[2.x] fix(realtime): reconstruct Pusher on iOS reconnect to bypass lives:2 budget - #4654

)

PR #4654 added a `visibilitychange` listener that forces a fresh
Pusher instance and refreshes the visible discussion list whenever
the tab has been hidden for >5s. That was needed on iOS Safari,
where backgrounding the page silently drops the WebSocket without
firing `close`. On every other platform the WebSocket survives tab
backgrounding fine — but the visibilitychange handler still fired,
causing an unnecessary `GET /api/discussions` request and full list
re-render every time the user switched away and back.

Gate the visibilitychange-triggered `forceReconnect()` on `isIOS()`
so the workaround only runs on the platform that actually needs it.
The `pageshow(persisted=true)` path stays unconditional — bfcache
restoration only fires on browsers that bfcached the page, and the
WebSocket was definitely torn down by then regardless of platform.

`isIOS()` is broader than the existing `isSafariMobile()` core
utility because all iOS browsers use WebKit and share the same
backgrounding pathology — iOS Chrome (`CriOS`) and iOS Firefox
(`FxiOS`) are excluded by `isSafariMobile()` but still need this
workaround.

Regression from #4654.

…h up missed events

WebKit suspends hidden pages on desktop Safari just like on iOS, so the
WebSocket dies silently (often without `close`) while pusher-js's
foreground-only activity timers cannot notice. The visibilitychange
recovery from flarum#4590/flarum#4654 was gated to isIOS() by flarum#4662 and never runs
there, and pusher-js's own reconnects perform no catch-up — so posts
that fired while the socket was down never appear, and the open
discussion silently stops live-updating (a >=2-post gap makes
PostStreamState.update()'s viewingEnd() guard refuse forever).

- Reconnect on visibility-restore when the connection is demonstrably
  unhealthy (state not 'connected', or no protocol frame within the
  activity window), in addition to the unconditional iOS path. Healthy
  desktop tabs keep receiving pongs while hidden, so plain tab switches
  still trigger no refetch (flarum#4662 intact).
- Catch up after every effective reconnect (pusher-internal or forced):
  refresh the discussion list and re-sync an open DiscussionPage,
  capturing viewingEnd() before the refetch grows postIds.
- Add PostStreamState.syncEnd() — update() without the 1-post drift
  bound — and capture end-ness in NewActivity before pushing event
  payloads, so a gap no longer permanently disables live-append.

Fixes flarum#4717.

Co-Authored-By: Claude Fable 5 <[email protected]>

…h up missed events

WebKit suspends hidden pages on desktop Safari just like on iOS, so the
WebSocket dies silently (often without `close`) while pusher-js's
foreground-only activity timers cannot notice. The visibilitychange
recovery from flarum#4590/flarum#4654 was gated to isIOS() by flarum#4662 and never runs
there, and pusher-js's own reconnects perform no catch-up — so posts
that fired while the socket was down never appear, and the open
discussion silently stops live-updating (a >=2-post gap makes
PostStreamState.update()'s viewingEnd() guard refuse forever).

- Reconnect on visibility-restore when the connection is demonstrably
  unhealthy (state not 'connected', or no protocol frame within the
  activity window), in addition to the unconditional iOS path. Healthy
  desktop tabs keep receiving pongs while hidden, so plain tab switches
  still trigger no refetch (flarum#4662 intact).
- Catch up after every effective reconnect (pusher-internal or forced):
  refresh the discussion list and re-sync an open DiscussionPage,
  capturing viewingEnd() before the refetch grows postIds.
- Add PostStreamState.syncEnd() — update() without the 1-post drift
  bound — and capture end-ness in NewActivity before pushing event
  payloads, so a gap no longer permanently disables live-append.

Fixes flarum#4717.

Co-Authored-By: Claude Opus 4.8 <[email protected]>

…h up missed events (#4718)

WebKit suspends hidden pages on desktop Safari just like on iOS, so the
WebSocket dies silently (often without `close`) while pusher-js's
foreground-only activity timers cannot notice. The visibilitychange
recovery from #4590/#4654 was gated to isIOS() by #4662 and never runs
there, and pusher-js's own reconnects perform no catch-up — so posts
that fired while the socket was down never appear, and the open
discussion silently stops live-updating (a >=2-post gap makes
PostStreamState.update()'s viewingEnd() guard refuse forever).

- Reconnect on visibility-restore when the connection is demonstrably
  unhealthy (state not 'connected', or no protocol frame within the
  activity window), in addition to the unconditional iOS path. Healthy
  desktop tabs keep receiving pongs while hidden, so plain tab switches
  still trigger no refetch (#4662 intact).
- Catch up after every effective reconnect (pusher-internal or forced):
  refresh the discussion list and re-sync an open DiscussionPage,
  capturing viewingEnd() before the refetch grows postIds.
- Add PostStreamState.syncEnd() — update() without the 1-post drift
  bound — and capture end-ness in NewActivity before pushing event
  payloads, so a gap no longer permanently disables live-append.

Fixes #4717.

Co-authored-by: Claude Opus 4.8 <[email protected]>