I think this is actually easier to implement than the version here. Instead of running a new validator class, I think you can add this validation to the existing IndexByName class. Just change addField to include the validation:

  private void addField(String name, int fieldId) {
    String fullName = name;
    if (!fieldNames().isEmpty()) {
      fullName = DOT.join(DOT.join(fieldNames().descendingIterator()), name);
    }

    Integer existingFieldId = nameToId.put(fullName, fieldId);
    if (existingFieldId != null) {
      throw new ValidationException(
          "Invalid schema: multiple fields for name %s: %s and %s", fullName, existingFieldId, fieldId);
    }
  }

Then all you'd need to do is to access lazyNameToId() in the constructor.

@rdblue, Good idea. Let me update it

@rdblue, sorry I forgot that this visitor is different from IndexByName visitor since it directly returns null in list and map function. That is because we don't push filedNames in case of List and Map when visiting the schema, so the testMixedTypes test fails if we apply validation in IndexByName.

For example, when we define a mixed typed schema such as:

    Schema schema = new Schema(
        required(1, "list_of_lists",
            ListType.ofOptional(2, ListType.ofOptional(3, StructType.of(
                required(4, "id", LongType.get()))
            ))));

It throws exception like:
Invalid schema: multiple fields for name list_of_lists.element: 3 and 2 org.apache.iceberg.exceptions.ValidationException: Invalid schema: multiple fields for name list_of_lists.element: 3 and 2

The map returned by TypeUtil.indexByName actually cannot express the concept of list of list.

@chenjunjiedada, ignoring the problem when the name is "element" seems to solve that case:

    Integer existingFieldId = nameToId.put(fullName, fieldId);
    if (existingFieldId != null && !"element".equals(name)) {
      throw new ValidationException(
          "Invalid schema: multiple fields for name %s: %s and %s", fullName, existingFieldId, fieldId);
    }

I think that's a reasonable fix for now. I think we should also go back and fix how these indexes are produced. The field names are only used in IndexByName, so I think we can fix this.

@chenjunjiedada, I submitted a fix for the indexing problem with the ValidationException in #627. Can you take a look?

Sure, will take a look.

@rdblue , ready for merge?