Re: [DataCache] Unhandled Cases in Networking Changes

>> There is nothing specified for what to do when there is no embedded server for the resource. [...]
> In case no embedded server is available, the user agent fetches from the server. I have clarified this in the spec. This behavior makes most sense.

Thanks for clarifying.

>> 2. In the off-line case there is a MutableHttpResponse that gets written to by the interception function. Here is one of the sub-steps:
>> [[
>> 10.4 Wait for the interception function to dispatch the dynamic response.
>> ]]
>> I think this is ambiguous. When does the interception function "dispatch" the response?
>> - it can explicitly dispatch by calling MutableHttpResponse.send;
> This is the correct way.
>> However what happens if send() is not called? What happens when:
>> - the interception function exits (either by exception or naturally)
> The normal network timeout logic should apply here.

Ahh, I see. I had not thought about that. Thanks.

In that case, the wording "Wait for the ..." could be interpreted
as waiting unconditionally. Maybe that could be clarified. I see
many developers getting caught forgetting send() and getting
weird, unexpected errors / behavior across different implementations.

>> - implicitly dispatch
>> - raise an exception and abort to normal behavior
>> I am currently siding with implicitly dispatching, which makes the send() optional (and unnecessary?). Do you see any disadvantage to this?
> Implicitly dispatching is a problem since the interception function may have to wait until a time some storage operation completes.

Good point. I thought of that later, and I agree. I guess I am
more used to how an Async-capable JavaScript testing library
QUnit [1] handles this. Its implicit by default, but if you
want it to be explicit, you can.

The workflow would then be:

  response.delay(); // explicitly saying, wait for the send()
  // trigger async action
  setTimeout(function() {
    // do work...
  }, ...);

I just bring that up for discussion purposes. I am fine with
always explicitly calling send().


Joseph Pecoraro

Received on Friday, 1 January 2010 21:08:01 UTC