Does this PR check to make sure that the new matches target is a valid one with traces? Or if not, do all references get cleaned up? For example if x2, x3, x4, x5 are all matching x and x is invalid and we make them all point to x2 but then x2 is invalid also do we then match them all to x3?

Does this PR check to make sure that the new matches target is a valid one with traces? Or if not, do all references get cleaned up? For example if x2, x3, x4, x5 are all matching x and x is invalid and we make them all point to x2 but then x2 is invalid also do we then match them all to x3?

It does only search within valid axes. So this should not be a problem.
BTW it would be great to test the demo posted on the PR description with more combinations.

One thing this approach misses is if the lost axis sets an explicit range or rangemode https://codepen.io/alexcjohnson/pen/wvBNXgY?editors=0010
That seems like something we want to support, right?

Ah drat. Also categoryarray and any other axis-related option I would guess. That is a drawback to this approach: you need to copy everything over basically

Then, should x.range override x2.range if both specified?

should x.range override x2.range if both specified?

The answer to that is "exactly whatever would have happened if there was data on the xy subplot" - I don't know how we handle that though I'd guess that xaxis.range wins.

Closing in respect to #4529.