In order to support the full richness of DFDL including properties like dfdl:separatorSuppressionPolicy in all its variations, the Daffodil sequence implementation needed refactoring.

The UML diagram below shows the introduction of the SequenceChild class in the Gram object tree, and how it relates to the runtime classes derived from SequenceChildParser and SequenceChildUnparser.

A sequence parser, for example, is a loop that drives a number of sequence child parsers, some of which are for scalars (elements or model groups that are not repeating/optional), others are for array/optional elements.

Array/Optional elements no longer have their own Parsers/Unparsers independent of the sequences that contain them. Rather, the OrderedSequenceParserBase class orchestrates the child parsing for all kinds of children, be they scalars, or occurrences of array/optional elements. That way they can deal with separators without special cases around array boundaries. So the iteration of recurring array/optional elements is driven from the OrderedSequenceParserBase's parse method. Each kind of sequence child provides a parseOne method designed to parse exactly one occurrence, driven by the framework.

This design also takes another step towards solving the Daffodil schema compiler space/time problem, and our inability to implement recursive types - both of which are due to the backpointers from every term to its point of use so that no term can be shared.

The SequenceChild is intended to eventually do all the calculations related to the sequence child instance, such as alignment, whether possibly trailing position in group, etc. Everything that depends on the context surrounding the child, when combined with characteristics of the child itself. This split allows the sequence child parser/unparsers to be shared. The complexities associated with use of  those parsers/unparsers within the context of a particular sequence, are handled by a SequenceChildParser (or Unparser) which delegates to the actual (sharable) child Parser/Unparser.

From the DSOM perspective.... the SequenceChild has a unique backpointer to the containing Sequence, and is effectively part of the Sequence object. The SequenceChild object does not own/contain its child object. It refers to it.

Hence, the DSOM object for the child itself does NOT have such a unique backpointer.

.

Below is the mixin architecture for creating the various flavors of SequenceChildParser and SequenceChildUnparser. We have mixable traits for separated/unseparated, and for RepParser, RepUnparser which are for array/optional elements (there is no Scalar mixin. It is the absence of a Rep mixin that makes it scalar.)