Reject ambiguous content models at compile time via a Glushkov position
automaton over a literal, QName-resolved view of the raw schema tree, built
just for this rule and independent of the instance-validation content model.
The literal view resolves a ref's QName through the schema's prefix bindings
(not the blanket target namespace) and does not expand substitution groups,
removing the two false-overlap sources a check over the compiled model
produced. Determinism is per particle: a particle competing with a repetition
of itself is never a violation, so a fixed count {n,n} unrolls into copies in
sequence while a true range or unbounded repeat uses a self-loop, and only two
distinct source particles with overlapping labels conflict. The overlap test
is QName-only, an under-rejection that can never over-reject.
A position cap bounds pathological nesting (skipping the check, never
over-rejecting). The element-declarations-consistent test loses its
incidentally-ambiguous (a?, a) case in favor of a deterministic (a, b, a).
XSTS: invalid schemas accepted 884 to 864, valid schemas rejected held at 72,
instance buckets unchanged.
Mihaela Mihaljevic
90ae16cba1