Are headers semantically passed by value or by (mutable) reference?

[dicej]

When I pass a headers parameter to new-outgoing-response and then edit those headers prior to passing the response to the host via set-response-outparam, should the edits be reflected in the response? And what if I edit the headers after calling set-response-outparam? Analogous questions apply for outgoing requests.

[lann]

I think ideally outgoing-response would hold a shared reference to headers until dropped. Maybe the best approximation would be documenting that you shouldn't mutate headers in that scope and call the result "unspecified behavior"? Or maybe just specify that doing so traps?

[pchickey]

In the preview 2 timeframe we are now going to just have reference-counted resources, so I would assume they are passed by a mutable reference.

We may want to add a method which performs a deep copy of a headers and returns a fresh (reference count of 1) resource to the copy.

Once we get owned/borrowed resources we can revisit and see if passing by value makes sense.

[lukewagner]

+1 to what @pchickey said.

That does mean that there is an unavoidable race if you hold onto a handle to headers after you pass an outgoing-request pointing to that same headers to outgoing-handler.handle. I'll file a PR to document this case.

[dicej]

To be clear, I do understand about reference-counted resources, and that they imply mutable aliasing for a type like fields. However, we have the option to say e.g. new-outgoing-response makes a private, immutable copy of headers if we want to. I.e. mutable aliasing is a choice rather than an inevitable consequence of the type of resource we're using.

Anyway, I'm fine with the consensus here and look forward to the documentation PR. Thanks, all!

[lukewagner]

Yes, good point, that is a reasonable alternative to consider. If we didn't have the future option of expressing exclusive ownership at the interface level, I'd want to more carefully dig into that option, b/c it does avoid some real composability footguns.

#43 is up, PTAL