[resize-observer] the spec text on explanatory "ResizeObservation" interface (not intended as a real interface) gets picked up by Reffy tool and baked into WPT tests

[dholbert]

https://wpt.fyi/results/resize-observer/idlharness.window.html shows a bunch of cross-browser test failures for the ResizeObservation interface, which come from chunk of IDL:

https://github.com/web-platform-tests/wpt/blob/a8d97c0c7c58cdc2def51573e30c2cad92923d75/interfaces/resize-observer.idl#L39-L43

interface ResizeObservation {
    readonly attribute Element target;
    readonly attribute ResizeObserverBoxOptions observedBox;
    readonly attribute ResizeObserverSize lastReportedSize;
};

This interface is mentioned in the spec, but it's clearly labeled as an "example" which is not supposed to be exposed to JS:

§ 3.1. ResizeObservation example struct
This section is not normative. ResizeObservation is an example struct that can be used in implementation of Resize Observer. It is being included here in order to help provide clarity during the processing model. It effectively holds observation information for a single Element. This interface is not visible to Javascript.
[...]
interface ResizeObservation {

https://drafts.csswg.org/resize-observer/#resize-observation-interface

Unfortunately, the WPT interfaces/resize-observer.idl file is autogenerated, so we can't just edit it to remove this example interface. It'll just come back whenever the file gets regenerated, presumably. So I think we need to edit the spec to make it clearer (to the Reffy tool) that this interface is non-normative and not something that should be tested. (And really, someone should probably add a WPT test to validate that the interface does not exist, since it's not supposed to.)

[dholbert]

Also, worth stating for the record -- both Chrome 76 and Firefox Nightly 68 correctly "fail" all of the ResizeObservation subtests right now, as shown on wpt.fyi. (only mentioning those two because other engines don't implement ResizeObserver at all.)

So, all implementations are correct here, and we just need a spec-fix (which will then implicitly fix the test, since the test is generated by the spec).

[gregwhitworth]

@dholbert I asked on the testing IRC channel, but I'll ask here too. I have no knowledge of the Reffy tool, so how do I change the spec to make it so that it isn't using this example IDL as actual IDL?

[dholbert]

I'm afraid I have no knowledge of the Reffy tool either. (I only figured out that it was involved via following the trail of clues to figure out where the test data was coming from.)

[bkardell]

(only mentioning those two because other engines don't implement ResizeObserver at all.)

ResizeObserver is in the latest webkit tech previews too

[gregwhitworth]

ok calling in the big guns - @gsnedders any ideas?

[gsnedders]

I can only suggest posting an issue on the Reffy repo or just asking @tidoust.

[tidoust]

Reffy extracts the IDL index at the end of the spec. That index contains ResizeObservation, so Reffy picks it up as well. To drop the interface from the index, you have to tell Bikeshed not to process it, which can be done by adding a class="extract" to the IDL block definition (see Turning off processing).

By the way, I do not know what you're trying to achieve here with ResizeObservation, but I note that the definition of observe currently normatively references ResizeObservation, which seems wrong to me (a spec should not normatively reference something that is defined informatively).