- Migrate user profile fields from `legacy{}` to `core{}` / `avatar{}` / `location{}` (with legacy fallback for older response shapes). Affects `user`, `whoami`, `status`, `followers`, `following`.
- Fix `user-posts` empty results: add missing `includePromotedContent` variable and update `instructions` path from `timeline_v2` to `timeline`.
- Switch `followers` / `following` to POST (Twitter changed the requirement).
- Refresh 16 stale `FALLBACK_QUERY_IDS` from fa0311/twitter-openapi.
- Drop `legacy{}` early-return in `parse_user_result`; key existence on `rest_id` so followers/following keep working when Twitter fully drops legacy.
- Add unit tests for the new core/avatar/location fallback chain and rest_id/typename guards.
- Early return when no entityRanges (skip unnecessary processing)
- Escape [ and ] in link labels to prevent nested bracket issues
- Encode ) in URLs as %29 to prevent malformed markdown links (e.g. Wikipedia)
- Add 3 new test cases for the above edge cases
Extract visibility metadata from TweetWithVisibilityResults wrapper
before unwrapping. Adds is_subscriber_only field to Tweet model,
with full serialization roundtrip and test coverage.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Twitter long tweets (>280 chars) store full text in
note_tweet.note_tweet_results.result.text rather than legacy.full_text.
The parser now prioritizes note_tweet text when available.
Split the monolithic client.py (1341 lines) into three focused modules:
- graphql.py (~200 lines): queryId resolution, URL building, JS bundle
scanning, feature flag management
- parser.py (~270 lines): Tweet/User/Media/Article parsing, utility functions
(_deep_get, _parse_int, _extract_cursor, _extract_media)
- client.py (~700 lines): TwitterClient class with HTTP engine, anti-detection,
session management, and all public API methods
Backward compatibility: client.py re-exports all previously public symbols
so existing test imports work without modification. 88/88 tests pass.
