@stll/folio-core 0.12.0 → 0.14.0

package/dist/docx/ensureParaIds.js

@@ -1,5 +1,6 @@
 import { deterministicHexId } from "../utils/hexId.js";
 import { isXmlNameBoundary } from "./selectiveXmlPatch.js";
+import { loadDocxArchive } from "./server/boundedArchive.js";
 import { TaggedError } from "better-result";
 import JSZip from "jszip";
 //#region src/docx/ensureParaIds.ts
@@ -350,18 +351,17 @@ const hasDigitalSignatureParts = (zip) => Object.values(zip.files).some(({ dir,
 * the buffer is not a WordprocessingML package or a part is malformed.
 */
 const ensureParaIdsInternal = async (docx, options) => {
-	const zip = await JSZip.loadAsync(docx);
-	const xmlPartNames = [];
-	zip.forEach((relativePath, entry) => {
+	const archive = await loadDocxArchive(docx);
+	const xmlPartNames = archive.entries.filter((relativePath) => {
 		const lower = relativePath.toLowerCase();
-		if (!entry.dir && lower.startsWith("word/") && lower.endsWith(".xml")) xmlPartNames.push(relativePath);
+		return lower.startsWith("word/") && lower.endsWith(".xml");
 	});
 	const documentPartName = xmlPartNames.find((name) => name.toLowerCase() === DOCUMENT_PART);
 	if (documentPartName === void 0) throw createEnsureParaIdsError("word/document.xml not found: not a WordprocessingML package");
 	const partTexts = /* @__PURE__ */ new Map();
 	for (const name of xmlPartNames) {
-		const entry = zip.file(name);
-		if (entry) partTexts.set(name, await entry.async("text"));
+		const text = await archive.readEntryString(name);
+		if (text !== null) partTexts.set(name, text);
 	}
 	const taken = /* @__PURE__ */ new Set();
 	for (const text of partTexts.values()) collectExistingParaIds(text, taken);
@@ -390,6 +390,7 @@ const ensureParaIdsInternal = async (docx, options) => {
 		deduplicated: 0,
 		alreadyComplete: true
 	};
+	const zip = await JSZip.loadAsync(docx);
 	if (hasDigitalSignatureParts(zip) && options.allowSignedPackageMutation !== true) throw createEnsureParaIdsError("Refusing to normalize a digitally signed package because rewriting OOXML invalidates its signatures. Warn the user and pass allowSignedPackageMutation only if invalidation is acceptable.");
 	for (const [partPath, content] of updates) zip.file(partPath, content, {
 		compression: "DEFLATE",

package/dist/docx/groupDrawingParser.js

@@ -150,12 +150,17 @@ const groupViewBox = (group, width, height) => {
 	};
 };
 const createSvg = (group, width, height, rels, media) => {
-	const content = getChildElements(group).slice(0, MAX_GROUP_SHAPES).map((child, index) => {
+	const children = getChildElements(group).slice(0, MAX_GROUP_SHAPES);
+	let content = "";
+	for (const [index, child] of children.entries()) {
 		const localName = getLocalName(child.name ?? "");
-		if (localName === "pic") return renderPicture(child, index, rels, media);
-		if (localName !== "wsp") return "";
-		return findChildByLocalName(child, "txbx") ? renderTextBox(child) : renderGeometry(child);
-	}).join("");
+		let piece;
+		if (localName === "pic") piece = renderPicture(child, index, rels, media);
+		else if (localName === "wsp") piece = findChildByLocalName(child, "txbx") ? renderTextBox(child) : renderGeometry(child);
+		else continue;
+		content += piece;
+		if (content.length > MAX_SVG_CHARACTERS) return null;
+	}
 	if (!content) return null;
 	const viewBox = groupViewBox(group, width, height);
 	return `<svg xmlns="http://www.w3.org/2000/svg" viewBox="${viewBox.x} ${viewBox.y} ${viewBox.width} ${viewBox.height}" width="${emuToPixels(width)}" height="${emuToPixels(height)}">${content}</svg>`;

package/dist/docx/hyperlinkParser.js

@@ -1,6 +1,6 @@
 import { getAttribute, getChildElements, mergeXmlnsDeclarations, parseNumericAttribute } from "./xmlParser.js";
-import { parseRun } from "./runParser.js";
 import { sanitizeExternalUrl, sanitizeLinkTarget } from "../utils/urlSecurity.js";
+import { parseRun } from "./runParser.js";
 //#region src/docx/hyperlinkParser.ts
 /**
 * Get the local name of an element (without namespace prefix)

