web-doc 0.10.0 → 0.12.0

package/dist/adapters/office.js

@@ -86,7 +86,6 @@ export class OfficeDocumentAdapter {
             useGoogleFonts: false,
             maxZipEntryBytes: context.limits.maxZipEntryBytes,
             mode: "main",
-            ...(reopening ? { progressiveLayout: true } : {}),
         };
         try {
             const kind = kindFor(format);
@@ -113,7 +112,10 @@ export class OfficeDocumentAdapter {
             }
             const buffer = exactArrayBuffer(data);
             if (kind === "presentation") {
-                const backend = await this.#loadPptx(buffer, engineOptions);
+                const backend = await this.#loadPptx(buffer, {
+                    ...engineOptions,
+                    ...(reopening ? { progressiveLayout: true } : {}),
+                });
                 throwIfAborted(context.signal, backend);
                 context.reportProgress({
                     phase: "parsing",

package/dist/adapters/pdf.js

@@ -90,14 +90,16 @@ class PdfJsBackend {
     #module;
     #document;
     #worker;
+    #workerPort;
     #loadingTask;
     #maxPixels;
     #pages = new Map();
     #closed = false;
-    constructor(module, document, worker, loadingTask, maxPixels) {
+    constructor(module, document, worker, workerPort, loadingTask, maxPixels) {
         this.#module = module;
         this.#document = document;
         this.#worker = worker;
+        this.#workerPort = workerPort;
         this.#loadingTask = loadingTask;
         this.#maxPixels = maxPixels;
         this.pageCount = document.numPages;
@@ -155,7 +157,7 @@ class PdfJsBackend {
             const document = await withTimeout(loadingTask.promise, context.limits.maxOperationMs, "PDF parsing");
             if (context.signal.aborted)
                 throw abortError();
-            return new PdfJsBackend(module, document, worker, loadingTask, context.limits.maxDecodedPixels);
+            return new PdfJsBackend(module, document, worker, workerPort, loadingTask, context.limits.maxDecodedPixels);
         }
         catch (error) {
             await loadingTask.destroy().catch(() => undefined);
@@ -168,9 +170,14 @@ class PdfJsBackend {
         }
     }
     async renderPage(target, pageIndex, zoom, devicePixelRatio, signal) {
-        this.#assertOpen();
-        const page = await this.#page(pageIndex);
-        if (signal?.aborted)
+        if (this.#closed || signal?.aborted)
+            throw abortError();
+        const page = await this.#page(pageIndex).catch((error) => {
+            if (this.#closed || signal?.aborted)
+                throw abortError();
+            throw error;
+        });
+        if (this.#closed || signal?.aborted)
             throw abortError();
         const viewport = page.getViewport({ scale: CSS_UNITS * zoom });
         const dpr = Math.max(1, devicePixelRatio);
@@ -200,11 +207,12 @@ class PdfJsBackend {
         signal?.addEventListener("abort", onAbort, { once: true });
         try {
             await renderTask.promise;
-            if (signal?.aborted)
+            if (this.#closed || signal?.aborted)
                 throw abortError();
         }
         catch (error) {
-            if (signal?.aborted ||
+            if (this.#closed ||
+                signal?.aborted ||
                 error instanceof this.#module.RenderingCancelledException ||
                 (error instanceof Error && error.name === "RenderingCancelledException"))
                 throw abortError();
@@ -251,11 +259,14 @@ class PdfJsBackend {
         this.#closed = true;
         this.#pages.clear();
         try {
-            await this.#document.cleanup();
+            // Destroy cancels and awaits active renders. PDF.js cleanup is only
+            // safe while idle, so it must not run before retiring this document.
             await this.#loadingTask.destroy();
         }
         finally {
             this.#worker.destroy();
+            // PDFWorker does not terminate an externally supplied port; we own it.
+            this.#workerPort.terminate();
         }
     }
     #page(pageIndex) {

