yt-dlp: [prosiebensat1] Unable to extract clip id

I haven’t looked at the atv.at pages or (recently) any of the sites supported in the existing ProSiebenSat1 extractor. However, you can get metadata from the page under test in two ways (at least). The method InfoExtractor._search_ld_json() returns an info_dict like this:

{
  'description': 'Wir haben eine Million Euro versteckt. In einem Koffer. Irgendwo in Deutschland. Und jede:r von euch hat die Chance, die Million zu finden.',
  'duration': 283,
  'thumbnail': 'https://mim.p7s1.io/pis/mw/3c4cjq5FgV8h73OMk30hSmQ3ksnTujGLThfZrJTemL4yREOdH4xmFJqCsiW3b1ZT2Z_aOLVU57XF-vIBm_Gxs8m-84sW3ArA5w2ymcfYya-jeg/profile:ezone-teaser940x528?w=1200',
  'timestamp': 1699557600,
  'title': 'Lesung & Konzert'
}

The method InfoExtractor._search_nextjs_data() returns the page hydration JSON, where the .props.pageProps.info member is full of metadata.

The atv.at extractor logic looks for hydration JSON in the webpage containing video IDs and then makes the vas-v4.p7s1video.net request using a JWT with those IDs (content_ids). However for prosieben.de, the videos object was empty ({}) when the web client made the equivalent request, and the video ID that was the key can be extracted from the end of the URL:

    _VALID_URL = r'''(?x)
                    https?://
                        (?:www\.)?
                        prosieben(?:maxx)?\.(?:de|at|ch)/
                        serien/(?:[a-z\d-]+/)+videos/
                        (?P<slug>[a-z\d-]+)-(?P<id>[a-z]_[a-z\d]+)
                 '''

The vas-v4.p7s1video.net getsources API is geo-restricted. As you saw, this API host is already used in the existing yt-dlp ProSiebenSat1 extractor, but with different endpoints (eg geturls) and relying on other API hosts as removed in PR #5593. Presumably it’s geturls vs getsources that causes the old code to return the “eingestellt” video.

The problem mentioned above is that ProSiebenSat1BaseIE is the base class of Puls4IE, which apparently still works. Therefore a separate or intermediate base class is needed for ProSiebenSat1IE, ideally one that could be used as a base class to simplify ATVAtIE. If any of the other sites supported by the existing ProSiebenSat1IE extractor might still be handled correctly, it will have to be cloned so that the module has ProSiebenSat1IE for those sites and ProSiebenSat1v2IE for prosieben.de and any other broken sites (eg, sat1.de from OP).

Thanks for the hint with the HAR files!

Here are more findings…

The JWT key is not in the HAR file. It is calculated using JavaScript in In https://oasis-player-prod.p7s1.io/web/15.18.0/bootstrap/bootstrap.js

        t.getJWT = function(e, t) {
            var n = e.encryption_key
              , a = e.access_id
              , o = Math.round(Date.now() / 1e3)
              , s = r.__assign(r.__assign({}, t), {
                iat: o,
                nbf: o + -300,
                exp: o + 300
            });
            return (0,
            i.default)(s, n, {
                kid: a
            })
        }

The JWT token is signed using algorithm “HS256”. The secret can be found using the Chrome Developer Tools (“Sources” tab). I could set a breakpoint at the sign() method in “webpack:///node_modules/.pnpm/jwt-encode@1.0.1/node_modules/jwt-encode/src/index.js” to see the secrets.

The secrets are:

• access_id = “x_supernovatvc-de”
• encryption_key = “Ahsh3soxiemusijophoophiodeevujup” (used for SHA256-HMAC)

function encode (data) {
  const stringifiedData = CryptoJS.enc.Utf8.parse(JSON.stringify(data));
  return base64url(stringifiedData);
}
var header = {
  "alg": "HS256",
  "typ": "JWT",
  "kid": "x_supernovatvc-de"
};
var payload = {
  "content_ids": {
    "v_yd0czlu26gxi": {}
  },
  "secure_delivery": true,
  "iat": Date.now(),
  "nbf": Date.now()-300,
  "exp": Date.now()+300
};
var payload = encode(header)+"."+encode(data);
var jwt = payload+"."+base64url(CryptoJS.HmacSHA256(payload, "Ahsh3soxiemusijophoophiodeevujup"))
console.log(jwt);

(It should be tested if the encryption key is individual for each video or not)

Unfortunately, I don’t have experience with Python or the yt-dlp code, so I have no idea how to implement it…