package/dist/docx/imageParser.js

@@ -1,5 +1,7 @@
 import { findByFullName, findChild, getAttribute, getChildElements, parseNumericAttribute } from "./xmlParser.js";
 import { emuToPixels } from "../utils/units.js";
+import { sanitizeExternalUrl } from "../utils/urlSecurity.js";
+import { sanitizeImageSrc } from "../utils/sanitizeImageSrc.js";
 import { WRAP_ELEMENT_NAMES, parsePositionH, parsePositionV, parseWrapElement } from "./drawingUtils.js";
 import { resolveTarget } from "./relsParser.js";
 import { isTextBoxDrawing } from "./textBoxParser.js";
@@ -326,7 +328,8 @@ function parseInline(inlineEl, rels, media) {
 	if (props.alt) image.alt = props.alt;
 	if (props.title) image.title = props.title;
 	if (props.decorative) image.decorative = true;
-	if (imageData.src) image.src = imageData.src;
+	const safeSrc = sanitizeImageSrc(imageData.src);
+	if (safeSrc) image.src = safeSrc;
 	if (imageData.mimeType) image.mimeType = imageData.mimeType;
 	if (imageData.filename) image.filename = imageData.filename;
 	if (padding) image.padding = padding;
@@ -334,8 +337,8 @@ function parseInline(inlineEl, rels, media) {
 	if (crop) image.crop = crop;
 	if (opacity !== void 0) image.opacity = opacity;
 	if (props.hlinkRId && rels) {
-		const href = resolveTarget(rels, props.hlinkRId);
-		if (href) image.hlinkHref = href;
+		const safeHref = sanitizeExternalUrl(resolveTarget(rels, props.hlinkRId));
+		if (safeHref) image.hlinkHref = safeHref;
 	}
 	return image;
 }
@@ -391,7 +394,8 @@ function parseAnchor(anchorEl, rels, media) {
 	if (props.alt) image.alt = props.alt;
 	if (props.title) image.title = props.title;
 	if (props.decorative) image.decorative = true;
-	if (imageData.src) image.src = imageData.src;
+	const safeSrc = sanitizeImageSrc(imageData.src);
+	if (safeSrc) image.src = safeSrc;
 	if (imageData.mimeType) image.mimeType = imageData.mimeType;
 	if (imageData.filename) image.filename = imageData.filename;
 	if (position) image.position = position;
@@ -402,8 +406,8 @@ function parseAnchor(anchorEl, rels, media) {
 	if (layoutInCell !== void 0) image.layoutInCell = layoutInCell;
 	if (allowOverlap !== void 0) image.allowOverlap = allowOverlap;
 	if (props.hlinkRId && rels) {
-		const href = resolveTarget(rels, props.hlinkRId);
-		if (href) image.hlinkHref = href;
+		const safeHref = sanitizeExternalUrl(resolveTarget(rels, props.hlinkRId));
+		if (safeHref) image.hlinkHref = safeHref;
 	}
 	return image;
 }