package/dist/edit/docx/engine.d.ts

@@ -3,10 +3,12 @@ import type { BatchMode, EditEngine, EngineBatch, EngineChange, MaterializedDocu
 import { OoxmlPackage } from "../ooxml/package.js";
 import type { EditFindOptions, EditOperation, ElementQuery, OperationIssue, PagePoint, TextTarget } from "../types.js";
 import { DocxModel, type AnyRecord } from "./model.js";
-import type { DocxElement, DocxRevision } from "./types.js";
+import type { DocxElement, DocxRevision, DocxTextStyle } from "./types.js";
+import { type TextSpan } from "../range-style.js";
 /** Reads the DOCX session adds on top of the core, served by the engine and the worker client alike. */
 export interface DocxEngineReads {
     revisions(id: string, signal: AbortSignal): Promise<readonly DocxRevision[]>;
+    textStyle(id: string, span: TextSpan | undefined, signal: AbortSignal): Promise<Partial<DocxTextStyle> | undefined>;
 }
 export declare class DocxEditEngine implements EditEngine, DocxEngineReads {
     #private;
@@ -25,6 +27,12 @@ export declare class DocxEditEngine implements EditEngine, DocxEngineReads {
     apply(input: EngineBatch | readonly EditOperation[], signal: AbortSignal): Promise<EngineChange>;
     /** The revisions of a paragraph; none for another element or an unknown id. */
     revisions(id: string, signal: AbortSignal): Promise<readonly DocxRevision[]>;
+    /**
+     * The style a span of a paragraph shows: what every run it covers shares;
+     * the paragraph mark's for an empty paragraph. None for another element,
+     * an unknown id or a span past the text.
+     */
+    textStyle(id: string, span: TextSpan | undefined, signal: AbortSignal): Promise<Partial<DocxTextStyle> | undefined>;
     materialize(purposeOrSignal?: "show" | "save" | AbortSignal, options?: MaterializeOptions, signal?: AbortSignal): Promise<Uint8Array>;
     /**
      * `save` is the package as the session changed it: ids written only on

package/dist/edit/docx/engine.js

@@ -7,9 +7,11 @@ import { NO_PAGE, toElement } from "./elements.js";
 import { docxHandlers } from "./handlers.js";
 import { freshParagraphId, paragraphsOf } from "./ids.js";
 import { DocxModel } from "./model.js";
+import { resolveTextStyle } from "./style.js";
 import { issueCollector, } from "./operations.js";
 import { docxOperationSchemas } from "./schemas.js";
 import { revisionsOf } from "./tracked.js";
+import { runsCovering, sharedStyle, spanFits, } from "../range-style.js";
 import { attributeProblem, namespacePatches } from "./write.js";
 /*
  * The DOCX edit engine: the package layer under a block index of the body
@@ -303,6 +305,26 @@ export class DocxEditEngine {
             return [];
         return revisionsOf(model.document, record.node);
     }
+    /**
+     * The style a span of a paragraph shows: what every run it covers shares;
+     * the paragraph mark's for an empty paragraph. None for another element,
+     * an unknown id or a span past the text.
+     */
+    async textStyle(id, span, signal) {
+        const model = await this.model(signal);
+        const record = model.byId.get(id);
+        if (!record || record.kind !== "paragraph")
+            return undefined;
+        const { text, items } = record.text;
+        const range = span ?? { start: 0, end: text.length };
+        if (!spanFits(range, text.length))
+            return undefined;
+        const covered = runsCovering(items, range);
+        const mark = record.pPr?.children.find((child) => child.local === "rPr");
+        return sharedStyle(covered.length > 0
+            ? covered.map((item) => resolveTextStyle(model.styles, record.pPr, item.rPr))
+            : [resolveTextStyle(model.styles, record.pPr, mark)]);
+    }
     async materialize(purposeOrSignal = "show", options = {}, signal = new AbortController().signal) {
         return (await this.materializeDocument(purposeOrSignal, options, signal))
             .bytes;

