opted out of and has some surprising effects.

Joi

Joi is the Cadillac of schema validation libraries. Its default of coercing strings to the target type makes it easy to adopt for environment validation. Unfortunately, Joi is written in JavaScript and its type definitions support a very limited form of inference when they work at all.

Zod

Hey, what's Zod doing here? Doesn't znv use Zod?

If you just want to parse some values against a certain schema, you might not need znv. Just use Zod directly.

znv is best-suited for environment validation: it automatically wraps your Zod schemas in preprocessors that coerce env vars, which are always strings, into the appropriate type. This is different from Zod's built-in z.coerce, which is often too naive to be useful. For example, z.coerce.boolean() will parse "false" into true, since the string "false" is truthy in JavaScript. znv will coerce "false" into false, which is probably what you expect. Check the section on coercion rules for more information.

Complementary tooling

The eslint-plugin-node rule no-process-env is recommended to restrict usage of process.env outside of the module that parses your schema.

znv also works great with dotenv.

How do I pronounce znv?

If you usually pronounce "z" as "zed," then you could say "zenv." If you usually pronounce "z" as "zee," you could say "zee en vee."

Or do your own thing. I'm not the boss of you.