For sync XHR consider setting the service-workers mode to "none" to align browsers

[asutherland]

What is the issue with the XMLHttpRequest Standard?

It's come up that the WPTs for service-worker sync XHR interception fail on blink and webkit and pass on gecko. Given that this has been a longstanding implementation divergence, perhaps we should standardize this by changing the initialization of the Request instance in XHR's send() method to set the service-workers mode to "none" for sync XHRs? Note that the req as initialized there is used for both sync and async XHR, and we would not want to impact async XHR where browsers are already aligned in intercepting the request.

xref #20

[annevk]

It seems unfortunate for service workers to not get all HTTP network activity, but we had already been discussing offering this as an option in fetch() as well so I suppose this is fine.

[martinthomson]

Just to add some spice, it has come to my attention that sync XHR and service workers have been proposed as a way of "fixing" asynchronous browser APIs. If there is a browser API that you don't want to have asynchronous (WebCrypto, say) you can wire that up with a synchronous XHR, routing the resulting fetch via a service worker that holds the response until the promise resolves, to turn it into a blocking function call. It's abominable, but thankfully it only seems to work in Firefox. I suggest that maybe it shouldn't work at all.