Content is reset at unexpected times

[Rich-Harris]

I'm sure there's a simpler repro than this, but I'm making a note of what I observed just now for further investigation later.

1. Go to https://learn.svelte.dev/tutorial/textarea-inputs
2. Hit 'solve'
3. Change - lists to - [links](https://example.com)
4. Click the newly-rendered link in the output section
5. Close the tab

The content will have been reset (this happens for me in Firefox and Chrome).

Weirdly, you have to actually close the tab — if you leave it open and navigate back to the tutorial, the content won't have been reset. And if you drag the newly opened tab into a separate window and then close it, nothing will happen. It's... strange.

[tomoam]

I found that the iframe was reloading.

reset-output.mov

---

The relevant code is here.

learn.svelte.dev/content/tutorial/common/src/__client.js

Lines 112 to 124 in 6736ff7

	function ping() {
	parent.postMessage(
	{
	type: 'ping',
	data: {
	path: location.pathname + location.search + location.hash
	}
	},
	'*'
	);
	}
	setInterval(ping, 100);

learn.svelte.dev/src/routes/tutorial/[slug]/Output.svelte

Lines 54 to 62 in 6736ff7

	clearTimeout(timeout);
	timeout = setTimeout(() => {
	if (dev && !iframe) return;
	// we lost contact, refresh the page
	loading = true;
	set_iframe_src($base + path);
	loading = false;
	}, 1000);

---

This is caused by throttling of setInterval and setTimeout.

https://developer.chrome.com/blog/timer-throttling-in-chrome-88/

The ping in __client.js runs every 100ms with setInterval and sends the message. The handle_message in Output.svelte receives that message. If the message does not come for more than 1000 milliseconds, the iframe is reloaded.

However, when the tab goes to the background, setInterval and setTimeout are throttled to run approximately every 1000 - 1500 milliseconds. This means that sometimes the setTimeout delay time in the handle_message is shorter than the ping delay time, at which time set_iframe_src is called and the iframe is reloaded.

I have found the cause, but I don't know how to fix it.

[Rich-Harris]

ah, interesting! good sleuthing. i wonder if we can emit ping-pause on the pagehide event

[tomoam]

Thanks for the advice, I'll try it.