Add steps to intercept requests using foreign fetch.

[mkruisselbrink]

If response is still null after a regular service worker has tried handling it
this gives a foreign fetch handler a chance to handle the request.

[annevk]

1. This lacks a same-origin check that I think we want. The fetch needs to actually be foreign. I'm not sure if you're handling that on the other side, but I think it would make more sense here. Happy to discuss of course.
2. A lot of this text is copy-and-pasted from the previous major step. If the requirements are actually identical, we should do the work to make that clear without duplication. I haven't quite figured out yet whether the "network error" requirements need to be the same, but CSP definitely makes sense.

[mkruisselbrink]

This lacks a same-origin check that I think we want. The fetch needs to actually be foreign. I'm not sure if you're handling that on the other side, but I think it would make more sense here. Happy to discuss of course.

Yeah, that check does exist on the service worker spec side as well, but I agree that it makes sense to have the check here as well. I also added the "is this a subresource" check here, to make it clearer which requests actually can be intercepted by foreign fetch.

A lot of this text is copy-and-pasted from the previous major step. If the requirements are actually identical, we should do the work to make that clear without duplication. I haven't quite figured out yet whether the "network error" requirements need to be the same, but CSP definitely makes sense.

Duplication is indeed unfortunate. Maybe one way to deal with this would be to move all the shared substeps to a separate step, so it roughly would be something like:

1. If [....] invoke handle fetch.
2. If response is null and [...], invoke handle foreign fetch.
3. If response is not null:
   1. Set actualResponse
   2. Do the network error stuff
   3. Do the CSP stuff

[annevk]

Yeah, that check does exist on the service worker spec side as well, but I agree that it makes sense to have the check here as well.

I think in the end we should only have it on one side.

With regards to duplication, what you suggest seems reasonable. An alternative might be to group what you suggest under a common skip-service-worker flag check. Not entirely sure what works better.

[mkruisselbrink]

Okay, factored out the common post-processing and skip-service-worker flag logic (which has the side effect of also fixing that previously things like "set response's CSP list" were called even when response was null).

[annevk]

"same" needs to reference "same origin" defined in HTML.

[annevk]

I like it!

[mkruisselbrink]

Addressed your comments.

[annevk]

Landed as 8510b2f. Thanks for working on this!