# Streams

Phasis implements the full [WHATWG Streams Standard](https://streams.spec.whatwg.org/). Every class, every reader mode, every helper. The same code that works in Node and modern browsers works in Phasis.

The implementation is what `fetch` uses internally — `Response.body` is a real `ReadableStream`, `Request` accepts streaming bodies, and the `tee()` you need to clone a Response works.

## What's shipped

| Class | Status |
|---|---|
| `ReadableStream` | Full |
| `ReadableStreamDefaultController` | Full |
| `ReadableByteStreamController` | Full |
| `ReadableStreamDefaultReader` | Full |
| `ReadableStreamBYOBReader` | Full |
| `ReadableStreamBYOBRequest` | Full |
| `WritableStream` | Full |
| `WritableStreamDefaultController` | Full |
| `WritableStreamDefaultWriter` | Full |
| `TransformStream` | Full |
| `TransformStreamDefaultController` | Full |
| `ByteLengthQueuingStrategy` | Full |
| `CountQueuingStrategy` | Full |

Plus the methods you reach for: `pipeTo`, `pipeThrough`, `tee`, `getReader`, async iteration, `Symbol.asyncIterator`, BYOB read with `{ min }`.

## Producing

```js
// Pull-based — only generates values when read
const counter = new ReadableStream({
  start(controller) { this.n = 0; },
  pull(controller) {
    if (this.n >= 5) { controller.close(); return; }
    controller.enqueue(this.n++);
  },
});

for await (const v of counter) console.log(v);  // 0,1,2,3,4
```

```js
// Push-based — start enqueues everything eagerly
const small = new ReadableStream({
  start(c) {
    [10, 20, 30].forEach(n => c.enqueue(n));
    c.close();
  },
});
```

## Consuming

Three ways to read:

```js
// 1. Reader
const reader = stream.getReader();
while (true) {
  const { value, done } = await reader.read();
  if (done) break;
  console.log(value);
}
reader.releaseLock();

// 2. Async iteration (preferred)
for await (const chunk of stream) {
  console.log(chunk);
}

// 3. tee + parallel pipes
const [a, b] = stream.tee();
await Promise.all([a.pipeTo(sinkA), b.pipeTo(sinkB)]);
```

## Transforming

```js
const upper = new TransformStream({
  transform(chunk, controller) {
    controller.enqueue(chunk.toUpperCase());
  },
});

const result = new ReadableStream({
  start(c) { ['hi', 'there'].forEach(s => c.enqueue(s)); c.close(); }
}).pipeThrough(upper);

for await (const v of result) console.log(v);  // HI, THERE
```

## Byte streams (BYOB)

```js
const bytes = new ReadableStream({
  type: 'bytes',
  start(controller) {
    controller.enqueue(new Uint8Array([1, 2, 3, 4, 5, 6]));
    controller.close();
  },
});

const reader = bytes.getReader({ mode: 'byob' });
const buf = new Uint8Array(4);
const { value, done } = await reader.read(buf);
// value is a Uint8Array of length 4, sharing buffer with `buf`
```

`{ min: N }` is supported — `read(view, { min: 2 })` resolves only when at least 2 elements are available.

## Backpressure

`desiredSize` and `ready` work as the spec describes. Use a queuing strategy to give the producer a back-channel:

```js
const stream = new ReadableStream({
  pull(controller) {
    if (controller.desiredSize <= 0) return;  // back off
    controller.enqueue(nextChunk());
  },
}, new CountQueuingStrategy({ highWaterMark: 4 }));
```

## With fetch

```js
// Response body is a ReadableStream
const res = await fetch('https://example.com/big.txt');
let total = 0;
for await (const chunk of res.body) {
  total += chunk.byteLength;
}

// Request bodies can stream too
const upload = new ReadableStream({
  start(c) { c.enqueue(new TextEncoder().encode('hello')); c.close(); }
});
await fetch('https://example.com/upload', { method: 'POST', body: upload });
```

## Tee for caching

Common pattern: read the body twice.

```js
const res = await fetch(url);
const [forCache, forUse] = res.body.tee();
await Promise.all([
  forCache.pipeTo(cacheWritable),
  parseJson(forUse),
]);
```

`Response.clone()` does this for you and returns a fresh `Response`.

## Caveats

- **Synchronous-only producers.** A `pull()` that performs blocking PHP I/O will block the script. Use the fetch transport hook for HTTP; for files, prefer reading into memory upfront. `setTimeout` / `setInterval` are available (see [Event loop](/docs/api#event-loop)) but the engine uses **virtual time** by default — a `setTimeout(cb, 1000)` inside a `pull()` does not sleep PHP, it advances the loop clock. Real-time pacing needs the host to drive `tickEventLoop()` against a wall clock.

## See also

- [Web APIs (WPT)](/docs/compatibility/web-apis) — WPT pass-rate table covering Streams subtests.
- [Fetch transport](/docs/interop/fetch-transport) — how request and response bodies cross the PHP/JS boundary.
- [AbortController patterns](/docs/interop/abort-patterns) — cancelling in-flight reads.