Currently, we traverse the chain of errors from the latest to deepest for merging data into a single structure.

We walk across the errors recursively for each field. It would be faster to do it once.