package/dist/docx/metadataPrivacy.js

@@ -44,6 +44,20 @@ const XML_DECLARATION_PATTERN = /^\s*<\?xml[^?]*\?>/u;
 const MAX_INPUT_BYTES = 50 * 1024 * 1024;
 const MAX_ARCHIVE_ENTRIES = 5e3;
 const MAX_CORE_PROPERTIES_BYTES = 1024 * 1024;
+const CORE_PROPERTIES_PATH_LOWER = "docprops/core.xml";
+/**
+* Look up a package part by its conventional lowercase path, falling back to
+* a case-insensitive scan. OPC part names are case-insensitive, so a
+* producer that wrote `docProps/Core.xml` (or another casing) must still be
+* found — otherwise the privacy rewrite silently no-ops and the unscrubbed
+* buffer round-trips untouched.
+*/
+const findZipEntryCaseInsensitive = (zip, lowerPath) => {
+	const direct = zip.file(lowerPath);
+	if (direct) return direct;
+	for (const [path, file] of Object.entries(zip.files)) if (!file.dir && path.toLowerCase() === lowerPath) return file;
+	return null;
+};
 const loadPrivacyArchive = async (buffer) => {
 	if (buffer.byteLength > MAX_INPUT_BYTES) throw new FolioDocumentPrivacyArchiveError({
 		message: "Document privacy input exceeded the compressed-size limit",
@@ -63,7 +77,7 @@ const loadPrivacyArchive = async (buffer) => {
 		message: "Document privacy input exceeded the package-entry limit",
 		reason: "too-many-entries"
 	});
-	const coreProperties = zip.file("docProps/core.xml");
+	const coreProperties = findZipEntryCaseInsensitive(zip, CORE_PROPERTIES_PATH_LOWER);
 	if (!coreProperties) return zip;
 	const data = "_data" in coreProperties ? coreProperties._data : void 0;
 	const declaredBytes = typeof data === "object" && data !== null && "uncompressedSize" in data ? data.uncompressedSize : void 0;
@@ -99,7 +113,7 @@ const rewriteCorePropertiesPrivacy = (xml, transforms) => {
 const rewriteDocxMetadataPrivacy = async (buffer, { transforms }) => {
 	const appliedTransforms = resolveFolioDocumentPrivacyTransforms(transforms);
 	const zip = await loadPrivacyArchive(buffer);
-	const coreProperties = zip.file("docProps/core.xml");
+	const coreProperties = findZipEntryCaseInsensitive(zip, CORE_PROPERTIES_PATH_LOWER);
 	if (!coreProperties) return {
 		buffer,
 		privacyReport: {
@@ -115,7 +129,7 @@ const rewriteDocxMetadataPrivacy = async (buffer, { transforms }) => {
 			removedMetadataProperties: []
 		}
 	};
-	zip.file("docProps/core.xml", rewritten.xml);
+	zip.file(coreProperties.name, rewritten.xml);
 	return {
 		buffer: await zip.generateAsync({
 			type: "arraybuffer",

package/dist/docx/paragraphParser.js

@@ -2,11 +2,13 @@ import { BorderStyleSchema, FrameWrapSchema, FrameXAlignSchema, FrameYAlignSchem
 import { elementToXml, findChild, findChildren, getAttribute, getChildElements, mergeXmlnsDeclarations, parseBooleanElement, parseNumericAttribute } from "./xmlParser.js";
 import { parseRun, parseRunProperties } from "./runParser.js";
 import { parseFieldType } from "./fieldParser.js";
+import { isValidHexId } from "../utils/hexId.js";
 import { parseBookmarkEnd as parseBookmarkEnd$1, parseBookmarkStart as parseBookmarkStart$1 } from "./bookmarkParser.js";
 import { parseHyperlink as parseHyperlink$1 } from "./hyperlinkParser.js";
 import { consolidateParagraphContent } from "./runConsolidator.js";
 import { parseSdtProperties } from "./sdtProperties.js";
 import { parseSectionProperties } from "./sectionParser.js";
+import { normalizeRevisionId } from "@stll/docx-core/model";
 //#region src/docx/paragraphParser.ts
 /**
 * Extract plain text from a math element (recursive text content extraction)
@@ -486,11 +488,13 @@ function parseTrackedChangeInfo(node) {
 	const rawDate = getAttribute(node, "w", "date");
 	const author = rawAuthor?.trim() ?? "";
 	const date = rawDate?.trim() ?? "";
+	const initials = (getAttribute(node, "w", "initials") ?? "").trim();
 	const info = {
-		id: Number.isInteger(parsedId) && parsedId >= 0 ? parsedId : 0,
+		id: normalizeRevisionId(parsedId),
 		author: author.length > 0 ? author : "Unknown"
 	};
 	if (date.length > 0) info.date = date;
+	if (initials.length > 0) info.initials = initials;
 	return info;
 }
 function parsePropertyChangeInfo(node) {
@@ -939,9 +943,9 @@ function parseParagraph(node, styles, theme, numbering, rels = null, media = nul
 		content: []
 	};
 	const paraId = getAttribute(node, "w14", "paraId") ?? getAttribute(node, "w", "paraId");
-	if (paraId) paragraph.paraId = paraId;
+	if (paraId && isValidHexId(paraId)) paragraph.paraId = paraId;
 	const textId = getAttribute(node, "w14", "textId") ?? getAttribute(node, "w", "textId");
-	if (textId) paragraph.textId = textId;
+	if (textId && isValidHexId(textId)) paragraph.textId = textId;
 	if (!options?.inHeaderFooter && paragraphStartsWithRenderedPageBreak(node)) paragraph.renderedPageBreakBefore = true;
 	const pPr = findChild(node, "w", "pPr");
 	if (pPr) {

