I was investigating a Firefox bug about custom properties with css-wide keyword fallbacks.
The following test-case (live) renders differently in Chromium and Safari. I can explain the difference, I believe.
Safari seems to first figure out what declarations apply, then you resolve dependencies between the values, and treat revert-layer as invalid at computed value time. This is similar to how longhands work.
Chromium seems to do something different, where their dependency-cycle detection also seems to deal with revert-layer somehow.
To me Safari's behavior seems easier to understand, but I might be biased because that's how Gecko implements custom-properties too. I think Chromium's behavior is more correct tho, but a clarification on the spec and sanity-checking my understanding of the situation would be great.
@mdubet / @andruud is my understanding above right?