[WebCryptoAPI] Remove race condition

[jugglinmike]

Configure testharness.js to wait until the global done function is
invoked before reporting results. This ensures that all
asynchronously-declared tests are defined and subsequently executed
regardless of variations in timing.

---

The instability is apparent when comparing the test results between Chrome, Firefox, and Safari, e.g. https://wpt.fyi/results/WebCryptoAPI/sign_verify?sha=58523789a0

Firefox and Safari are missing results from tests that depend on WebCryptoAPI/sign_verify/ecdsa.js due to the race condition which this patch corrects.

[jimsch]

I would just like to verify that it does not matter when the call to setup is made. Specifically that it does not matter if it is not called from all of the run_test functions.

[jugglinmike]

setup should be called before the load event on the document has fired. In worker contexts (where there is no document), explicit_done is automatically set to true. We could move the setting to the HTML file, but I feel that it's better to have it in-line with the code that requires it. There's more detail in the framework's documentation: https://web-platform-tests.org/writing-tests/testharness-api.html

I'm glad you mentioned the other run_test functions. I was focused on the instability we're experiencing today, which happens to be limited to the tests for ECDSA. However, rsa.js and hmac.js use the same pattern and could exhibit the same problem, so I've updated those as well.

[Ms2ger]

This is good. However, the tests are written as

var promise = importVectorKeys(vector, ["verify"], ["sign"]).then(function(vectors) {
  promise_test(function(test) { /* test stuff */ });
}, function(err) {
  promise_test(function(test) { assert_unreached(/* ... */) });
});

… while I would expect

promise_test(function(test) {
  return importVectorKeys(vector, ["verify"], ["sign"]).then(function(vectors) {
    /* test stuff */
  });
});

Why is that?

[jimsch]

The reason is that not all algorithms and not all sizes may be supported. This provides a uniform error in those cases.

[jugglinmike]

@jimsch Thanks for the review. Would you mind merging on my behalf?

[jimsch]

I am used to this happening automatically, but two of the tests never finished

[jugglinmike]

Those "checks" are intended to produce test results to assist in review. They're still under development, though, which is why they are not required for merging.

@lukebjerring Would you like me to let you know when/if I see issues with those checks?

[lukebjerring]

Yep; this bad state shouldn't occur anymore, and I've also updated the body of the check summary to include a link for reporting issues, e.g. https://github.com/web-platform-tests/wpt/pull/14291/checks?check_run_id=39572969

Note: wpt.fyi checks are still in beta!
See something wrong? Please file an issue!

[jugglinmike]

Very nice! Thanks, Luke