v0.5.6

Sharpens the machine-readable locations added in 0.5.5: schema (RSC-005)
content-model findings now point at the exact offending node, and the
element_path form is corrected so it actually resolves in the XPath engine most
consumers use. Both build on 0.5.5's data.element_path; corpus recall is
unchanged (600/607, 1.1% false positives).

Added

• RSC-005 content-model findings now carry a real line:column and
  element_path. The RELAX NG engine reports which node collapsed the
  content model, so an OPF or XHTML schema violation points at the offending
  element — or the offending attribute (…/@name) when the violation is
  attribute-level — instead of anchoring the whole document at its root.
  (issue #17, reported by
  Doitsu on the MobileRead forum.)

Changed

• data.element_path now binds every namespaced name to a non-empty prefix.
  The 0.5.5 form left default-namespaced names bare and recorded the URI under an
  empty-string key, which is not resolvable in libxml2 / lxml (XPath 1.0 has no
  default namespace). Each namespace URI now gets a bound prefix — a readable
  well-known one for the common EPUB namespaces (opf, dc, h for XHTML,
  svg, …) or a generated ns… — so a path resolves directly with
  root.xpath(path, namespaces=data["namespaces"]).
  (issue #18, reported by
  Jens Tröger.)

Full notes: CHANGELOG.md. Also on crates.io.