fix: close #77 SAX rich-typing edge cases (FieldOccurrence refactor + ns-decl injection)

[teaguesterling]

Brings in @bendrucker's sax-ns-order follow-up to PR #75, closing both edge cases tracked in #77. Opening on his behalf to keep release-prep moving — his offer from #75: "feel free to merge that branch into this PR if you want, or merge this as-is and I'll open a follow up to close #77." All commits authored by him (verbatim from bendrucker:sax-ns-order); branch hosted on origin only because GitHub blocks cross-fork PRs from the upstream owner.

Commit 1 — 20b92c3 preserve document order for mixed text/XML SAX occurrences

Replaces the four parallel accumulator maps (current_values / current_xml_values / current_lists / current_xml_lists) with a single current_fields → vector<FieldOccurrence> ordered sequence per field, each occurrence tagged {is_xml, payload} and appended in element-close order.

This is the "cleaner shape ... one map of (name) → Variant<Text, XmlFragment> that preserves document order" Ben himself flagged as out-of-scope in PR #75's body. Without this, the prior release would publish the parallel-maps shape that the author had already replaced.

• SAXEndElementNs: the 3-branch "promote scalar → list / append / new scalar" dance collapses to current_fields[name].push_back({is_xml, std::move(value)}). Singleton-vs-list semantics become consumer-derived from occs.size().
• InferSchemaFromStream: 4 separate emit loops → 1 with a ternary on occ.is_xml. Closes the synthetic-doc-order leak.
• AccumulatorToRow LIST + VARCHAR branches: one ordered loop per field replaces the parallel-list dispatch. This is the actual read_xml SAX rich-typing: namespace declarations dropped on reparsed fragments and list order lost for mixed text/XML occurrences #77 case-2 fix — the prior "Document order between text and XML items is not preserved" comment retires.
• STRUCT branch: now reads occs.front() explicitly, making the implicit "XML wins over text" semantics visible.

Net −62 lines.

Commit 2 — 3520d3b inject namespace declarations into reparsed SAX fragments

Closes the second #77 edge case: with namespaces:='keep', SAX captures a prefixed nested element (ns:tag) into a raw fragment but not the originating xmlns:ns declaration. ExtractValueFromXmlFragment then reparses under strict (non-recovery) mode and the unresolved prefix collapses the value to typed NULL.

• SAXStartElementNs accumulates prefixed namespace declarations into a std::map<string,string> on the accumulator (skips default namespace — only prefixes need resolving; extraction matches by local name).
• Persistent across records (NOT cleared by Reset()) since root-level xmlns: decls appear before the first record.
• BuildNamespaceDeclarations() serializes as xmlns:prefix="uri"... with XmlEscapeAttr on URIs (legitimately needed for & in query strings).
• InferSchemaFromStream declares accumulated namespaces on the synthetic <root> (uses records.back() — monotonic accumulation means it's the union of every decl seen while sampling).
• AccumulatorToRow builds ns_decls once per row, threads through 4 ExtractValueFromXmlFragment callsites via a defaulted extra_ns_decls="" parameter on the function added in feat: STRUCT widening and SAX rich-typing for read_xml(union_by_name) #75 itself. Fully backwards-compatible.
• Default namespaces:='strip' is untouched — extra cost paid only when needed.

Tests

• test/sql/github_issue_77_sax_edge_cases.test: ns-decl injection (DOM-ref + SAX-via-maximum_file_size:=1 × keep mode + strip-mode control), mixed-order assertions for the FieldOccurrence refactor, and a sax_ns_later_record.xml case exercising the cumulative-across-records inference path (prefix first declared on a non-root record).
• test/sql/xml_sax_accumulator_parity.test: minor adjustments for the new internal API.
• Fixtures: sax_mixed_order.xml, sax_ns_keep.xml, sax_ns_later_record.xml.

Closes

Closes #77.

cc @bendrucker — happy to amend the PR description if anything misrepresents the change.

[Copilot]

Pull request overview

This PR addresses SAX rich-typing edge cases for XML reading, focused on preserving mixed text/XML occurrence order and carrying namespace declarations into reparsed fragments.

Changes:

• Refactors SAX field accumulation to ordered FieldOccurrence sequences.
• Injects accumulated namespace declarations into synthetic/reparsed XML.
• Adds regression tests and fixtures for issue #77 SAX edge cases.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/xml_sax_reader.cpp	Refactors SAX accumulation/materialization and adds namespace declaration capture/serialization.
src/include/xml_sax_reader.hpp	Adds FieldOccurrence and namespace declaration state/API.
src/xml_schema_inference.cpp	Extends fragment wrapping to accept injected namespace declarations.
src/include/xml_schema_inference.hpp	Updates ExtractValueFromXmlFragment signature with a defaulted namespace parameter.
test/sql/github_issue_77_sax_edge_cases.test	Adds regression coverage for mixed order and namespace handling.
test/sql/xml_sax_accumulator_parity.test	Updates comments for the new accumulator model.
test/xml/sax_mixed_order.xml, test/xml/sax_ns_keep.xml, test/xml/sax_ns_later_record.xml	Adds XML fixtures for the new regression tests.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.