@fougere/adapter-env

The process environment as a source — MEILI_URL is Meili's url

// entities/Meili.ts
export default class Meili extends entity({ url: url(), key: text(), timeoutMs: number({ integer: true, default: 2000 }) }) {}

// fougere.config.ts
import '@fougere/adapter-env';

export default defineFougere({
  sources: {
    main: { type: 'sqlite', path: '.data/app.db' },
    env: { type: 'env', values: ['Meili'] },
  },
});

// services/Search.ts
export default class Search {
  constructor(private meili: Meili) {}
}

Search receives { url, key, timeoutMs }, judged at boot. A variable that is missing, or that does not hold what its field declares, refuses the boot and names it:

meili.key: Required — sources.env, MEILI_KEY

The names

A field is read from ENTITY_FIELD, in SNAKE_CASE and with no prefix: timeoutMs is MEILI_TIMEOUT_MS. The text is read as the field's shape says — a number, a boolean (true, false, 1, 0), JSON for an object or a list — and what does not read that way is refused by the field.

Under rows:, a segment of the name is the key, read from the END, since the fields are known: INDEX_BOOKS_URL is the row books, INDEX_PROD_AUTHORS_URL the row prod_authors. Two fields that end alike — url and booksUrl — would read one variable two ways, and are refused when the storage is made.

What it does not do

Nothing is written: a write is refused, naming the variable to change. And it is never the source that holds the rest — a type that is read-only cannot be, so env beside main leaves main the default.

One environment per deployment, not a pattern inside it: production and staging are two processes, each started with its own variables. A value that differs by file is read from a directory the config computes:

config: { type: 'file', path: `config/${process.env.APP_ENV}`, values: ['Catalog'] },