FilteredLattice

Decompose a state into potentially smaller parts. Guarantees for any two states a and b that decompose(a).fold(b)(merge) == bmergea, i.e., merging the decomposed values into b has the same result as merging the full a into b (assuming b is normalized).

Note that the goal here is small individual storage size at reasonable computational cost. Minimalism of returned results is not guaranteed. It is also not guaranteed that the result does not overlap. The result may be the empty sequence.

By assumption: associative, commutative, idempotent.

Implementation note: If it matters, assume that left is the current state and right is an added delta. All code should assume that left is the larger state (and optimize for this). If left == right, prefer to return left.

Computes delta without state. Overriding this is discouraged.

Lattice order is derived from merge, but should be overridden for efficiency

Some types have multiple structural representations for semantically the same value, e.g., they may contain redundant or replaced parts. This can lead to semantically equivalent values that are not structurally equal. Normalize tries to fix this. Overriding this is discouraged.