using mercure somehow breaks bfcache

[asdfzdfj]

in short: mercure + bfcache + onbeforeunload => page cannot be stored in bfcache, causing every navigation to do a page load

possible solution: rewrite teardown function to either use pagehide event or stimulusjs disconnect() function

note that I might be off here, if anyone knows what's more likely them I'm all ears

EDIT 1: ok, maybe onbeforeunload isn't tha main culprit, it seems like having an open connection to mercure at all breaks bfcache, still needs more investigating

---

based on discussion/discovery here

You know what's great... It is mercure... I had it enabled and it is disabled on kbin.run
Now I have disabled it and no more extra requests...

Originally posted by @BentiGorlich in #689 (comment)

merde*, I think I have idea why:

• this looks like a browser bfcache territory
• the notification controller sets up onbeforeunload event listener to tear down the mercure connection
  

  mbin/assets/controllers/notifications_controller.js

  Lines 13 to 21 in 3e68837

  	connect() {
  	this.es(this.getTopics());
  	window.onbeforeunload = function (event) {
  	if (window.es !== undefined) {
  	window.es.close();
  	}
  	};
  	}

• this firefox docs says that the page is ineligible if beforeunload listener is used (the docs seems ancient but that's the only page of firefox bfcache docs I could find atm)

* somehow this sound came up in my head upon reading that

Originally posted by @asdfzdfj in #689 (comment)

[BentiGorlich]

I just tried on my server changing window.onbeforeunload to window.onpagehide and now it works properly even on Fennec

[asdfzdfj]

on my own testing it's still some bit inconsistent, you sure that you didn't forgot to enable mercure back when testing?

[BentiGorlich]

I am very sure that mercure was deactivated when testing, however Fennec still had the problems, because the notifications controller is always loaded, even when mercure is disabled

[BentiGorlich]

But enabling mercure still breaks the cache, that's for sure 😁

[asdfzdfj]

yeah that seems to be it, it looks like the other half/part seems to be having mercure connection opened at all, so that's more to investigate

[BentiGorlich]

I just tried enabling mercure, but have it not be reachable, so the request fails instantly. That solves the issue. But I think our code behaves differently when the connection is established versus not established... It is really a bummer that it doesn't just use web sockets, but this weird "eventsource" call...

[BentiGorlich]

It is really a bummer that it doesn't just use web sockets, but this weird "eventsource" call...

Maybe I am wrong about it though. Seems like eventsources are way older and have wider support than websockets... Still weird

[github-actions]

This issue is stale because it has been open 50 days with no activity. Remove stale label or comment or this will be closed in 6 days.