package/dist/edit/docx/provider.d.ts

@@ -2,7 +2,8 @@ import type { EditEngineContext } from "../engine.js";
 import { type OoxmlEditProviderOptions } from "../ooxml/worker.js";
 import { WorkerEngineClient } from "../worker-engine.js";
 import type { DocxEngineReads } from "./engine.js";
-import type { DocxRevision } from "./types.js";
+import type { TextSpan } from "../range-style.js";
+import type { DocxRevision, DocxTextStyle } from "./types.js";
 export type DocxEditProviderOptions = OoxmlEditProviderOptions;
 /**
  * Starts the OOXML edit worker for `original` as a DOCX engine. Loaded
@@ -12,5 +13,6 @@ export declare function loadDocxEditEngine(original: Uint8Array, context: EditEn
 export declare class DocxEditEngineClient extends WorkerEngineClient implements DocxEngineReads {
     readonly schemas: import("../types.js").OperationSchemaSet;
     revisions(id: string, signal: AbortSignal): Promise<readonly DocxRevision[]>;
+    textStyle(id: string, span: TextSpan | undefined, signal: AbortSignal): Promise<Partial<DocxTextStyle> | undefined>;
     start(original: Uint8Array): Promise<void>;
 }

package/dist/edit/docx/provider.js

@@ -24,6 +24,9 @@ export class DocxEditEngineClient extends WorkerEngineClient {
     revisions(id, signal) {
         return this.request("edit-docx-revisions", { id }, signal);
     }
+    textStyle(id, span, signal) {
+        return this.request("edit-docx-text-style", { id, span }, signal);
+    }
     async start(original) {
         const data = original.slice().buffer;
         const open = {

package/dist/edit/docx/session.d.ts

@@ -1,7 +1,7 @@
 import type { DescribeOptions, DocumentDescription, EditCheckpoint, OutlineOptions, OutlineResult, TargetCandidate, TargetQuery, ToolCall, ToolCallOptions, ToolResult, ToolSet } from "../ai/types.js";
 import type { EditSessionAccess, EditSessionCore } from "../engine.js";
-import type { ApplyOptions, AssetOptions, EditFindOptions, EditOperation, EditReceipt, EditState, ElementQuery, HistoryOptions, OperationSchemaSet, PagePoint, PageRect, ReadItem, ReadOptions, ReadResult, SavedDocument, TextTarget } from "../types.js";
-import type { DocxDeleteElementOperation, DocxEditSession, DocxElement, DocxFields, DocxInsertImageOperation, DocxInsertParagraphOperation, DocxInsertTableOperation, DocxMoveElementOperation, DocxOperation, DocxReplaceTextOperation, DocxRevision, DocxSaveOptions, DocxSetParagraphStyleOperation, DocxSetTableCellOperation, DocxSetTextStyleOperation } from "./types.js";
+import type { ApplyOptions, AssetOptions, EditFindOptions, EditOperation, EditReceipt, EditState, ElementQuery, HistoryOptions, OperationSchemaSet, PagePoint, PageRect, ReadItem, ReadOptions, ReadResult, SavedDocument, TextRange, TextTarget } from "../types.js";
+import type { DocxDeleteElementOperation, DocxEditSession, DocxElement, DocxFields, DocxInsertImageOperation, DocxInsertParagraphOperation, DocxInsertTableOperation, DocxMoveElementOperation, DocxOperation, DocxReplaceTextOperation, DocxRevision, DocxSaveOptions, DocxSetParagraphStyleOperation, DocxSetTableCellOperation, DocxSetTextStyleOperation, DocxTextStyle } from "./types.js";
 /** Per paragraph id (eight hex digits), the union of its runs per page. */
 type Placement = ReadonlyMap<string, ReadonlyMap<number, PageRect>>;
 export declare class DocxSession implements DocxEditSession {
@@ -33,6 +33,10 @@ export declare class DocxSession implements DocxEditSession {
     get tools(): ToolSet;
     callTool(call: ToolCall, options?: ToolCallOptions): Promise<ToolResult>;
     getRevisions(elementId: string, options?: ReadOptions): Promise<ReadResult<DocxRevision>>;
+    getTextStyle(fields: {
+        readonly target: string;
+        readonly range?: TextRange;
+    }, options?: ReadOptions): Promise<ReadItem<Partial<DocxTextStyle>>>;
     replaceText(fields: DocxFields<DocxReplaceTextOperation>, options?: ApplyOptions): Promise<EditReceipt>;
     setTextStyle(fields: DocxFields<DocxSetTextStyleOperation>, options?: ApplyOptions): Promise<EditReceipt>;
     setParagraphStyle(fields: DocxFields<DocxSetParagraphStyleOperation>, options?: ApplyOptions): Promise<EditReceipt>;

