JTPlay

A audio/video player for jailbroken PS5s where the majority of the code was produced by claude.ai

JavaScript plugins

Besides UPnP/DLNA servers and the local filesystem, media sources can be added with JavaScript. Every *.js file in the plugin directory (--plugins DIR, default <assets>/../plugins) is evaluated by QuickJS and becomes one provider; the script returns an object implementing the same interface as browse::Provider and browse::Source:

return {
  name: "My provider",
  discover() {
    return [
      {
        name: "My source",         // row title
        detail: "somewhere",       // row subtitle (optional)
        icon: "📻",                // emoji glyph (optional)
        root: "/",                 // root folder id (optional)
        browse(id) {               // list the children of 'id'
          return [
            {
              id: "bbc1",                        // optional; defaults to uri
              type: "audio",                     // folder|audio|video|image
              name: "A stream",                  // required
              description: "Some station",       // optional
              image: "https://example.com/a.jpg",// optional
              uri: "https://example.com/stream",
            },
          ];
        },
      },
    ];
  },
};

The script body runs inside a function (like a CommonJS module), so it ends with return rather than an export, and top-level const/let stay private to the plugin. In browse(), this is the source object, so a plugin can keep per-source state on it.

Each script gets its own QuickJS runtime with the standard objects (JSON, Math, Date, RegExp, ...), console.log()/warn()/error() writing to the RmlUi log, and the http object below. There is no event loop: discover() and browse() must return their result synchronously (an async function returns a promise and is reported as an error), and a call that spends more than 10 seconds executing is interrupted.

An entry has six fields and no more: id, type, name, description, image and uri (see browse::Entry in src/browse/source.h). Only name is required; a folder entry is browsed by handing its id back to browse(), anything else is opened by its uri. A source that serves signed or otherwise short-lived URIs can leave uri out and add a resolve(id, entry) function returning a fresh one (as a string, or as an entry object with a uri) at the moment playback starts.

See plugins/sr.js (Sveriges Radio live channels) for a working example.

Fetching over the network

Plugins talk to the network through http, backed by libcurl, so https, redirects and compressed responses work. The calls block (plugins already run on the browsing worker thread) and return the response directly:

const res = http.get("https://api.example.com/things", {
  headers: { Accept: "application/json" },
  timeout: 20,        // seconds; default 15, capped at 60
});
if (!res.ok) throw new Error("HTTP " + res.status);
const things = JSON.parse(res.body);

http.post(url, body, options);   // options as above
http.request({ method, url, headers, body, timeout, redirect, binary });

A response is { ok, status, url, headers, body }, where url is the final URL after redirects and headers keys are lower-cased. Following redirects can be turned off with redirect: false, and binary: true returns the body as an ArrayBuffer instead of a string (a string body is decoded as UTF-8).

An HTTP error status is not an exception: check res.ok. Transport failures (DNS, TLS, timeout) throw an Error, so wrap calls in try/catch if a source should survive an unreachable server. Only http and https URLs are permitted, and a response over 8 MiB is refused. Time spent waiting on a request does not count towards the 10-second watchdog.

If the build's libcurl has no default CA bundle, point it at one with the SSL_CERT_FILE (or CURL_CA_BUNDLE) environment variable, otherwise https requests fail with a certificate error.

Building

Besides SDL2, RmlUi, FFmpeg and tinyxml2, the build needs QuickJS and libcurl.