package/dist/docx/rezip.js

@@ -4,7 +4,8 @@ import { parseEndnotes, parseFootnotes } from "./footnoteParser.js";
 import { assertValidFolioDocumentModel } from "./modelValidation.js";
 import { isPreservableDocxEntry } from "./unzip.js";
 import { applyReplyThreadMarkers } from "./commentReplyMarkers.js";
-import { buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds, extractParagraphXml } from "./selectiveXmlPatch.js";
+import { withoutOrphanCommentRanges } from "./commentRangeIntegrity.js";
+import { buildParagraphOffsetIndex, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds } from "./selectiveXmlPatch.js";
 import { escapeXml } from "./serializer/xmlUtils.js";
 import { ensureThreadedCommentParaIds, serializeComments, serializeCommentsExtended } from "./serializer/commentSerializer.js";
 import { serializeDocument } from "./serializer/documentSerializer.js";
@@ -15,6 +16,7 @@ import { serializeFontTableXml } from "./serializer/fontTableSerializer.js";
 import { serializeSettingsXml } from "./serializer/settingsSerializer.js";
 import { serializeStylesXml } from "./serializer/stylesSerializer.js";
 import { serializeThemeXml } from "./serializer/themeSerializer.js";
+import { isAllowedExternalWatermarkImageUrl } from "../watermark/index.js";
 import { panic } from "better-result";
 import JSZip from "jszip";
 //#region src/docx/rezip.ts
@@ -369,7 +371,7 @@ async function processNewHyperlinks(newHyperlinks, zip, compressionLevel) {
 async function repackDocx(doc, options = {}) {
 	if (!doc.originalBuffer) panic("Cannot repack document: no original buffer for round-trip. Use createDocx() for new documents.");
 	const { compressionLevel = 6, updateModifiedDate = true, modifiedBy } = options;
-	const exportDocument = doc;
+	const exportDocument = withoutOrphanCommentRanges(doc);
 	const originalZip = await JSZip.loadAsync(doc.originalBuffer);
 	const newZip = new JSZip();
 	for (const [path, file] of Object.entries(originalZip.files)) {
@@ -431,7 +433,7 @@ async function repackDocx(doc, options = {}) {
 */
 async function repackDocxFromRaw(doc, rawContent, options = {}) {
 	const { compressionLevel = 6, updateModifiedDate = true, modifiedBy } = options;
-	const exportDocument = doc;
+	const exportDocument = withoutOrphanCommentRanges(doc);
 	const newZip = new JSZip();
 	for (const [path, file] of Object.entries(rawContent.originalZip.files)) {
 		if (file.dir) {
@@ -854,10 +856,11 @@ async function rebindWatermarkRelIds(doc, zip, compressionLevel) {
 	const changedPaths = /* @__PURE__ */ new Set();
 	for (const { watermark, relsPath, partPath } of pending) {
 		let canonical;
-		if (watermark.imageTarget !== void 0) canonical = watermark.imageTargetExternal ? {
+		if (watermark.imageTarget !== void 0) if (watermark.imageTargetExternal) canonical = isAllowedExternalWatermarkImageUrl(watermark.imageTarget) ? {
 			mode: "external",
 			url: watermark.imageTarget
-		} : {
+		} : resolveCanonical(watermark.imageRId);
+		else canonical = {
 			mode: "internal",
 			absolute: watermark.imageTarget
 		};
@@ -1000,15 +1003,23 @@ async function patchNotePartIntoZip(conventionalLowerPath, currentXml, baselineF
 * (re-parsed original) serialization — i.e. the ones actually edited. A
 * paragraph is only considered when its `paraId` resolves uniquely in both,
 * so it can be spliced safely.
+*
+* Builds one {@link buildParagraphOffsetIndex} per side (a single linear scan
+* each) instead of calling `extractParagraphXml` per candidate id, which
+* re-scanned the whole XML per id — O(note count * XML size) for a document
+* with many footnotes/endnotes. The index turns each lookup below into O(1).
 */
 function collectChangedNoteParaIds(baselineXml, currentXml) {
 	const changed = /* @__PURE__ */ new Set();
 	const baselineIds = collectParaIds(baselineXml);
+	const baselineOffsets = buildParagraphOffsetIndex(baselineXml);
+	const currentOffsets = buildParagraphOffsetIndex(currentXml);
 	for (const [id, count] of collectParaIds(currentXml)) {
 		if (count !== 1 || baselineIds.get(id) !== 1) continue;
-		const before = extractParagraphXml(baselineXml, id);
-		const after = extractParagraphXml(currentXml, id);
-		if (before !== null && after !== null && before !== after) changed.add(id);
+		const beforeRange = baselineOffsets.get(id);
+		const afterRange = currentOffsets.get(id);
+		if (!beforeRange || !afterRange) continue;
+		if (baselineXml.slice(beforeRange.start, beforeRange.end) !== currentXml.slice(afterRange.start, afterRange.end)) changed.add(id);
 	}
 	return changed;
 }