package/dist/edit/docx/session.js

@@ -2,6 +2,7 @@ import { readDescription, readOutline } from "../ai/outline.js";
 import { resolveTargets } from "../ai/targets.js";
 import { buildToolSet, callTool as runTool } from "../ai/tools.js";
 import { ViewerError } from "../../errors.js";
+import { spanOnTarget } from "../range-style.js";
 export class DocxSession {
     format = "docx";
     #core;
@@ -134,6 +135,10 @@ export class DocxSession {
     getRevisions(elementId, options) {
         return this.#core.readItems(options, (engine, signal) => docxReads(engine).revisions(elementId, signal));
     }
+    async getTextStyle(fields, options) {
+        const span = spanOnTarget(fields.target, fields.range);
+        return this.#core.readItem(options, (engine, signal) => docxReads(engine).textStyle(fields.target, span, signal));
+    }
     replaceText(fields, options) {
         return this.apply([{ op: "replaceText", ...fields }], options);
     }
@@ -461,7 +466,8 @@ function matchesQuery(element, query) {
 }
 function docxReads(engine) {
     const candidate = engine;
-    if (typeof candidate.revisions !== "function")
+    if (typeof candidate.revisions !== "function" ||
+        typeof candidate.textStyle !== "function")
         throw new ViewerError("edit-unsupported", "The engine does not provide the DOCX reads", { details: { format: "docx", reason: "no-reads" } });
     return candidate;
 }

package/dist/edit/docx/style.d.ts

@@ -34,7 +34,7 @@ export declare class DocxStyles {
     /** A style and the styles it is based on, nearest first; cycles stop. */
     chain(id: string | undefined): StyleRecord[];
 }
-/** `w:b`, `w:i`, `w:default`-style toggles: present without a value means on. */
+/** `w:b`/`w:i` toggle values: a present element without `w:val` means on. */
 export declare function isOn(value: string | undefined): boolean;
 /** The run property holders in resolution order for a run of a paragraph. */
 export declare function runPropertySources(styles: DocxStyles, pPr: XmlElement | undefined, rPr: XmlElement | undefined): readonly XmlElement[];

package/dist/edit/docx/style.js

@@ -95,11 +95,12 @@ export class DocxStyles {
             if (!id)
                 continue;
             const basedOn = part.attribute(node.children.find((child) => child.local === "basedOn") ?? node, "w:val");
+            const defaultAttribute = part.attribute(node, "w:default");
             const record = {
                 id,
                 type: part.attribute(node, "w:type") ?? "paragraph",
                 ...(basedOn && basedOn !== id ? { basedOn } : {}),
-                isDefault: isOn(part.attribute(node, "w:default")),
+                isDefault: defaultAttribute !== undefined && isOn(defaultAttribute),
                 ...pick(node, "rPr", "pPr"),
             };
             styles.set(id, record);
@@ -175,7 +176,7 @@ async function readTheme(pkg, themePart, signal) {
         return none;
     }
 }
-/** `w:b`, `w:i`, `w:default`-style toggles: present without a value means on. */
+/** `w:b`/`w:i` toggle values: a present element without `w:val` means on. */
 export function isOn(value) {
     return value === undefined || !/^(0|false|off)$/i.test(value);
 }

