~jZdZddlmZddlZddlZddlmZddlZddlm Z ddl m Z ddl m Z mZmZmZejeZdZGd d ZdS) u!YouTube provider — list Max Amini's recent uploads, fetch transcripts. Two pieces, deliberately different in auth: 1. Listing recent videos uses the YouTube Data API v3 `search.list` endpoint with the user-supplied `YOUTUBE_API_KEY` (just an API key, no OAuth). Plain httpx — we don't pull in the heavy `google-api-python-client` since we only need this one GET. 2. Transcript extraction uses the third-party `youtube-transcript-api` library. It hits the public `timedtext` endpoint — no OAuth, no API key needed. (The official Data API `captions.download` endpoint requires OAuth + ownership of the video, which we do not have for Max Amini's channel. This is the standard approach used across the ecosystem.) The library is synchronous, so we run its calls via `asyncio.to_thread`. Per-video transcript failures are non-fatal: we log + skip that video so the stage can still surface the videos that DID have transcripts. The stage only fails if the initial Data API listing itself fails (network/retryable). ) annotationsN)Any)YouTubeTranscriptApi)settings)JsonExtractionErrorNonRetryableProviderErrorRetryableProviderError_build_semaphorez,https://www.googleapis.com/youtube/v3/searchc8eZdZddZddZddd Zdd Zdd ZdS) YouTubeClientreturnNonectjtj|_t tj|_t|_ dS)N)timeout) httpx AsyncClientrYOUTUBE_TIMEOUT_httpr YOUTUBE_MAX_CONCURRENCY_semr_ttaselfs ? N N N,Y;Ls;L;LMMM N N N N N N N N N N N N N N N N N N N N N N N N N N N N % S==FcMM(FHMJJ J S==+Ivx}MM M ^==??DD ^ ^ ^%&S&&S&S&STTZ] ] ^(*HHWb))  D,"11)<r~s&#""""" 777777  8 $ $?jjjjjjjjjjr