cloudflare/miniflare v2.0.0-rc.5 on GitHub

This update is mostly focused on improving mounts for service bindings. It also matches some other core behaviour of the Workers runtime, like blocking asynchronous I/O outside request handlers, and limiting request recursion depth.

Breaking Changes

instanceof checks on primitives such as Object, Array, Promise, etc. from outside the Miniflare sandbox no longer succeed by default. This was initially added to make it easier to run Rust workers in Miniflare, as wasm-bindgen frequently generates this code. However, the implementation caused Object prototype/constructor checks to fail in JavaScript. The previous behaviour can be restored with the --proxy-primitive/miniflare.proxy_primitive_instanceof/proxyPrimitiveInstanceOf CLI/wrangler.toml/API option. This should be enabled when developing Rust workers. See this comment for more details. Closes issues #109, #137, #141 and cloudflare/wrangler2#91. Thanks @EmilianoSanchez, @lukaszczerpak, @SirJosh3917 and @johnhenry. NOTE: this change will be reverted in the next release. I've thought of an alternative solution that doesn't need this flag.
Mount paths defined in wrangler.toml are now resolved relative to the directory containing the wrangler.toml file
Mount names must match the name field in wrangler.toml if its defined

Features

To match the behaviour of the Workers runtime, some functionality, such as asynchronous I/O (fetch, Cache API, KV), timeouts (setTimeout, setInterval), and generating cryptographically-secure random values (crypto.getRandomValues, crypto.subtle.generateKey), can now only be performed while handling a request.
- This behaviour can be disabled by setting the --global-async-io/globalAsyncIO, --global-timers/globalTimers and --global-random/globalRandom options respectively, which may be useful for tests or libraries that need async I/O for setup during local development. Note the Miniflare Jest environment automatically enables these options.
- KV namespaces and caches returned from Miniflare#getKVNamespace() and getCaches() are unaffected by this change, so they can still be used in tests without setting any additional options.
Adds highly experimental support for service bindings. This is primarily meant for internal testing, and users outside the beta can't deploy workers using this feature yet, but feel free to play around with them locally and let us know what you think in the Cloudflare Workers Discord server.
To enable these, mount your service (so Miniflare knows where to find it) then add the binding. Note the bound service name must match the mounted name:
```
$ miniflare --mount auth=./auth --service AUTH_SERVICE=auth # or -S
```
```
# wrangler.toml
experimental_services = [
  # Note environment is currently ignored
  { name = "AUTH_SERVICE", service = "auth", environment = "production" }
]

[miniflare.mounts]
auth = "./auth"
```
```
const mf = new Miniflare({
  mounts: { auth: "./auth" },
  serviceBindings: { AUTH_SERVICE: "auth" },
});
```
...then to use the service binding:
```
export default {
  async fetch(request, env, ctx) {
    const res = await env.AUTH_SERVICE.fetch("...");
    // ...
  },
};
```
If ./auth/wrangler.toml contains its own service bindings, those services must also be mounted in the root worker (i.e. in wrangler.toml not ./auth/wrangler.toml). Nested mounts are not supported.
To match the behaviour of the Workers runtime, Miniflare now enforces recursion depth limits. Durable Object fetches can recurse up to 16 times, and service bindings can recurse up to 32 times. This means if a Durable Object fetch triggers another Durable Object fetch, and so on 16 times, an error will be thrown.
The root worker is now routable using route/routes as with mounts if it has a name set. This means if it has more specific routes than other mounts, they'll be used instead. The root worker is always used as a fallback if no mounts' routes match.
Mounted workers can now access Durable Objects defined in other mounts or the root worker (assuming a name is set) using the script_name option.
Allow the subrequest limit to be customised using the MINIFLARE_SUBREQUEST_LIMIT environment variable. Setting this to a negative number disables the limit. Setting this to 0 disables subrequests. The majority of users should never need to touch this, hence it's configured via an environment variable, not a CLI option. This also makes the implementation much simpler. :slight_smile: Closes issue #132, thanks @DanielAGW.

Fixes

Allow env_path to be set in wrangler.toml
WebSocket client fetches now contribute to the subrequest limit
Errors raised when reloading (e.g. missing Durable Object classes) in response to mount file changes are now logged, and don't crash
Fixed issue where deleting [miniflare.mounts] in wrangler.toml wouldn't unmount everything