@ai-ecoverse/slicc-bios 1.51.7 → 1.51.8

package/README.md

@@ -617,7 +617,7 @@ Limits:
 
 A nudge first sends the old worker `{ streams: 'mine' }` and waits for its answer (2 s at most), which makes sure it is running. It then sends the waiting worker `{ activate: 'now' }`. `sw.js` takes that message only from a window client and only with nothing else in it, and does nothing but call `self.skipWaiting()` again. Activation still waits until the old worker has no work, so a half-read download keeps holding it, and *Seven updated* still shows after 10 s.
 
-**A slow live stream doesn't hold it (#710).** Chromium leaves an updated worker `installed` while the old one streams a slow, live response, such as a dev server's live-reload SSE that sends an event a second, or a long poll. A download the tab keeps pulling doesn't hold it: Chromium activates the update, and the old worker finishes the download. So the old worker acts only once a new worker has been waiting 5 s, which an update with nothing holding it never does (with the nudge, it activates in about 3 s). It notes when it first sees the waiting worker, on that worker's `statechange`, when a stream opens and when a tab asks for its streams. Then it ends every stream it serves that is at least 5 s old, so a load during an update is never touched. An event stream (`text/event-stream`) ends cleanly, so an `EventSource` reconnects by itself, to the new worker. A `fetch()` reader of an event stream sees the same clean end, as when its server closes the connection: the worker can't tell it from an `EventSource`, and an error would close an `EventSource` for good, since Chromium reports it to the page as an abort. Anything else ends with an error (`sw: seven updated, so <path> stopped`), so a reader knows it didn't get everything. The cut ends every stream 5 s old, in every tab, so a download that wasn't holding anything fails too if it runs while another stream holds an update. It can be fetched again. Nothing happens while a worker is only installing, so an install that fails cuts nothing. The update is active about 5 s after it's `installed`, before *Seven updated* could show. A stream that nobody reads any more is the exception (#699): Chromium lets the old worker go only once the reader takes more, so a half-read download still holds the update, and *Seven updated* names it after 10 s.
+**A slow live stream doesn't hold it (#710).** Chromium leaves an updated worker `installed` while the old one streams a slow, live response, such as a dev server's live-reload SSE that sends an event a second, or a long poll. A download the tab keeps pulling doesn't hold it: Chromium activates the update, and the old worker finishes the download. So the old worker acts only once a new worker has been waiting 5 s, which an update with nothing holding it never does (with the nudge, it activates in about 3 s). It notes when it first sees the waiting worker, on that worker's `statechange`, when a stream opens and when a tab asks for its streams. Then it ends every stream it serves that is at least 5 s old, so a load shorter than that is never touched. An event stream (`text/event-stream`) ends cleanly, so an `EventSource` reconnects by itself, to the new worker. A `fetch()` reader of an event stream sees the same clean end, as when its server closes the connection: the worker can't tell it from an `EventSource`, and an error would close an `EventSource` for good, since Chromium reports it to the page as an abort. Anything else ends with an error (`sw: seven updated, so <path> stopped`), so a reader knows it didn't get everything. The cut ends every stream 5 s old, in every tab, so a download that wasn't holding anything fails too if it runs while another stream holds an update. It can be fetched again. Nothing happens while a worker is only installing, so an install that fails cuts nothing. The update is active about 5 s after it's `installed`, before *Seven updated* could show. A stream that nobody reads any more is the exception (#699): Chromium lets the old worker go only once the reader takes more, so a half-read download still holds the update, and *Seven updated* names it after 10 s. Cutting the other streams can't release it, so while any stream the worker serves has gone 5 s without its reader asking for more, the worker cuts nothing and looks again 5 s later (#1104). Without that pause, every stream opened meanwhile would end 5 s after it opened, and an `EventSource` would reconnect every few seconds until that tab reloads.
 
 ## Certified packages
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai-ecoverse/slicc-bios",
-  "version": "1.51.7",
+  "version": "1.51.8",
   "description": "Slim loader that boots SLICC into OPFS and serves the UI from there",
   "type": "module",
   "license": "Apache-2.0",