Should we clarify how to compute the destination of a request?

[hbenl]

The spec doesn't specify how to compute the destination of a request, it only links to the fetch spec which contains this incomplete table and current implementations in Chrome and Firefox sometimes don't follow that table and also disagree with each other (for example they both set destination to "" for top-level navigations and Chrome sets it to "" for fonts).
Related: #698 - the initiator type can be used as another hint to distinguish the different kinds of requests.

[jgraham]

The table in fetch is informative, but the intent is that it's summarizing data that's normatively specified in the algorithms. If the data is incorrect or doesn't match browsers that's an issue that we should file in fetch, or in the spec that invokes fetch.

[hbenl]

If the data is incorrect or doesn't match browsers that's an issue that we should file in fetch

I've checked some requests in a service worker and there the browsers seem to follow the table in the fetch spec, i.e. I saw both Chrome and Firefox setting destination to "document" for top-level navigations. So the discrepancy is only in the implementations of the BiDi protocol.
We could close this as being an implementation issue only. On the other hand since both implementations don't follow the spec, a few words of clarification (and a direct link to that table in the fetch spec) could help?

[juliandescottes]

In practice the actual specs hooking with fetch should set the destination. See HTML spec for navigation requests: https://html.spec.whatwg.org/#create-navigation-params-by-fetching

So I don't think this needs to be clarified in our spec.
If the implementation seems wrong (after cross checking with the actual spec triggering the request, not only the fetch note), then it's an implementation issue.

I agree it's painful to find out where a specific property gets set, especially when various specs interact together, but that's another topic.

[OrKoN]

Might be somewhat helpful but this page allows seeing in which specs the definition for destination is referenced: https://dontcallmedom.github.io/webdex/d.html#destination%40%40request%40dfn A non-normative table in this spec would be really great though. The Chromium implementation is based on interpreting some of the bits currently available in CDP so it's not 100% accurate (GoogleChromeLabs/chromium-bidi#2947).