@doxi/docx

DOCX (OOXML) export and import for Doxiva. Zero runtime dependencies — hand-written ZIP reader/writer, hand-written OOXML builder, hand-rolled XML parser with a DOMParser fast path.

Status: v0.10.x preview. Round-trips Doxiva-authored DOCX losslessly within the supported schema subset; foreign docs (Word / LibreOffice / Google Docs export) come in best-effort with explicit warnings.

Install

pnpm add @doxi/docx @doxi/core

Export

import { exportDocx } from '@doxi/docx'

const bytes: Uint8Array = exportDocx(doc) // pure function. Deterministic output.
// Save to disk, send to server, trigger a browser download — your call.

Pure function: no DOM, no view state, no IO. Deterministic byte output, suitable for golden-file snapshots.

Import

import { importDocx } from '@doxi/docx'

const { doc, warnings } = await importDocx(bytes)
// warnings: ImportWarning[] — surface in your UI
view.dispatch(view.state.tr.replace(0, view.state.doc.content.size, doc))

Async because ZIP decompression goes through the platform DecompressionStream (Web Streams). Returns a Doxiva document tree plus structured warnings for anything the importer couldn't handle.

Supported schema subset

Paragraphs (alignment + line height), headings, lists (bullet + numbered, multiple instances), tables (incl. colspan/rowspan + header rows), text with bold/italic/underline/color/font-size/font-family marks, hyperlinks, horizontal rule, forced page break, blockquote.

Self-round-trip guarantee

@doxi/docx ships a property-test harness (fast-check) with 200 runs per arbitrary across 5 arbitraries (1000 runs total). Every generator round-trips through importDocx(exportDocx(doc)) and asserts deep equality against the input.

Honest deferrals

• Images — word/media/ binaries and image*.xml rels are read past silently; no image nodes materialize yet.
• page_var field codes (PAGE, NUMPAGES) — exported as literal placeholder text; imported as literal text.
• Multi-section documents — only the body's first <w:sectPr> is considered.
• Headers / footers from header*.xml / footer*.xml parts are not yet wired into the imported pageMeta.
• Track changes / comments / footnotes — ignored.
• DEFLATE compression — exports are ~2-3× the size of typical .docx. Word reads them without complaint.

Security

importDocx rejects <w:hyperlink> targets matching javascript:, vbscript:, and non-image data: URLs. An 'unsafe-content' ImportWarning is emitted so the host UI can surface "1 unsafe link was removed". See docs/security-review.md.

License

MIT