package/dist/docx/runParser.js

@@ -5,8 +5,17 @@ import { parseGroupDrawing } from "./groupDrawingParser.js";
 import { parseShapeFromDrawing, shouldPreserveRawShapeDrawing } from "./shapeParser.js";
 import { parseVmlImageContent } from "./vmlImageParser.js";
 import { resolveThemeFontRef } from "./themeParser.js";
+import { normalizeRevisionId } from "@stll/docx-core/model";
 //#region src/docx/runParser.ts
 /**
+* Sanity cap on `w:lang` `@w:val`/`@w:eastAsia`/`@w:bidi` tag length. BCP-47
+* tags top out well under this; a hostile/corrupt tag here would drive the
+* hyphenation dictionary lookup and segmenter cache keying with an
+* attacker-sized string per run.
+*/
+const MAX_LANGUAGE_TAG_LENGTH = 35;
+const truncateLanguageTag = (value) => value === void 0 ? void 0 : value.slice(0, MAX_LANGUAGE_TAG_LENGTH);
+/**
 * Parse color value from attributes
 */
 function parseColorValue(rgb, themeColor, themeTint, themeShade) {
@@ -267,9 +276,9 @@ function parseRunProperties(rPr, theme, _styles) {
 	}
 	const lang = propertyChildren.lang;
 	if (lang) {
-		const val = getAttribute(lang, "w", "val") || void 0;
-		const eastAsia = getAttribute(lang, "w", "eastAsia") || void 0;
-		const bidi = getAttribute(lang, "w", "bidi") || void 0;
+		const val = truncateLanguageTag(getAttribute(lang, "w", "val") || void 0);
+		const eastAsia = truncateLanguageTag(getAttribute(lang, "w", "eastAsia") || void 0);
+		const bidi = truncateLanguageTag(getAttribute(lang, "w", "bidi") || void 0);
 		if (val || eastAsia || bidi) formatting.language = {
 			...val ? { val } : {},
 			...eastAsia ? { eastAsia } : {},
@@ -332,7 +341,7 @@ function parsePropertyChangeInfo(changeElement) {
 	const date = (getAttribute(changeElement, "w", "date") ?? "").trim();
 	const rsid = (getAttribute(changeElement, "w", "rsid") ?? "").trim();
 	const info = {
-		id: Number.isInteger(parsedId) && parsedId >= 0 ? parsedId : 0,
+		id: normalizeRevisionId(parsedId),
 		author: author.length > 0 ? author : "Unknown"
 	};
 	if (date.length > 0) info.date = date;

