TubeFeed versionshistorik - 3 versioner
Var försiktig med äldre versioner! Dessa versioner visas för test och referensändamål.Du bör alltid använda den senaste versionen av ett tillägg.
Senaste version
Version 0.6.13
Släppt 14 jun 2026 - 108,36 kBFungerar med firefox 140.0 och senarev0.6.13
Scroll-to-seen- Fixed scroll-to-seen: videos you scroll past are marked "seen" and now actually drop out of the toolbar badge count, as intended. Previously the badge ignored the seen state, so scrolling through the list never reduced the count. Seen videos still stay visible and undimmed in the lists — they're just no longer counted as new.
- Added a Settings → General toggle, "Mark videos as seen when scrolled into view" (on by default). Turn it off to have the toolbar badge count every unwatched video regardless of scrolling.
How videos are pulled (efficiency & reliability)- Duration backfill now uses a single batched YouTube Data API call when an API key is configured, instead of fetching each video's watch page individually — far faster and far less bandwidth for API-key users. Without a key, the watch-page method is unchanged.
- Fixed a duration-repair queue problem: livestreams, premieres, and upcoming videos never expose a duration, and previously they could monopolize each poll's repair budget indefinitely, starving normal videos. Repair now gives up on a video after several failed attempts so it always reaches repairable ones. The Settings → Data "Refresh all durations" button still retries everything on demand.
- Polls now write only new or changed videos to the database instead of rewriting every video every cycle, reducing disk churn — especially noticeable with many channels.
- The channel-page scrape now treats a bare HTTP 403 ("automated queries") the same as other rate-limit signals, backing off for 2 hours instead of repeatedly hitting a flagged session.
Källkod släppt under MIT-licens
Äldre versioner
Version 0.6.12
Släppt 13 jun 2026 - 108,14 kBFungerar med firefox 140.0 och senarev0.6.12
How videos are pulled (efficiency & reliability)- Duration backfill now uses a single batched YouTube Data API call when an API key is configured, instead of fetching each video's watch page individually — far faster and far less bandwidth for API-key users. Without a key, the watch-page method is unchanged.
- Fixed a duration-repair queue problem: livestreams, premieres, and upcoming videos never expose a duration, and previously they could monopolize each poll's repair budget indefinitely, starving normal videos. Repair now gives up on a video after several failed attempts so it always reaches repairable ones. The Settings → Data "Refresh all durations" button still retries everything on demand.
- Polls now write only new or changed videos to the database instead of rewriting every video every cycle, reducing disk churn — especially noticeable with many channels.
- The channel-page scrape now treats a bare HTTP 403 ("automated queries") the same as other rate-limit signals, backing off for 2 hours instead of repeatedly hitting a flagged session.
Källkod släppt under MIT-licens
Version 0.6.10
Släppt 12 jun 2026 - 107,37 kBFungerar med firefox 140.0 och senarev0.6.10
Reliability fixes- Fixed the background poller not picking up channel list changes: adding or removing a channel in the reader now immediately notifies the background worker to drop its in-memory channel cache. Previously, while the worker stayed alive (which the reader tab's heartbeat keeps it), newly added channels were not background-polled and removed channels kept being fetched until the worker restarted.
- Channel cache invalidation now happens after the database transaction commits, closing a window where a concurrent read could re-cache pre-write data and serve a stale channel list indefinitely.
- Each channel poll now has a 60-second overall ceiling, so one hung fetch chain (RSS → scrape → API) can no longer stall the entire sequential poll loop.
- Background poll trigger errors (e.g. a temporarily unavailable database) are now caught and logged instead of producing an unhandled rejection every minute.
- Read-state tombstones are now pruned when their videos age past the re-fetch horizon, keeping the
storage.localtombstone set bounded instead of growing forever. - Fixed a Settings race where an external settings refresh during the 500 ms input debounce could silently discard a number field edit before it was saved.
- Adding a channel now validates the background script's resolve response instead of assuming it is well-formed.
Performance & internals- Unread counts (sidebar and toolbar badge) are computed from a single bulk read instead of a row-by-row cursor walk.
- Video list mark-read/seen handlers have stable identities, so list items no longer re-render on unrelated state changes.
- Consolidated four duplicated fetch-timeout implementations into one shared helper that also clears its abort timer once the request settles.
- Removed two phantom IndexedDB index definitions that were never used (one of them could never work: booleans are not valid IndexedDB keys). Existing profiles are unaffected.
Accessibility- The channel remove (×) control is now keyboard-operable (Tab + Enter/Space) and announces itself as "Remove <channel>" to screen readers.
- The watched/unwatched toggle on video thumbnails now has a proper screen-reader label.
Källkod släppt under MIT-licens