Vendor W3C MathML 4 grammar, rework metanorma-mathml wrapper (MathML 3 → 4)

[opoudjis]

Refs metanorma/metanorma#585

Child of the MathML 3 → 4 migration: swap the vendored grammar at the source of truth.

Scope

• Vendor the W3C MathML 4 RelaxNG Compact grammar (https://www.w3.org/Math/RelaxNG/mathml4/ — mathml4.rnc, mathml4-presentation.rnc, mathml4-content.rnc, mathml4-strict-content.rnc; generated from https://github.com/w3c/mathml-schema) into grammars/mathml/, replacing mathml3*.rnc. Pin to a dated draft snapshot and record provenance (source URL, retrieval date, licence, spec status) in grammars/mathml/README.adoc — MathML 4 is a Working Draft nearing CR; re-sync at CR and at REC.
• Rework the wrapper grammars/mathml/metanorma-mathml.rnc: include "mathml4.rnc"; re-apply the data-metanorma-numberformat augmentation on mn; re-apply the foreign-markup bans (annotation-xml &= notAllowed, anyElement &= notAllowed) — re-audit both against MathML 4's restructured content/annotation model; the combine-form encoding exists for libxml2 compatibility and must be preserved.
• Update basicdoc.rnc:1074 external reference if filenames change; update the README statement that "v4 is not yet officially published".
• Confirm jing/libxml2 compatibility of the compiled grammar, including whether the xsd:ID collision that currently forces id_check: false in metanorma-standoc (schema.rb) behaves the same under the v4 grammar.

Non-goals

• No dual-version validation: the schema validates exactly one MathML model (MathML 4 full — not MathML Core, which drops Content MathML).
• No v3→v4 ingestion transform here (design position recorded on the umbrella).

Downstream compile/propagation is metanorma/metanorma-standoc#1212.

🤖

[opoudjis]

Implementation landed in #40, with one significant discovery to record.

The W3C MathML 4 draft grammar is not usable as generated. The plan assumed an unmodified vendoring, mirroring the v3 work. In practice, jing rejects the full-profile composition outright: the draft-generated schema violates RELAX NG's duplicate-attribute restriction pervasively (the core, presentation and content profiles re-declare each other's attributes: data-other, dir/mathsize/mathvariant, id, alttext, maxsize, href, attributionsrc, displaystyle, encoding), and libxml2 additionally rejects the compiled grammar on the embedded-HTML/SVG token content (textorHTML svg branch), which would have broken every passthrough validation in metanorma-standoc.

The vendoring therefore carries 13 marked modifications, each tagged in place with a METANORMA MODIFICATION (basicdoc-models#39) comment preserving the upstream text, summarised in grammars/mathml/README.adoc. Three classes:

1. Duplicate-attribute repairs (the bulk): dropping re-declarations where a profile re-adds an attribute already carried by MathMLPGlobalAttributes or a sibling profile; maxsize merges the value spaces (length-percentage | "infinity") instead.
2. Foreign-markup surface: the anyElement/annotation-xml combine re-opens and a nested-include override are adjusted so the wrapper's bans stay expressible for both jing and libxml2; the textorHTML embedded-XHTML/SVG branches are removed at source (libxml2-fatal, and metanorma never emits them).
3. Provenance: README documents the snapshot date (2026-07-16), draft status (WD nearing CR), and instructs a re-audit of all modifications at CR/REC re-sync — upstream may fix the duplicates by then, and these defects are candidates for upstream reports to w3c/mathml-schema.

Behaviour verified with jing and libxml2: MathML 4 intent/arg + data-metanorma-numberformat accepted, text-only annotation accepted, plurimath's own emitted MathML accepted, foreign annotation-xml rejected; metanorma-standoc's schema/validation/passthrough specs green against the compiled set (metanorma/metanorma-standoc#1212).

🤖

[opoudjis]

Delivered and merged in #40: MathML 4 grammar vendored (2026-07-16 draft snapshot, 13 marked modifications documented in place and in the README), wrapper reworked, behaviour verified under jing and libxml2. Downstream propagation tracked on metanorma/metanorma-standoc#1212; re-sync at CR/REC to be handled when it happens.

🤖