// The WebSocket shim tag, put into guest app documents so it runs before any app
// script: as the first child of <head>, or before the first tag when there is no
// <head> tag (the parser then opens head for it). Comments, doctypes and the contents
// of script and style elements are skipped while looking. The document's bytes are
// otherwise untouched; the scan reads them as latin1, one character per byte, so any
// ASCII-compatible charset survives.
import { WS_SHIM_PATH } from "../shared/vurls";

const TAG = new TextEncoder().encode(`<script src="${WS_SHIM_PATH}"></script>`);
const RAW_TEXT = new Set(["script", "style"]);

/** Where the shim tag goes in `html`. */
export function shimOffset(html: string): number {
  let firstTag = -1;
  let at = 0;
  while ((at = html.indexOf("<", at)) !== -1) {
    if (html.startsWith("<!--", at)) {
      const end = html.indexOf("-->", at + 4);
      if (end === -1) break;
      at = end + 3;
      continue;
    }
    const m = /^<\/?([A-Za-z][^\s/>]*)/.exec(html.slice(at, at + 64));
    if (!m) {
      // <!doctype ...>, <?...>, or a stray "<" in text.
      at += 1;
      continue;
    }
    const close = html.indexOf(">", at);
    if (close === -1) break;
    if (firstTag === -1) firstTag = at;
    const name = m[1]!.toLowerCase();
    const isEnd = html[at + 1] === "/";
    if (!isEnd && name === "head") return close + 1;
    if (!isEnd && RAW_TEXT.has(name)) {
      const end = html.toLowerCase().indexOf(`</${name}`, close);
      at = end === -1 ? html.length : end;
      continue;
    }
    at = close + 1;
  }
  return firstTag === -1 ? html.length : firstTag;
}

export function injectShim(body: Uint8Array): Uint8Array {
  const at = shimOffset(new TextDecoder("latin1").decode(body));
  const out = new Uint8Array(body.length + TAG.length);
  out.set(body.subarray(0, at));
  out.set(TAG, at);
  out.set(body.subarray(at), at + TAG.length);
  return out;
}

/** The HTML patterns of the MIME Sniffing standard's rules for an unknown type: a body
 *  with no content-type that starts like this is rendered as HTML. */
const SNIFFED_HTML = /^[\t\n\f\r ]*<(?:!doctype html|html|head|script|iframe|h1|div|font|table|a|style|title|b|body|br|p|!--)[ >]/i;

/** Only a document runs scripts, and only an uncompressed HTML body can be edited. */
export function wantsShim(mode: RequestMode, headers: Headers, body: Uint8Array): boolean {
  const encoding = headers.get("content-encoding");
  if (mode !== "navigate" || (encoding !== null && encoding.toLowerCase() !== "identity")) return false;
  const type = headers.get("content-type");
  if (type !== null) return /^text\/html\b/i.test(type);
  return SNIFFED_HTML.test(new TextDecoder("latin1").decode(body.subarray(0, 512)));
}