package/dist/docx/sectionParser.js

@@ -2,7 +2,20 @@ import { BorderStyleSchema, ThemeColorSlotSchema, narrowEnum } from "./parserEnu
 import { findChild, findChildren, getAttribute, getChildElements, getLocalName, parseBooleanElement, parseNumericAttribute } from "./xmlParser.js";
 import { parseFooterReference, parseHeaderReference } from "./headerFooterRefParser.js";
 import { parseEndnoteProperties, parseFootnoteProperties } from "./notePropertiesParser.js";
+import { normalizeRevisionId } from "@stll/docx-core/model";
 //#region src/docx/sectionParser.ts
+/**
+* Sanity cap on `w:cols/@w:num`. Word's column picker tops out well below
+* this; a hostile/corrupt value here would force the layout engine to
+* generate a proportional band per column.
+*/
+const MAX_SECTION_COLUMNS = 45;
+/**
+* Sanity cap on `w:pgSz` `@w:w`/`@w:h`, in twips (~22in). Matches the
+* existing `w:defaultTabStop` cap (`settingsParser.ts`) and bounds the
+* ruler tick generators, which are sized off the page dimensions.
+*/
+const MAX_PAGE_DIMENSION_TWIPS = 31680;
 const serializedSectionPropertyChildNames = /* @__PURE__ */ new Set([
 	"headerReference",
 	"footerReference",
@@ -55,7 +68,7 @@ function parsePropertyChangeInfo(node) {
 	const date = (getAttribute(node, "w", "date") ?? "").trim();
 	const rsid = (getAttribute(node, "w", "rsid") ?? "").trim();
 	const info = {
-		id: Number.isInteger(parsedId) && parsedId >= 0 ? parsedId : 0,
+		id: normalizeRevisionId(parsedId),
 		author: author.length > 0 ? author : "Unknown"
 	};
 	if (date.length > 0) info.date = date;
@@ -170,9 +183,9 @@ function parseSectionProperties(sectPr) {
 	const pgSz = findChild(sectPr, "w", "pgSz");
 	if (pgSz) {
 		const w = parseNumericAttribute(pgSz, "w", "w");
-		if (w !== void 0) props.pageWidth = w;
+		if (w !== void 0) props.pageWidth = Math.min(w, MAX_PAGE_DIMENSION_TWIPS);
 		const h = parseNumericAttribute(pgSz, "w", "h");
-		if (h !== void 0) props.pageHeight = h;
+		if (h !== void 0) props.pageHeight = Math.min(h, MAX_PAGE_DIMENSION_TWIPS);
 		const orientation = parseOrientation(getAttribute(pgSz, "w", "orient"));
 		if (orientation) props.orientation = orientation;
 	}
@@ -196,7 +209,7 @@ function parseSectionProperties(sectPr) {
 	const cols = findChild(sectPr, "w", "cols");
 	if (cols) {
 		const num = parseNumericAttribute(cols, "w", "num");
-		if (num !== void 0) props.columnCount = num;
+		if (num !== void 0) props.columnCount = Math.min(num, MAX_SECTION_COLUMNS);
 		const space = parseNumericAttribute(cols, "w", "space");
 		if (space !== void 0) props.columnSpace = space;
 		const equalWidth = getAttribute(cols, "w", "equalWidth");

package/dist/docx/selectiveSave.js

@@ -2,7 +2,9 @@ import { parseNumbering } from "./numberingParser.js";
 import { RELATIONSHIP_TYPES } from "./relsParser.js";
 import { parseCommentsExtended } from "./commentParser.js";
 import { validateFolioDocumentModel } from "./modelValidation.js";
+import { isPreservableDocxEntry } from "./unzip.js";
 import { hasUnsynthesizedReplyRanges } from "./commentReplyMarkers.js";
+import { withoutOrphanCommentRanges } from "./commentRangeIntegrity.js";
 import { buildPatchedDocumentXml, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds } from "./selectiveXmlPatch.js";
 import { ensureThreadedCommentParaIds, serializeComments, serializeCommentsExtended } from "./serializer/commentSerializer.js";
 import { serializeDocument } from "./serializer/documentSerializer.js";
@@ -182,6 +184,7 @@ async function attemptSelectiveSave(doc, originalBuffer, options) {
 	if (hasNewImagesOrHyperlinks(content)) return null;
 	if (hasUnmaterializedHeaderFooter(doc)) return null;
 	if (hasModelDrivenPictureWatermark(doc)) return null;
+	if (withoutOrphanCommentRanges(doc) !== doc) return null;
 	if (hasUnsynthesizedReplyRanges(doc)) return null;
 	if (!validateFolioDocumentModel(doc).valid) return null;
 	const comments = doc.package.document.comments ?? [];
@@ -189,6 +192,7 @@ async function attemptSelectiveSave(doc, originalBuffer, options) {
 	const headerFooterUpdates = collectHeaderFooterUpdates(doc);
 	try {
 		const zip = await (await import("jszip")).default.loadAsync(originalBuffer);
+		for (const [path, file] of Object.entries(zip.files)) if (!file.dir && !isPreservableDocxEntry(path)) return null;
 		const updates = /* @__PURE__ */ new Map();
 		if (changedParaIds.size > 0) {
 			const docXmlFile = zip.file("word/document.xml");