package/dist/edit/docx/types.d.ts

@@ -1,4 +1,4 @@
-import type { ApplyOptions, BinaryData, EditColor, EditElement, EditReceipt, EditSessionBase, ReadOptions, ReadResult, SavedDocument, SaveOptions, TextRange } from "../types.js";
+import type { ApplyOptions, BinaryData, EditColor, EditElement, EditReceipt, EditSessionBase, ReadItem, ReadOptions, ReadResult, SavedDocument, SaveOptions, TextRange } from "../types.js";
 export type DocxElementKind = "paragraph" | "table" | "image" | "other";
 export type DocxTextAlign = "left" | "center" | "right" | "justify";
 /** Word's sixteen highlight colours (`w:highlight/@w:val`). */
@@ -193,6 +193,16 @@ export interface DocxEditSession extends EditSessionBase<DocxOperation, DocxElem
     save(options?: DocxSaveOptions): Promise<SavedDocument>;
     /** The tracked changes a paragraph holds, in document order; empty for other elements. */
     getRevisions(elementId: string, options?: ReadOptions): Promise<ReadResult<DocxRevision>>;
+    /**
+     * The text style a range of a paragraph shows: each property every run it
+     * covers shares, a property they differ on left out. Without a range, the
+     * whole text; a collapsed range reads the run before it, whose style text
+     * typed there takes. `undefined` for an element without text.
+     */
+    getTextStyle(fields: {
+        readonly target: string;
+        readonly range?: TextRange;
+    }, options?: ReadOptions): Promise<ReadItem<Partial<DocxTextStyle>>>;
     /** Replaces the whole text of a paragraph, or the part a range covers. */
     replaceText(fields: DocxFields<DocxReplaceTextOperation>, options?: ApplyOptions): Promise<EditReceipt>;
     /** Changes run properties; unspecified ones keep their bytes. */

package/dist/edit/engine.d.ts

@@ -1,6 +1,6 @@
 import type { DocumentFormat, RegisteredFont, ResourceLimits, TextRun, ViewerWarning } from "../contracts.js";
 import type { EditSession } from "./sessions.js";
-import type { ChangeMode, EditableFormat, EditElement, EditFindOptions, EditOperation, EditSessionBase, ElementQuery, OperationIssue, OperationSchemaSet, PagePoint, ReadItem, ReadOptions, ReadResult, TextTarget } from "./types.js";
+import type { ChangeMode, EditableFormat, EditElement, EditFindOptions, EditOperation, EditSessionBase, ElementQuery, OperationIssue, OperationSchemaSet, PagePoint, ReadItem, ReadOptions, ReadResult, TextTarget, TextAnchorMigration } from "./types.js";
 export interface EditEngineContext {
     readonly format: EditableFormat;
     readonly fileName?: string;
@@ -74,6 +74,7 @@ export interface EngineChange {
     readonly removedIds: readonly string[];
     /** Old id → new id, when the format had to rename an element. */
     readonly remappedIds?: Readonly<Record<string, string>>;
+    readonly textAnchorMigrations?: readonly TextAnchorMigration[];
     /** A superset of the pages whose content changed. */
     readonly changedPages: readonly number[];
     /**

