Code review fixes: faster start, no update wait before playback, Deno on Android

[firsttris]

Implements the verified points of a code review of the add-on. Every point was checked against the code, and measured or reproduced where possible. Each change is its own commit.

Faster playback start

• yt-dlp stable now installs the yt-dlp.tar.gz release asset. It is built with make all and ships lazy_extractors.py; the tag archive does not, so every start imported all ~1800 extractor modules. Checked against the real 2026.08.19 asset. Releases without the asset fall back to the tag archive on HTTP 404.
• Runtime updates no longer run before playback. An installed yt-dlp or Deno is used without network access. The update check runs after setResolvedUrl, also after a failed resolve, and the new version is used from the next playback. A first install still happens up front.
• reuselanguageinvoker keeps the interpreter between calls, so yt-dlp is imported only once. A yt-dlp version switch still takes effect immediately: activate_runtime cleans sys.path and unloads a stale yt_dlp. This was tested by switching between two real yt-dlp trees in one interpreter.
• Subtitles download in parallel. The DASH range probe only fetches the bytes that are missing.

Measurement (x86)	Before	After
import yt_dlp + YoutubeDL(), cold cache	1.8 s	0.6 s
same, warm cache	0.45 s	0.22 s

Bugs

• Deno on Android: Python reports Linux/aarch64 there, so the glibc Linux build was downloaded (100+ MB unpacked) even though it cannot run. Android is now detected, the leftover download is removed, and the settings actions explain why.
• Downloads in RAM: the Deno zip and the unpacked binary were held fully in memory. Downloads now stream to disk.
• Opening the settings could wait up to 20 s on GitHub, because the yt-dlp status always looked up the latest release.
• An empty maxresolution setting raised ValueError. The dead deno_enabled setting is removed.
• The playlist start video was resolved without a progress dialog. The docs claimed playlists are never downloaded, which is wrong and is now fixed.

Why the strptime patch is needed (#177)

It is not a thread race. Kodi runs every add-on call in a sub-interpreter and destroys it afterwards. Up to Python 3.12, the C datetime module caches _strptime in a static variable shared by all interpreters. After the interpreter that filled the cache is gone, every strptime() call raises 'NoneType' object is not callable (bpo-27400, fixed in 3.13).

This was reproduced with _testcapi.run_in_subinterp on 3.11 and 3.12. The warmup suggested in #177 makes things worse, because it poisons the next call. The patch is now idempotent, and a new test replays two Kodi-style calls; it runs on CI's Python 3.12.

Other changes

• The release zip no longer contains tests, docs or dev config. It shrinks from over 1 MB to 70 KB.
• service.py is now a main() function that returns instead of calling exit(), and it has smoke tests against fake Kodi modules.
• The repeated blocks in select_playback_source were moved into helpers without changing the order of decisions. Two branches that had no tests now have them, and those tests also pass on the old code.

Rejected after checking

• Caching parse_options: a call takes 2 ms.
• Splitting the DASH timeout constant: it would change behavior and needs a product decision.
• Unifying requests and urllib: no benefit.
• Coverage only in CI: it adds about 1 s locally.

Testing

• 341 tests: 340 pass and 1 is skipped (that test is also skipped on master, which has 301). Run on Python 3.13 and on 3.12 with coverage, as in CI.
• Not tested on a real Kodi device. Before merging, please try reuselanguageinvoker and the post-playback update on Kodi 21. Check playback, a playlist, the settings actions and a yt-dlp version switch without restarting Kodi.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MpqNbBhrj8pan7qLdtFbdy

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 90.60052% with 36 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.48%. Comparing base (b5e27a0) to head (895ab5b).
⚠️ Report is 2 commits behind head on master.

Files with missing lines	Patch %	Lines
service.py	73.40%	25 Missing ⚠️
core/deno_manager.py	90.90%	7 Missing ⚠️
core/playback_selection.py	96.96%	1 Missing ⚠️
core/runtime/playback.py	95.65%	1 Missing ⚠️
core/service_runtime.py	80.00%	1 Missing ⚠️
core/ytdlp_manager.py	98.38%	1 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #194      +/-   ##
==========================================
+ Coverage   75.88%   84.48%   +8.60%     
==========================================
  Files          14       14              
  Lines        2380     2591     +211     
==========================================
+ Hits         1806     2189     +383     
+ Misses        574      402     -172     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.