Clarify `EventSource` behavior for documents entering and restoring from BFCache

[zp1996]

What is the issue with the HTML Standard?

EventSource interaction with BFCache seems underspecified today.

As far as I can tell, the HTML Standard does not clearly define what should happen when an EventSource associated Document becomes non-fully-active due to entering the back/forward cache, and then later becomes fully active again when restored.

This appears to leave room for interop differences:

• Safari allows the page to enter BFCache, closes the EventSource, and reconnects when the page is restored.
• Chrome is moving toward matching Safari's behavior.
• Firefox appears to keep such pages out of BFCache.

So I think it would be helpful to clarify the expected behavior for EventSource, for example along these lines:

• when a document becomes non-fully-active due to BFCache, its EventSource should stop delivering events;
• the user agent may close / fail the active connection at that point;
• events that would have arrived while the document is in BFCache should not be queued for later delivery;
• when the document is restored and becomes fully active again, the EventSource may resume through its normal reconnection behavior.

Relatedly, there is already BFCache-oriented guidance for spec in the TAG document “Supporting BFCached Documents”, but it seems like the normative clarification should live in the HTML Standard's EventSource processing model.

Would such a clarification be acceptable here? If so, I’d be happy to put together a PR.

cc: @ricea @rakina @annevk @zcorpan

[ricea]

I agree it would be great to have this behaviour normatively specified.

[zp1996]

Hi @zcorpan, what do you think? The expected behavior is inconsistent with Firefox's current implementation.