package/dist/edit/pdf/engine/document.d.ts

@@ -1,6 +1,6 @@
 import type { EngineBatch, EngineChange, MaterializedDocument, RestoreTarget } from "../../engine.js";
 import type { EditFindOptions, EditOperation, ElementQuery, OperationIssue, PagePoint, PageRect, TextPosition, TextRange, TextTarget } from "../../types.js";
-import type { PageLayout, PdfElement, PdfOperation, TextFont, TextLayout } from "../types.js";
+import type { PageLayout, PdfElement, PdfOperation, PdfTextParagraph, TextFont, TextLayout } from "../types.js";
 import type { EditWorkerBitmap } from "../../../worker-protocol.js";
 import { FontLibrary } from "./fonts.js";
 import { ImageCache } from "./images.js";
@@ -76,7 +76,9 @@ export declare class PdfEditDocument {
     /** Elements under a point, top-most (drawn last) first. */
     elementsAt(pageIndex: number, point: PagePoint): PdfElement[];
     findText(query: string, options: EditFindOptions): TextTarget[];
-    /** Lines, glyph boxes and styles of a text, text box or table element. */
+    /** Resolves the canonical imported paragraph, when its rows can be grouped safely. */
+    textParagraph(elementId: string): PdfTextParagraph | undefined;
+    /** Lines, glyph boxes and styles of a text, text box, paragraph or table element. */
     textLayout(elementId: string): TextLayout | undefined;
     /** The browser face of the font a text or text box element is drawn in, see `TextFont`. */
     textFont(elementId: string): TextFont | undefined;

