Bug 303737 – [WebXR Hit Test] imported/w3c/web-platform-tests/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html is failing

• Component: WebXR
• Status: NEW

[Fujii Hironori]

[WebXR Hit Test] imported/w3c/web-platform-tests/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html is failing

FAIL Ensures hit test result returns null pose w/unlocalizable space - viewer space - webgl assert_true: Pose should be null since input source is not localizable expected true got false
FAIL Ensures hit test result returns null pose w/unlocalizable space - viewer space - webgl2 assert_true: Pose should be null since input source is not localizable expected true got false

[Fujii Hironori]

This test was added by https://chromium-review.googlesource.com/c/chromium/src/+/3733395
There was a following change.

~~~
diff --git a/third_party/blink/renderer/modules/xr/xr_hit_test_result.cc b/third_party/blink/renderer/modules/xr/xr_hit_test_result.cc
index ab8173eca7473..a367f6608b729 100644
--- a/third_party/blink/renderer/modules/xr/xr_hit_test_result.cc
+++ b/third_party/blink/renderer/modules/xr/xr_hit_test_result.cc
@@ -25,9 +25,18 @@ XRHitTestResult::XRHitTestResult(
   }
 
   auto maybe_other_space_native_from_mojo = other->NativeFromMojo();
-  DCHECK(maybe_other_space_native_from_mojo);
+  if (!maybe_other_space_native_from_mojo) {
+    return nullptr;
+  }
 
   auto mojo_from_this = TransformationMatrix(mojo_from_this_.ToTransform());
~~~

There were a review comments.

Alex Cooper: What made this change?
Piotr Bialecki: Previous version looked incorrect - I don't see why other->NativeFromMojo() would guarantee that it's non-nullopt in this case. Let me add a WPT to verify.
Alex Cooper: Hmm, yeah I'm not seeing any spec text that requires space to be localizable. So this is fine.
Piotr Bialecki: WPT added, PTAL.

This is the reason the test case was added.

[Radar WebKit Bug Importer]

<rdar://problem/166522898>

[Fujii Hironori]

I built and tested Chromium. If I modified the test case, it fails.

diff --git a/third_party/blink/web_tests/external/wpt/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html b/third_party/blink/web_tests/external/wpt/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html
index d1cb1a5a5c444..b113d0fc1331d 100644
--- a/third_party/blink/web_tests/external/wpt/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html
+++ b/third_party/blink/web_tests/external/wpt/webxr/hit-test/ar_hittest_subscription_unlocalizable.https.html
@@ -43,6 +43,7 @@ let testFunctionGenerator = function(ray, refSpaceName) {
       handedness: "right",
       targetRayMode: "tracked-pointer",
       pointerOrigin: IDENTITY_TRANSFORM,
+      gripOrigin: IDENTITY_TRANSFORM,
       profiles: []
     });
 

> This is a testharness.js-based test.
> [FAIL] Ensures hit test result returns null pose w/unlocalizable space - viewer space - webgl
>   assert_true: Pose should be null since input source is not localizable expected true got false
> [FAIL] Ensures hit test result returns null pose w/unlocalizable space - viewer space - webgl2
>   assert_true: Pose should be null since input source is not localizable expected true got false
> Harness: the test ran to completion.

Thus, if gripOrigin is null, the input source is unlocalizable.

[Fujii Hironori]

Looking into it more.

XRTargetRaySpace::MojoFromNative() returns nullopt in the test case because input_source_->MojoFromInput() is nullopt.
https://github.com/chromium/chromium/blob/9ffe3999a201b3d58e05873d438c43f5676afaa6/third_party/blink/renderer/modules/xr/xr_target_ray_space.cc#L38-L42

XRInputSource::MojoFromInput() returns nullopt because mojo_from_input_ is null.
https://github.com/chromium/chromium/blob/9ffe3999a201b3d58e05873d438c43f5676afaa6/third_party/blink/renderer/modules/xr/xr_input_source.cc#L262-L263

mojo_from_input_ is null because state->mojo_from_input is null XRInputSource::CreateOrUpdateFrom.
https://github.com/chromium/chromium/blob/9ffe3999a201b3d58e05873d438c43f5676afaa6/third_party/blink/renderer/modules/xr/xr_input_source.cc#L107

mojo_from_input_ is set in setGripOrigin().
https://github.com/chromium/chromium/blob/868cd2197aceec8d0641c47efcfd865f4adb0537/third_party/blink/web_tests/external/wpt/resources/chromium/webxr-test.js#L2011

setGripOrigin() is called if fakeInputSourceInit.gripOrigin != null.
https://github.com/chromium/chromium/blob/868cd2197aceec8d0641c47efcfd865f4adb0537/third_party/blink/web_tests/external/wpt/resources/chromium/webxr-test.js#L1978-L1979

[Sergio Villar Senin]

Looks like a WPT test added for something specific to chromium? If that's the case I think it's safe to skip, as you mentioned there seem to be nothing related to that in the spec.

Actually I'd open a bug against the spec, just to ask about that issue.

[Fujii Hironori]

This is about the spec of WebXR Test API.

It's unclear to me the relationship of pointerOrigin and gripOrigin.
https://immersive-web.github.io/webxr-test-api/

There is a comment in Chromium code <https://github.com/chromium/chromium/blob/9ffe3999a201b3d58e05873d438c43f5676afaa6/third_party/blink/renderer/modules/xr/xr_target_ray_space.cc#L39>

> // mojo_from_pointer is just: MojoFromInput*InputFromPointer;

IIUC, this means pointerOrigin is relative to gripOrigin.
That's the reason input_source.targetRaySpace is unlocalizable if gripOrigin is null.