Add a plugin API for reading HTTP response bodies

Plugins read bodies by response id through ctx.httpResponse.body()
instead of opening HttpResponse.bodyPath themselves. The accessors are
named after fetch's, minus the single-use semantics, since the bytes are
durable and re-reading should work.

Underneath is a chunked pull over the existing plugin protocol, so the
host can move bodies off the filesystem without plugins noticing.
text() now decodes with the response's charset rather than assuming
UTF-8, and the buffering accessors refuse past 32 MiB and point at
chunks().

packages/plugin-runtime/src/responseBody.ts

@@ -0,0 +1,151 @@
+import type { HttpResponseBody, ReadHttpResponseBodyOptions } from "@yaakapp/api";
+
+/** Bytes pulled from the host per round trip, when the caller doesn't say. */
+const DEFAULT_CHUNK_SIZE = 1024 * 1024;
+
+/**
+ * The most a plugin buffers by default.
+ *
+ * Reading a body used to be unbounded, so any ceiling is an improvement; this
+ * one is set well above what an API returns and well below what makes the
+ * plugin runtime fall over. `chunks()` has no ceiling, and any caller that
+ * really wants the whole thing can raise `maxBytes`.
+ */
+const DEFAULT_MAX_BYTES = 32 * 1024 * 1024;
+
+/** Fetch one window of body bytes from the host. */
+export type ReadResponseBodyChunk = (offset: number, length: number) => Promise<Uint8Array>;
+
+export interface ResponseBodyInfo {
+  responseId: string;
+  contentLength: number;
+  contentType: string | null;
+}
+
+export function createResponseBody(
+  info: ResponseBodyInfo,
+  readChunk: ReadResponseBodyChunk,
+): HttpResponseBody {
+  const { responseId, contentLength, contentType } = info;
+
+  async function* chunks(
+    options?: Pick<ReadHttpResponseBodyOptions, "chunkSize">,
+  ): AsyncIterable<Uint8Array> {
+    const chunkSize = Math.max(1, Math.floor(options?.chunkSize ?? DEFAULT_CHUNK_SIZE));
+    let offset = 0;
+    // Bounded by the length the host reported, but a short read still ends it:
+    // the body may have been rewritten between the two calls.
+    while (offset < contentLength) {
+      const chunk = await readChunk(offset, Math.min(chunkSize, contentLength - offset));
+      if (chunk.byteLength === 0) return;
+      yield chunk;
+      offset += chunk.byteLength;
+    }
+  }
+
+  async function readAll(accessor: string, options?: ReadHttpResponseBodyOptions) {
+    const maxBytes = options?.maxBytes ?? DEFAULT_MAX_BYTES;
+    refuseIfTooBig(accessor, contentLength, maxBytes);
+
+    const parts: Uint8Array[] = [];
+    let total = 0;
+    for await (const chunk of chunks(options)) {
+      total += chunk.byteLength;
+      // The size the host reported is a claim about a moment ago, so check the
+      // bytes actually arriving too.
+      refuseIfTooBig(accessor, total, maxBytes);
+      parts.push(chunk);
+    }
+
+    const bytes = new Uint8Array(total);
+    let offset = 0;
+    for (const part of parts) {
+      bytes.set(part, offset);
+      offset += part.byteLength;
+    }
+    return bytes;
+  }
+
+  return {
+    responseId,
+    contentLength,
+    contentType,
+    chunks,
+    async arrayBuffer(options) {
+      const bytes = await readAll("arrayBuffer", options);
+      return bytes.buffer as ArrayBuffer;
+    },
+    async text(options) {
+      return decodeBody(await readAll("text", options), contentType);
+    },
+    async json<T>(options?: ReadHttpResponseBodyOptions) {
+      return JSON.parse(decodeBody(await readAll("json", options), contentType)) as T;
+    },
+  };
+}
+
+function refuseIfTooBig(accessor: string, bytes: number, maxBytes: number) {
+  if (bytes <= maxBytes) return;
+  throw new Error(
+    `Response body is ${formatBytes(bytes)}, over the ${formatBytes(maxBytes)} limit for ` +
+      `${accessor}(). Read it with chunks() instead, or pass a larger maxBytes.`,
+  );
+}
+
+/**
+ * Decode using the charset the response declared.
+ *
+ * Assuming UTF-8 mangles every response that isn't, and the header is right
+ * there. An unknown label is the one case worth guessing on, since the
+ * alternative is refusing to read a body we can very likely still read.
+ */
+function decodeBody(bytes: Uint8Array, contentType: string | null): string {
+  const charset = parseCharset(contentType);
+  if (charset != null) {
+    try {
+      return new TextDecoder(charset).decode(bytes);
+    } catch {
+      // Not a label this runtime knows.
+    }
+  }
+  // TextDecoder drops a leading BOM on its own.
+  return new TextDecoder("utf-8").decode(bytes);
+}
+
+function parseCharset(contentType: string | null): string | null {
+  const match = contentType?.match(/;\s*charset\s*=\s*"?([^";]+)"?/i);
+  return match?.[1]?.trim() || null;
+}
+
+function formatBytes(bytes: number): string {
+  if (bytes === Infinity) return "unlimited";
+  if (bytes < 1024) return `${bytes} B`;
+  const units = ["KB", "MB", "GB"];
+  let value = bytes / 1024;
+  let unit = 0;
+  while (value >= 1024 && unit < units.length - 1) {
+    value /= 1024;
+    unit++;
+  }
+  return `${value.toFixed(1)} ${units[unit]}`;
+}
+
+/**
+ * Decode a chunk that arrived as base64.
+ *
+ * The desktop transport is a WebSocket carrying JSON text frames, so bytes
+ * have to be spelled out. A host that can pass an ArrayBuffer along skips this.
+ */
+export function decodeBase64Chunk(data: string): Uint8Array {
+  if (typeof Buffer !== "undefined") {
+    const buf = Buffer.from(data, "base64");
+    return new Uint8Array(buf.buffer, buf.byteOffset, buf.byteLength);
+  }
+
+  const binary = atob(data);
+  const bytes = new Uint8Array(binary.length);
+  for (let i = 0; i < binary.length; i++) {
+    bytes[i] = binary.charCodeAt(i);
+  }
+  return bytes;
+}