Kohsuke KAWAGUCHI wrote:
> Path-expression which is used for (c) requires another constraint. That
> is, the specified path expression may not match more than one node.
> Say every <section> element is uniquely identified by the value of its
> <name> element. If a schema allows <name> element to appear more than
> once in one section, then there is something wrong, isn't there.
Yes, of course. I should have thought about this. (On the other hand, one
could argue that a section can have more than one names.)
In my previous mail, I have already demonstrated construction of grammars
such that an element matches the given path expression if
and only if a successful computation assigns a marked named pattern. For
a given grammar G and a given pattern p, let G:p be the constructed grammar.
First, consider those define elements which have marked named patterns.
Replase each terminal symbol a with marked(a), which is a
new element name. Then, we replace all <attribute> in G:p with <empty>.
Let marked(G:p) be the modified grammar.
Third, we create a determinisitc tree automaton that accepts
trees containing one and only one marked symbol. The tree
automaton is shown below. Final states are s1.
s0 ::= not-marked<s0*>
s1 ::= marked<s0*>
s1 ::= not-marked<s0* s1 s0*>
s2 ::= marked <(s0 | s1 | s2)* - s0*)>
s2 ::= not-marked <(s0 | s1 | s2)* - (s0* s1 s0*) - s0*)>
If we use s0 and s2 as final states, this automaton accepts
trees containing no marked symbols or more than one marked symbol.
Finally, we compute the intersection of marked(G:p) and this tree
automaton. If the result is empty, for any tree t valid against G,
one and at most one node in t matches the given pattern p.
Cheers,
Makoto