package/dist/edit/pdf/engine/document.js

@@ -7,6 +7,8 @@ import { textFaceOf } from "./text-font.js";
 import { fontRequestsOf } from "./text-box.js";
 import { issueCollector, } from "./operations.js";
 import { insertTextBox } from "./text-box.js";
+import { discoverParagraphs, paragraphElement, } from "./paragraph.js";
+import { replaceParagraphText } from "./paragraph-edit.js";
 import { replaceText, setTextStyle } from "./existing-text.js";
 import { deleteElement, moveElement, resizeElement } from "./transform.js";
 import { deletePage, insertPage, movePage, rotatePage } from "./pages.js";
@@ -90,6 +92,7 @@ function referenceOf(operation) {
     return typeof target === "string" ? parseReference(target) : undefined;
 }
 const handlers = {
+    replaceParagraphText: replaceParagraphText,
     insertTextBox: insertTextBox,
     replaceText: replaceText,
     setTextStyle: setTextStyle,
@@ -209,6 +212,7 @@ export class PdfEditDocument {
         const createdIds = [];
         const createdByOperation = [];
         const removedIds = [];
+        const textAnchorMigrations = [];
         const changedPages = new Set();
         const warnings = [];
         // The first change of a signed, certified, tagged or PDF/A file is the
@@ -230,6 +234,8 @@ export class PdfEditDocument {
             createdIds.push(...result.createdIds);
             createdByOperation[operationIndex] = [...result.createdIds];
             removedIds.push(...(result.removedIds ?? []));
+            for (const migration of result.textAnchorMigrations ?? [])
+                textAnchorMigrations.push({ ...migration, operationIndex });
             for (const pageIndex of result.changedPages)
                 changedPages.add(pageIndex);
             warnings.push(...result.warnings);
@@ -238,6 +244,7 @@ export class PdfEditDocument {
         return {
             createdIds,
             removedIds,
+            ...(textAnchorMigrations.length ? { textAnchorMigrations } : {}),
             changedPages: [...changedPages].sort((a, b) => a - b),
             pageCount: this.pageCount,
             warnings,
@@ -368,7 +375,13 @@ export class PdfEditDocument {
         const pageIndex = this.#pageIndexOf(id);
         if (pageIndex === undefined)
             return undefined;
-        return this.#elementsOf(pageIndex).find((element) => element.id === id);
+        const element = this.#elementsOf(pageIndex).find((element) => element.id === id);
+        if (element)
+            return element;
+        const target = this.#pages[pageIndex]?.paragraphs?.get(id);
+        return target?.paragraph.id === id
+            ? paragraphElement(target.paragraph)
+            : undefined;
     }
     /** Elements under a point, top-most (drawn last) first. */
     elementsAt(pageIndex, point) {
@@ -395,25 +408,60 @@ export class PdfEditDocument {
             });
         return targets;
     }
-    /** Lines, glyph boxes and styles of a text, text box or table element. */
+    /** Resolves the canonical imported paragraph, when its rows can be grouped safely. */
+    textParagraph(elementId) {
+        return this.#paragraphOf(elementId)?.paragraph;
+    }
+    #paragraphOf(id) {
+        const pageIndex = this.#pageIndexOf(id);
+        if (pageIndex === undefined)
+            return undefined;
+        this.#elementsOf(pageIndex);
+        return this.#pages[pageIndex]?.paragraphs?.get(id);
+    }
+    /** Lines, glyph boxes and styles of a text, text box, paragraph or table element. */
     textLayout(elementId) {
         const pageIndex = this.#pageIndexOf(elementId);
         if (pageIndex === undefined)
             return undefined;
-        return this.#scanText(pageIndex, (scan) => {
+        const paragraph = this.#paragraphOf(elementId);
+        return this.#scanText(pageIndex, (raw) => {
+            const scan = paragraph?.paragraph.id === elementId
+                ? this.#paragraphScan(raw, paragraph)
+                : raw;
             const element = scan.elements.find((entry) => entry.id === elementId);
-            return element ? layoutOf(this.#pdfium, scan, element) : undefined;
+            const layout = element
+                ? layoutOf(this.#pdfium, scan, element)
+                : undefined;
+            return layout && paragraph?.paragraph.id === elementId
+                ? {
+                    ...layout,
+                    lines: [...layout.lines]
+                        .sort((a, b) => a.range.start.offset - b.range.start.offset)
+                        .map((line) => ({
+                        ...line,
+                        fontSize: paragraph.paragraph.textStyle.fontSize,
+                    })),
+                }
+                : layout;
         });
     }
     /** The browser face of the font a text or text box element is drawn in, see `TextFont`. */
     textFont(elementId) {
         const location = this.#locate(elementId);
-        const element = location && this.getElement(elementId);
-        if (!element || (element.kind !== "text" && element.kind !== "textBox"))
+        if (!location)
+            return undefined;
+        // A paragraph not yet edited is not listed among the elements; its id
+        // still locates its rows.
+        const kind = this.getElement(elementId)?.kind ??
+            (this.#paragraphOf(elementId)?.paragraph.id === elementId
+                ? "paragraph"
+                : undefined);
+        if (kind !== "text" && kind !== "textBox" && kind !== "paragraph")
             return undefined;
         const { lib } = this.#pdfium;
         return this.#withPage(location.pageIndex, (page) => {
-            // A text box's lines share their font; its first line answers.
