http/tests/devtools/elements/shadow/shadow-distribution.js is failing with IncrementalShadowDOM flag.

https://test-results.appspot.com/data/layout_results/linux_chromium_rel_ng/77814/layout-test-results/results.html

It looks the cause is:

- devtool's inspector maintains its own distributedNodes as slot's *expanded* children. It looks expanded children are updated in InspectorDOMAgent::DidPerformSlotDistribution [1].
- probe::didPerformSlotDistribution is called if and only if actual slot's distributed_nodes are actually updated [2].

[1] https://cs.chromium.org/chromium/src/out/Debug/gen/third_party/blink/renderer/core/CoreProbesInl.h?type=cs&q=didPerformSlotDistribution&l=124
[2] https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/html/html_slot_element.cc?type=cs&l=554

Given that, it would be possible that inspector shows outdated distributed nodes unless devtools explicitly try to update distribution because the actual distribution is calculated lazily as per Blink's current architecture. An inspector might show outdated distributed nodes if the distribution is dirty; NeedsDistributionRecal flag is set.

I don't think this is a serious issue because I believe that end users wouldn't face this issue in most cases. The distribution is surely updated in most cases when end users are using an inspector.

However, it looks this can be observable in layout testing, where the rendering pipeline does't have any chance to update distribution between DOM mutation and assertions used in tests.

In fact, when we turn on IncrementalShadowDOM flag, this can be observable via shadow-distribution.js test.

I've investigated why this can not be observable in testing without IncrementalShadowDOM flag. It turned out an slotchange event triggers to update distribution. The slotchange event happens at the timing of between DOM mutation and assertion used in testing; slotchange event is dispatched at the end of microtask timing.

In non-IncrementalShadowDOM, an event path calculation needs to update distribution, however, this is no longer true for IncrementalShadowDOM, that is one of the benefits of IncrementalShadowDOM.

I guess the test shouldn't depend on an slotchange event implicitly for updating distribution.
I guess the shadow-distribuiton.js passes without any intention to depend on slotchange events.
IncrementalShadowDOM just revealed that.