Hmm. I'm not sure this is right. react-reconciler is not meant to be in sync with react-dom. They have a separate release cycle. There is no guarantee that any arbitrary version of react-reconciler will work with any arbitrary version of react-dom. The fact that they share the same method (and the one from react-reconciler "works" on inputs from react-dom) is an accident, not an intentional feature.

react-reconciler/reflection was only intended for us by consumers of react-reconciler itself (i.e. people creating custom renderers). Those don't need to be in sync with react-dom.

Thanks for clarifying @gaearon, I can revert. The fact that so much code needs to be duplicated in the adapter is a strong indicator that the react-dom testing utilities are insufficient, so it would be great if we could finally address that.

I agree, open to suggestions how to fix those :-) There was a proposal from Enzyme side for us to expose toTree in the test renderer. But even that didn’t end up being used. We’re totally open to Enzyme taking more ownership over this (and potentially exposing toTree on ReactDOM in some special testing mode), but we should probably start “blessed” integration from the test renderer.

We're beginning to bump into problems caused by this at Facebook. Specifically, we've changed some internal constants, but we have to remember to fork both the Enzyme adapter and this deep dependency that we don't normally check in.

It would be great to undo this abstraction as soon as possible.