streaming.ts

import { ReadableStream, type Response } from "./_shims/mod.ts";
import { OpenAIError } from "./error.ts";

import { APIError } from "./error.ts";

type Bytes = string | ArrayBuffer | Uint8Array | Buffer | null | undefined;

export type ServerSentEvent = {
  event: string | null;
  data: string;
  raw: string[];
};

export class Stream<Item> implements AsyncIterable<Item> {
  controller: AbortController;

  constructor(
    private iterator: () => AsyncIterator<Item>,
    controller: AbortController,
  ) {
    this.controller = controller;
  }

  static fromSSEResponse<Item>(
    response: Response,
    controller: AbortController,
  ) {
    let consumed = false;

    async function* iterator(): AsyncIterator<Item, any, undefined> {
      if (consumed) {
        throw new Error(
          "Cannot iterate over a consumed stream, use `.tee()` to split the stream.",
        );
      }
      consumed = true;
      let done = false;
      try {
        for await (const sse of _iterSSEMessages(response, controller)) {
          if (done) continue;

          if (sse.data.startsWith("[DONE]")) {
            done = true;
            continue;
          }

          if (sse.event === null) {
            let data;

            try {
              data = JSON.parse(sse.data);
            } catch (e) {
              console.error(`Could not parse message into JSON:`, sse.data);
              console.error(`From chunk:`, sse.raw);
              throw e;
            }

            if (data && data.error) {
              throw new APIError(undefined, data.error, undefined, undefined);
            }

            yield data;
          } else {
            let data;
            try {
              data = JSON.parse(sse.data);
            } catch (e) {
              console.error(`Could not parse message into JSON:`, sse.data);
              console.error(`From chunk:`, sse.raw);
              throw e;
            }
            // TODO: Is this where the error should be thrown?
            if (sse.event == "error") {
              throw new APIError(
                undefined,
                data.error,
                data.message,
                undefined,
              );
            }
            yield { event: sse.event, data: data } as any;
          }
        }
        done = true;
      } catch (e) {
        // If the user calls `stream.controller.abort()`, we should exit without throwing.
        if (e instanceof Error && e.name === "AbortError") return;
        throw e;
      } finally {
        // If the user `break`s, abort the ongoing request.
        if (!done) controller.abort();
      }
    }

    return new Stream(iterator, controller);
  }

  /**
   * Generates a Stream from a newline-separated ReadableStream
   * where each item is a JSON value.
   */
  static fromReadableStream<Item>(
    readableStream: ReadableStream,
    controller: AbortController,
  ) {
    let consumed = false;

    async function* iterLines(): AsyncGenerator<string, void, unknown> {
      const lineDecoder = new LineDecoder();

      const iter = readableStreamAsyncIterable<Bytes>(readableStream);
      for await (const chunk of iter) {
        for (const line of lineDecoder.decode(chunk)) {
          yield line;
        }
      }

      for (const line of lineDecoder.flush()) {
        yield line;
      }
    }

    async function* iterator(): AsyncIterator<Item, any, undefined> {
      if (consumed) {
        throw new Error(
          "Cannot iterate over a consumed stream, use `.tee()` to split the stream.",
        );
      }
      consumed = true;
      let done = false;
      try {
        for await (const line of iterLines()) {
          if (done) continue;
          if (line) yield JSON.parse(line);
        }
        done = true;
      } catch (e) {
        // If the user calls `stream.controller.abort()`, we should exit without throwing.
        if (e instanceof Error && e.name === "AbortError") return;
        throw e;
      } finally {
        // If the user `break`s, abort the ongoing request.
        if (!done) controller.abort();
      }
    }

    return new Stream(iterator, controller);
  }

  [Symbol.asyncIterator](): AsyncIterator<Item> {
    return this.iterator();
  }

  /**
   * Splits the stream into two streams which can be
   * independently read from at different speeds.
   */
  tee(): [Stream<Item>, Stream<Item>] {
    const left: Array<Promise<IteratorResult<Item>>> = [];
    const right: Array<Promise<IteratorResult<Item>>> = [];
    const iterator = this.iterator();

    const teeIterator = (
      queue: Array<Promise<IteratorResult<Item>>>,
    ): AsyncIterator<Item> => {
      return {
        next: () => {
          if (queue.length === 0) {
            const result = iterator.next();
            left.push(result);
            right.push(result);
          }
          return queue.shift()!;
        },
      };
    };

    return [
      new Stream(() => teeIterator(left), this.controller),
      new Stream(() => teeIterator(right), this.controller),
    ];
  }

