XProc step(s) to validate Office Open XML container files (for example MS Word '.docx').

Dynamic schema selection (strict / transitional) and MCE preprocessing

tr:wml-check (xpl/wml-check.xpl) now supports both ISO/IEC 29500 conformance flavors and performs Markup Compatibility preprocessing before RelaxNG validation.

How the schema is chosen

The flavor is detected per package part from the root element namespace:

• http://purl.oclc.org/ooxml/… → strict (validates against schema/strict/)
• http://schemas.openxmlformats.org/… → transitional (validates against schema/)
• anything else (e.g. customXml/item*.xml with application-specific vocabularies) → base (transitional) schema set, which is identical in both flavors for these permissive schemas.

The conformance option overrides detection for the whole run.

New options

option	default	values	effect
conformance	auto	auto, strict, transitional	select the schema set; auto = detect from root namespace
mce-preprocessing	yes	yes, no	strip mc:Ignorable extension content and resolve mc:AlternateContent before validation

With conformance=transitional and mce-preprocessing=no the step behaves like the previous release (raw validation, transitional schemas only).

Why MCE preprocessing

ECMA-376 RelaxNG schemas do not declare Markup Compatibility attributes (mc:Ignorable, mc:AlternateContent) nor the Microsoft extension namespaces (w14, w15, w16 …). Validating a Word-generated file without preprocessing therefore reports dozens of findings that every MCE-conforming consumer (including Word) must ignore. The preprocessing step (xsl/mce-normalize.xsl) implements the consumer side of ECMA-376 Part 3:

• attributes and elements in namespaces listed in mc:Ignorable are removed,
• mc:AlternateContent is replaced by the first mc:Choice whose @Requires prefixes are all understood (not ignorable), otherwise by the mc:Fallback,
• mc:ProcessContent is honored in its simple form (children of an ignorable element are processed if it carries a non-empty mc:ProcessContent).

Genuine schema violations — wrong child order in w:pPr/w:tblPr, empty w:rsidR, non-integer w:id, … — are of course still reported.

Known simplifications: namespace prefixes are resolved against the part's root element only, and per-element namespace redeclarations are not considered. Both are irrelevant for Word-generated parts.

Schema provenance

• schema/ — transitional set, unchanged: RelaxNG versions of the ECMA-376 5th edition, Part 4 (OfficeOpenXML-RELAXNG-Transitional.zip), converted with trang 20091111.
• schema/strict/ — strict set: RelaxNG versions of the ECMA-376 5th edition, Part 1 (OfficeOpenXML-RELAXNG-Strict.zip), converted with the same trang version. A reconversion of the transitional wml.rnc is byte-identical to the committed schema/wml.rng, so both sets were produced identically.

Testing

test/part-check.xpl is a standalone harness (standard XProc 1.0 steps only, runs with any XML Calabash) that exercises flavor detection, MCE preprocessing and validation for a single part:

calabash -i source=test/fxt/transitional-mixed.xml -o result=report.xml \
  test/part-check.xpl conformance=auto mce-preprocessing=yes \
  rng-name=WordprocessingML_Main_Document.rng

Fixtures and expected results:

fixture	options	expected
transitional-mixed.xml	auto, mce=yes	transitional; exactly 2 real errors (shd order, empty rsidR); no MCE errors
transitional-mixed.xml	auto, mce=no	previous behavior: additionally mc:Ignorable, w14:paraId, w15:custom, 2× AlternateContent errors
transitional-numbering.xml	auto, mce=yes	0 errors (reproduces the w15:restartNumberingAfterBreak noise)
strict-sample.xml	auto	strict; 0 errors against schema/strict/
strict-sample.xml	conformance=transitional	root element mismatch (proves the flavor selection)

The full tr:wml-check pipeline requires the transpect Calabash extensions (unzip, rng-extension) and therefore runs in the transpect environment only.