IndexedDB/keypath-special-identifiers.htm failure in Edge is misleading

[foolip]

https://wpt.fyi/IndexedDB/keypath-special-identifiers.htm

Here it looks like there's a different test in Edge than in the other browsers. Actually, for some reason it seems like the test has been treated as file_is_test, probably at this line:
https://github.com/w3c/web-platform-tests/blob/62c85b4892519965b4e15f4291a4f7acc6e13ab5/resources/testharness.js#L2915

This should arguably have been a harness error, and the setup code should then have been wrapped in setup() to fix the problem.

[foolip]

@pwnall, can you take a look? Test is from 279eb98 (one of our earliest exports)

@jugglinmik, is this the category of weirdness you mentioned last week? The behavior we have here sure doesn't seem very good.

[foolip]

As for the reason for the failure, it's that Edge doesn't seem to support the File constructor.

[foolip]

@jugglinmike I meant to poke you here but failed.

[jugglinmike]

It would be nice if the harness could identify this as a faulty test. I don't think this is possible as long as we support "single page tests," at least in the way we do today. I've filed gh-10853 to talk about resolving that.

Separately, if code in a setup function throws an error, testharness.js assigns the status "ERROR" to the test. Do you think that should change?

If we wanted this particular test to be interpreted as harness status "okay" with many test failures, then we could wrap the instantiation logic in dedicated functions, e.g.

   {
     type: 'File',
     property: 'name',
-    instance: new File([''], 'foo'),
+    instance: () => new File([''], 'foo'),
   },