  /**
   * Converts this stream to a newline-separated ReadableStream of
   * JSON stringified values in the stream
   * which can be turned back into a Stream with `Stream.fromReadableStream()`.
   */
  toReadableStream(): ReadableStream {
    const self = this;
    let iter: AsyncIterator<Item>;
    const encoder = new TextEncoder();

    return new ReadableStream({
      async start() {
        iter = self[Symbol.asyncIterator]();
      },
      async pull(ctrl: any) {
        try {
          const { value, done } = await iter.next();
          if (done) return ctrl.close();

          const bytes = encoder.encode(JSON.stringify(value) + "\n");

          ctrl.enqueue(bytes);
        } catch (err) {
          ctrl.error(err);
        }
      },
      async cancel() {
        await iter.return?.();
      },
    });
  }
}

export async function* _iterSSEMessages(
  response: Response,
  controller: AbortController,
): AsyncGenerator<ServerSentEvent, void, unknown> {
  if (!response.body) {
    controller.abort();
    throw new OpenAIError(`Attempted to iterate over a response with no body`);
  }

  const sseDecoder = new SSEDecoder();
  const lineDecoder = new LineDecoder();

  const iter = readableStreamAsyncIterable<Bytes>(response.body);
  for await (const sseChunk of iterSSEChunks(iter)) {
    for (const line of lineDecoder.decode(sseChunk)) {
      const sse = sseDecoder.decode(line);
      if (sse) yield sse;
    }
  }

  for (const line of lineDecoder.flush()) {
    const sse = sseDecoder.decode(line);
    if (sse) yield sse;
  }
}

/**
 * Given an async iterable iterator, iterates over it and yields full
 * SSE chunks, i.e. yields when a double new-line is encountered.
 */
async function* iterSSEChunks(
  iterator: AsyncIterableIterator<Bytes>,
): AsyncGenerator<Uint8Array> {
  let data = new Uint8Array();

  for await (const chunk of iterator) {
    if (chunk == null) {
      continue;
    }

    const binaryChunk = chunk instanceof ArrayBuffer
      ? new Uint8Array(chunk)
      : typeof chunk === "string"
      ? new TextEncoder().encode(chunk)
      : chunk;

    let newData = new Uint8Array(data.length + binaryChunk.length);
    newData.set(data);
    newData.set(binaryChunk, data.length);
    data = newData;

    let patternIndex;
    while ((patternIndex = findDoubleNewlineIndex(data)) !== -1) {
      yield data.slice(0, patternIndex);
      data = data.slice(patternIndex);
    }
  }

  if (data.length > 0) {
    yield data;
  }
}

function findDoubleNewlineIndex(buffer: Uint8Array): number {
  // This function searches the buffer for the end patterns (\r\r, \n\n, \r\n\r\n)
  // and returns the index right after the first occurrence of any pattern,
  // or -1 if none of the patterns are found.
  const newline = 0x0a; // \n
  const carriage = 0x0d; // \r

  for (let i = 0; i < buffer.length - 2; i++) {
    if (buffer[i] === newline && buffer[i + 1] === newline) {
      // \n\n
      return i + 2;
    }
    if (buffer[i] === carriage && buffer[i + 1] === carriage) {
      // \r\r
      return i + 2;
    }
    if (
      buffer[i] === carriage &&
      buffer[i + 1] === newline &&
      i + 3 < buffer.length &&
      buffer[i + 2] === carriage &&
      buffer[i + 3] === newline
    ) {
      // \r\n\r\n
      return i + 4;
    }
  }

  return -1;
}

class SSEDecoder {
  private data: string[];
  private event: string | null;
  private chunks: string[];

  constructor() {
    this.event = null;
    this.data = [];
    this.chunks = [];
  }

  decode(line: string) {
    if (line.endsWith("\r")) {
      line = line.substring(0, line.length - 1);
    }

    if (!line) {
      // empty line and we didn't previously encounter any messages
      if (!this.event && !this.data.length) return null;

      const sse: ServerSentEvent = {
        event: this.event,
        data: this.data.join("\n"),
        raw: this.chunks,
      };

      this.event = null;
      this.data = [];
      this.chunks = [];

      return sse;
    }

    this.chunks.push(line);

    if (line.startsWith(":")) {
      return null;
    }

    let [fieldname, _, value] = partition(line, ":");

    if (value.startsWith(" ")) {
      value = value.substring(1);
    }

    if (fieldname === "event") {
      this.event = value;
    } else if (fieldname === "data") {
      this.data.push(value);
    }

    return null;
  }
}

/**
 * A re-implementation of httpx's `LineDecoder` in Python that handles incrementally
 * reading lines from text.
 *
 * https://github.com/encode/httpx/blob/920333ea98118e9cf617f246905d7b202510941c/httpx/_decoders.py#L258
 */
class LineDecoder {
  // prettier-ignore
  static NEWLINE_CHARS = new Set(["\n", "\r"]);
  static NEWLINE_REGEXP = /\r\n|[\n\r]/g;

  buffer: string[];
  trailingCR: boolean;
  textDecoder: any; // TextDecoder found in browsers; not typed to avoid pulling in either "dom" or "node" types.