package/dist/docx/selectiveXmlPatch.d.ts

@@ -33,6 +33,27 @@ declare function findParagraphOffsets(xml: string, paraId: string): {
  * from a fully serialized document.xml string.
  */
 declare function extractParagraphXml(serializedXml: string, paraId: string): string | null;
+type ParagraphOffsets = {
+  start: number;
+  end: number;
+};
+/**
+ * Single linear-scan index of every `<w:p>` element's start/end offsets,
+ * keyed by `w14:paraId`. Generalizes {@link findParagraphOffsets} (one
+ * regex-scan-plus-depth-walk per lookup) to build every paragraph's offsets
+ * in a single pass, so a caller that needs many paragraphs from the same XML
+ * (e.g. `collectChangedNoteParaIds` in rezip.ts, walking every note paraId)
+ * does O(1) map lookups afterward instead of re-scanning the whole XML once
+ * per id — O(ids * XML length) collapses to O(XML length).
+ *
+ * A `<w:p>` nested inside another (e.g. inside `mc:AlternateContent`) is
+ * indexed too via a depth stack, matching every element `findParagraphOffsets`
+ * can resolve. A paraId that appears on more than one element is ambiguous —
+ * mirroring {@link findParagraphOffsets}'s single-match requirement — and is
+ * omitted from the index, so a lookup misses exactly where the per-id
+ * function would return null.
+ */
+declare function buildParagraphOffsetIndex(xml: string): Map<string, ParagraphOffsets>;
 /**
  * Count <w:p> elements in an XML string (top-level paragraph count).
  * Counts opening <w:p tags that are NOT self-closing.
@@ -109,4 +130,4 @@ declare function collectChangedNumberingDefs(baselineXml: string, currentXml: st
  */
 declare function buildPatchedNumberingXml(originalXml: string, currentXml: string, changed: ChangedNumberingDefs): string | null;
 //#endregion
-export { ChangedNumberingDefs, PatchSafetyOptions, PatchValidationResult, buildPatchedDocumentXml, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds, countParagraphElements, extractParagraphXml, findParagraphOffsets, isXmlNameBoundary, validatePatchSafety };
+export { ChangedNumberingDefs, ParagraphOffsets, PatchSafetyOptions, PatchValidationResult, buildParagraphOffsetIndex, buildPatchedDocumentXml, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds, countParagraphElements, extractParagraphXml, findParagraphOffsets, isXmlNameBoundary, validatePatchSafety };

package/dist/docx/selectiveXmlPatch.js

@@ -74,6 +74,66 @@ function extractParagraphXml(serializedXml, paraId) {
 	return serializedXml.slice(offsets.start, offsets.end);
 }
 /**
+* Single linear-scan index of every `<w:p>` element's start/end offsets,
+* keyed by `w14:paraId`. Generalizes {@link findParagraphOffsets} (one
+* regex-scan-plus-depth-walk per lookup) to build every paragraph's offsets
+* in a single pass, so a caller that needs many paragraphs from the same XML
+* (e.g. `collectChangedNoteParaIds` in rezip.ts, walking every note paraId)
+* does O(1) map lookups afterward instead of re-scanning the whole XML once
+* per id — O(ids * XML length) collapses to O(XML length).
+*
+* A `<w:p>` nested inside another (e.g. inside `mc:AlternateContent`) is
+* indexed too via a depth stack, matching every element `findParagraphOffsets`
+* can resolve. A paraId that appears on more than one element is ambiguous —
+* mirroring {@link findParagraphOffsets}'s single-match requirement — and is
+* omitted from the index, so a lookup misses exactly where the per-id
+* function would return null.
+*/
+function buildParagraphOffsetIndex(xml) {
+	const seenCount = /* @__PURE__ */ new Map();
+	const ranges = /* @__PURE__ */ new Map();
+	const stack = [];
+	let pos = 0;
+	const noteOpen = (paraId) => {
+		if (paraId) seenCount.set(paraId, (seenCount.get(paraId) ?? 0) + 1);
+	};
+	const noteRange = (paraId, start, end) => {
+		if (paraId) ranges.set(paraId, {
+			start,
+			end
+		});
+	};
+	while (pos < xml.length) {
+		const tagStart = xml.indexOf("<", pos);
+		if (tagStart === -1) break;
+		if (xml.startsWith("</w:p>", tagStart)) {
+			const frame = stack.pop();
+			const end = tagStart + 6;
+			if (frame) noteRange(frame.paraId, frame.start, end);
+			pos = end;
+			continue;
+		}
+		if (!xml.startsWith("<w:p", tagStart) || !isXmlNameBoundary(xml[tagStart + 4])) {
+			pos = tagStart + 1;
+			continue;
+		}
+		const tagEnd = xml.indexOf(">", tagStart);
+		if (tagEnd === -1) break;
+		const openTag = xml.slice(tagStart, tagEnd + 1);
+		const paraId = /\bw14:paraId="(?<id>[^"]*)"/u.exec(openTag)?.groups?.["id"];
+		noteOpen(paraId);
+		if (xml[tagEnd - 1] === "/") noteRange(paraId, tagStart, tagEnd + 1);
+		else stack.push({
+			start: tagStart,
+			paraId
+		});
+		pos = tagEnd + 1;
+	}
+	const index = /* @__PURE__ */ new Map();
+	for (const [id, range] of ranges) if (seenCount.get(id) === 1) index.set(id, range);
+	return index;
+}
+/**
 * Count <w:p> elements in an XML string (top-level paragraph count).
 * Counts opening <w:p tags that are NOT self-closing.
 */
@@ -484,4 +544,4 @@ function escapeRegExp(str) {
 	return str.replace(/[.*+?^${}()|[\]\\]/gu, "\\$&");
 }
 //#endregion
-export { buildPatchedDocumentXml, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds, countParagraphElements, extractParagraphXml, findParagraphOffsets, isXmlNameBoundary, validatePatchSafety };
+export { buildParagraphOffsetIndex, buildPatchedDocumentXml, buildPatchedNoteXml, buildPatchedNumberingXml, collectChangedNumberingDefs, collectParaIds, countParagraphElements, extractParagraphXml, findParagraphOffsets, isXmlNameBoundary, validatePatchSafety };

