The goal is to avoid recomputing field several times. Consider, for instance,
two fields F and G, such that G depends on F. Suppose that G is recomputed
before F. Saving G to database proceeds well, but when F is saved to database,
G is invalidated and marked for recomputation. Field G is possibly recomputed
twice on some records.
Avoid this situation by chosing a field such that none of its dependencies must
be recomputed; use a topological sort based on field dependencies for that
purpose. In the example above, G will never be recomputed before F.
Raphael Collet
ac98af1e9f