Re: Hierarchical databases and semistructured data

"Morten" <morten_at_kikobu.com> wrote in message news:3C73F433.2090001_at_kikobu.com...

>

The latter. Peter Buneman actually discusses several possiblities in his tutorial which are more or less equivalent. The version where only edges are labeled is mathematically more elegant, but the one with the labeled leaves is more common.

> The mismatch is illustrated
> by the figure in the article, and the figure on page 33 in the book,
> basically the leaves of the tree in the book have value, whereas they
> are "void" in the article, but have a label of a "type" (rather than a
> symbol I assume)

Peter Buneman actually gives three models. The first model has unlabeled nodes and labeled edges where the labels can be symbols or a value. The second model has edges labeled with symbols, internal nodes unlabeled and the leaf nodes are labeled with a value. The third model also labels the internal nodes with symbols. I believe, by the way, that he makes a small mistake when he presents the third model. The formulation should have been:

  type base = int | string | ... | symbol   type tree = base | label x set(label x tree)

otherwise the type base is never used in the tree.

• Jan Hidders