package/dist/docx/serializer/blockSdtSerializer.js

@@ -1,3 +1,4 @@
+import { isSingleWellFormedElement } from "./xmlUtils.js";
 import { reconcileRawSdtPr } from "../sdtPropertiesPatch.js";
 //#region src/docx/serializer/blockSdtSerializer.ts
 function escapeXmlAttr(value) {
@@ -73,14 +74,14 @@ function extractDateFullDate(blockSdt) {
 }
 function serializeBlockSdt(blockSdt, serializeChild) {
 	const props = blockSdt.properties;
-	const baseSdtPr = props.rawPropertiesXml ?? serializeFallbackSdtPr(props);
+	const baseSdtPr = props.rawPropertiesXml && isSingleWellFormedElement(props.rawPropertiesXml, "sdtPr") ? props.rawPropertiesXml : serializeFallbackSdtPr(props);
 	const dateFullDate = extractDateFullDate(blockSdt);
 	const dropdownLastValue = extractDropdownLastValue(blockSdt);
 	const sdtPrXml = reconcileRawSdtPr(baseSdtPr, props, {
 		...dateFullDate !== void 0 ? { dateFullDate } : {},
 		...dropdownLastValue !== void 0 ? { dropdownLastValue } : {}
 	});
-	const sdtEndPrXml = props.rawEndPropertiesXml ?? "";
+	const sdtEndPrXml = props.rawEndPropertiesXml && isSingleWellFormedElement(props.rawEndPropertiesXml, "sdtEndPr") ? props.rawEndPropertiesXml : "";
 	const contentXml = blockSdt.content.map(serializeChild).join("");
 	return `<w:sdt>${sdtPrXml}${sdtEndPrXml}${props.rawSdtChildrenBeforeContent ?? ""}<w:sdtContent>${contentXml}</w:sdtContent>${props.rawSdtChildrenAfterContent ?? ""}</w:sdt>`;
 }

package/dist/docx/serializer/commentSerializer.js

@@ -27,8 +27,8 @@ function serializeRunContent(run) {
 */
 function commentParagraphOpenTag(p) {
 	const attrs = [];
-	if (p.paraId) attrs.push(`w14:paraId="${p.paraId}"`);
-	if (p.textId) attrs.push(`w14:textId="${p.textId}"`);
+	if (p.paraId) attrs.push(`w14:paraId="${escapeXml(p.paraId)}"`);
+	if (p.textId) attrs.push(`w14:textId="${escapeXml(p.textId)}"`);
 	return attrs.length > 0 ? `<w:p ${attrs.join(" ")}>` : "<w:p>";
 }
 function serializeParagraph(p) {
@@ -173,8 +173,8 @@ function serializeCommentsExtended(comments) {
 	if (!entries) return null;
 	let xml = COMMENTS_EXTENDED_HEADER;
 	for (const entry of entries) {
-		const parentAttr = entry.paraIdParent !== void 0 ? ` w15:paraIdParent="${entry.paraIdParent}"` : "";
-		xml += `<w15:commentEx w15:paraId="${entry.paraId}"${parentAttr} w15:done="${entry.done ? "1" : "0"}"/>`;
+		const parentAttr = entry.paraIdParent !== void 0 ? ` w15:paraIdParent="${escapeXml(entry.paraIdParent)}"` : "";
+		xml += `<w15:commentEx w15:paraId="${escapeXml(entry.paraId)}"${parentAttr} w15:done="${entry.done ? "1" : "0"}"/>`;
 	}
 	xml += "</w15:commentsEx>";
 	return xml;