  constructor() {
    this.buffer = [];
    this.trailingCR = false;
  }

  decode(chunk: Bytes): string[] {
    let text = this.decodeText(chunk);

    if (this.trailingCR) {
      text = "\r" + text;
      this.trailingCR = false;
    }
    if (text.endsWith("\r")) {
      this.trailingCR = true;
      text = text.slice(0, -1);
    }

    if (!text) {
      return [];
    }

    const trailingNewline = LineDecoder.NEWLINE_CHARS.has(
      text[text.length - 1] || "",
    );
    let lines = text.split(LineDecoder.NEWLINE_REGEXP);

    // if there is a trailing new line then the last entry will be an empty
    // string which we don't care about
    if (trailingNewline) {
      lines.pop();
    }

    if (lines.length === 1 && !trailingNewline) {
      this.buffer.push(lines[0]!);
      return [];
    }

    if (this.buffer.length > 0) {
      lines = [this.buffer.join("") + lines[0], ...lines.slice(1)];
      this.buffer = [];
    }

    if (!trailingNewline) {
      this.buffer = [lines.pop() || ""];
    }

    return lines;
  }

  decodeText(bytes: Bytes): string {
    if (bytes == null) return "";
    if (typeof bytes === "string") return bytes;

    // Node:
    if (typeof Buffer !== "undefined") {
      if (bytes instanceof Buffer) {
        return bytes.toString();
      }
      if (bytes instanceof Uint8Array) {
        return Buffer.from(bytes).toString();
      }

      throw new OpenAIError(
        `Unexpected: received non-Uint8Array (${bytes.constructor.name}) stream chunk in an environment with a global "Buffer" defined, which this library assumes to be Node. Please report this error.`,
      );
    }

    // Browser
    if (typeof TextDecoder !== "undefined") {
      if (bytes instanceof Uint8Array || bytes instanceof ArrayBuffer) {
        this.textDecoder ??= new TextDecoder("utf8");
        return this.textDecoder.decode(bytes);
      }

      throw new OpenAIError(
        `Unexpected: received non-Uint8Array/ArrayBuffer (${
          (bytes as any).constructor.name
        }) in a web platform. Please report this error.`,
      );
    }

    throw new OpenAIError(
      `Unexpected: neither Buffer nor TextDecoder are available as globals. Please report this error.`,
    );
  }

  flush(): string[] {
    if (!this.buffer.length && !this.trailingCR) {
      return [];
    }

    const lines = [this.buffer.join("")];
    this.buffer = [];
    this.trailingCR = false;
    return lines;
  }
}

/** This is an internal helper function that's just used for testing */
export function _decodeChunks(chunks: string[]): string[] {
  const decoder = new LineDecoder();
  const lines: string[] = [];
  for (const chunk of chunks) {
    lines.push(...decoder.decode(chunk));
  }

  return lines;
}

function partition(str: string, delimiter: string): [string, string, string] {
  const index = str.indexOf(delimiter);
  if (index !== -1) {
    return [
      str.substring(0, index),
      delimiter,
      str.substring(index + delimiter.length),
    ];
  }

  return [str, "", ""];
}

/**
 * Most browsers don't yet have async iterable support for ReadableStream,
 * and Node has a very different way of reading bytes from its "ReadableStream".
 *
 * This polyfill was pulled from https://github.com/MattiasBuelens/web-streams-polyfill/pull/122#issuecomment-1627354490
 */
export function readableStreamAsyncIterable<T>(
  stream: any,
): AsyncIterableIterator<T> {
  if (stream[Symbol.asyncIterator]) return stream;

  const reader = stream.getReader();
  return {
    async next() {
      try {
        const result = await reader.read();
        if (result?.done) reader.releaseLock(); // release lock when stream becomes closed
        return result;
      } catch (e) {
        reader.releaseLock(); // release lock when stream becomes errored
        throw e;
      }
    },
    async return() {
      const cancelPromise = reader.cancel();
      reader.releaseLock();
      await cancelPromise;
      return { done: true, value: undefined };
    },
    [Symbol.asyncIterator]() {
      return this;
    },
  };
}

declare let Buffer: any;

type Buffer = any;