[aaltat]

Summary

Follow-up to #5326; the tests from #5358 guard this change.

The video size of a context was worked out twice: _record_video resolves recordVideo["size"] (size given, else viewport, else 1280x720), and then every route worked it out again for the context cache, each in its own way. Now the resolved size is the one that is cached.

 New Context / New Persistent Context / register_adopted_context
-  context_cache.add(id, _video_size(recordVideo, viewport))   # viewport, or None for adopted
+  if recordVideo: context_cache.add(id, recordVideo["size"])  # resolved by _record_video

 _record_video(record_video, viewport)
   size = _video_size(record_video, viewport)                  # the only caller left

• A context without a video caches no size. Only _embed_video reads the cache, and it returns before reading it when there is no video_path; Node sends video_path: null for a context that does not record.
• Types: the video dict Node returns ({video_path, contextUuid}) is ContextVideo in data_types.py, next to the other internal TypedDicts; _video_size and _record_video take RecordVideo and ViewportDimensions instead of Mapping[str, Any].

Evidence

• Before: new_context(viewport=...) without a video caches 800x600 for a context that never embeds a video.
  After: test_context_without_video_caches_no_video_size (written red first): no cache entry for New Context, New Persistent Context or an adopted context without a video.
• The test: guard the adopted context options and video sizes before unifying them #5358 tests pass unchanged: each route caches 800x600 from its viewport and 640x480 from an explicit size, and the embedded video uses it. utest/test_get_video_size.py passes {} instead of None (the narrower signature); its approved output is unchanged.
• inv utest: 857 passed. inv atest suites Video, Persistent State, Tracing Groups, 05 JS Tests: all passed.

Merge Danger

Door: two-way

Blast Radius: small

register_adopted_context now reads options["recordVideo"]["size"], which adopted_context_options always fills (its docstring says options come from there). Options built by hand with a recordVideo without size get a KeyError instead of a 1280x720 fallback. The Electron library passes the options of adopted_context_options unchanged.

🤖 Generated with Claude Code