+            // A text box's lines and a paragraph's rows share their font; the first answers.
             const object = location.indexes
                 .map((index) => lib.FPDFPage_GetObject(page, index))
                 .find((entry) => lib.FPDFPageObj_GetType(entry) === OBJECT_TEXT);
@@ -465,7 +513,10 @@ export class PdfEditDocument {
             if (!Number.isSafeInteger(pixels) ||
                 pixels > this.#limits.maxDecodedPixels)
                 throw new ViewerError("resource-limit", "The rendered page exceeds maxDecodedPixels", { details: { actual: pixels, limit: this.#limits.maxDecodedPixels } });
-            const wanted = new Set(elementIds);
+            const wanted = new Set(elementIds.flatMap((id) => {
+                const paragraph = this.#paragraphOf(id)?.paragraph;
+                return paragraph?.id === id ? paragraph.memberIds : [id];
+            }));
             const records = this.#objectsOf(pageIndex, page);
             const suppressed = [];
             records.forEach((record, index) => {
@@ -529,7 +580,12 @@ export class PdfEditDocument {
         if (pageIndex === undefined ||
             pageIndex !== this.#pageIndexOf(range.end.elementId))
             return [];
-        return this.#scanText(pageIndex, (scan) => rectsOf(this.#pdfium, scan, range));
+        const paragraph = range.start.elementId === range.end.elementId
+            ? this.#paragraphOf(range.start.elementId)
+            : undefined;
+        return this.#scanText(pageIndex, (scan) => rectsOf(this.#pdfium, paragraph?.paragraph.id === range.start.elementId
+            ? this.#paragraphScan(scan, paragraph)
+            : scan, range));
     }
     dispose() {
         this.#fonts.release(this.#pdfium, this.#document.handle);
@@ -561,10 +617,14 @@ export class PdfEditDocument {
             // part of the base, so a replay meets them and picks the same id.
             newId: (pageIndex, suffix = "") => `${this.#unusedId(pageIndex, `${this.#pages[pageIndex].key}:n${stateId}.${operationIndex}.${created++}`)}${suffix}`,
             withPage: (pageIndex, use) => this.#writePage(pageIndex, use),
+            readPage: (pageIndex, use) => this.#withPage(pageIndex, use),
+            paragraph: (id) => this.#paragraphOf(id),
+            pageElements: (pageIndex) => this.#elementsOf(pageIndex),
             appendObjects: (pageIndex, records) => {
                 const page = this.#pages[pageIndex];
                 page.objects = [...(page.objects ?? []), ...records];
                 delete page.elements;
+                delete page.paragraphs;
             },
             locate: (id) => this.#locate(id),
             element: (id) => this.getElement(id),
@@ -591,6 +651,7 @@ export class PdfEditDocument {
                     return;
                 delete page.geometry;
                 delete page.elements;
+                delete page.paragraphs;
             },
             spliceObjects: (pageIndex, start, count, records) => {
                 const page = this.#pages[pageIndex];
@@ -598,13 +659,16 @@ export class PdfEditDocument {
                 objects.splice(start, count, ...records);
                 page.objects = objects;
                 delete page.elements;
+                delete page.paragraphs;
             },
         };
     }
     /** Cached elements carry page indexes, which a structure change makes stale. */
     #forgetElements() {
-        for (const page of this.#pages)
+        for (const page of this.#pages) {
             delete page.elements;
+            delete page.paragraphs;
+        }
     }
     #locate(id) {
         const pageIndex = this.#pageIndexOf(id);
@@ -617,7 +681,17 @@ export class PdfEditDocument {
                 indexes.push(index);
         });
         const record = records[indexes[0] ?? -1];
-        return record ? { pageIndex, indexes, record } : undefined;
+        if (record)
+            return { pageIndex, indexes, record };
+        const paragraph = this.#paragraphOf(id);
+        const first = paragraph && records[paragraph.indexes[0] ?? -1];
+        return paragraph?.paragraph.id === id && first
+            ? {
+                pageIndex,
+                indexes: paragraph.indexes,
+                record: { ...first, id, mark: paragraph.spec },
+            }
+            : undefined;
     }
     /** Loads a page, lets `use` change it, regenerates its content stream. */
     #writePage(pageIndex, use) {
@@ -636,6 +710,7 @@ export class PdfEditDocument {
                     details: { stage: "apply", pageIndex },
                 });
             delete this.#pages[pageIndex].elements;
+            delete this.#pages[pageIndex].paragraphs;
             return result;
         }
         finally {
@@ -657,7 +732,13 @@ export class PdfEditDocument {
             return record.elements;
         return this.#withPage(pageIndex, (page, textPage, geometry) => {
             const { elements } = scanPage(this.#pdfium, page, textPage, pageIndex, geometry, this.#objectsOf(pageIndex, page));
-            record.elements = Object.freeze(elements);
+            record.paragraphs = discoverParagraphs(this.#pdfium, page, pageIndex, geometry, this.#objectsOf(pageIndex, page), elements, textPage, this.#measurer);
+            record.elements = Object.freeze(elements.map((element) => {
+                const paragraph = record.paragraphs?.get(element.id)?.paragraph;
+                return paragraph && paragraph.id !== element.id
+                    ? { ...element, textEditingTarget: paragraph.id }
+                    : element;
+            }));
             return record.elements;
         });
     }
@@ -779,6 +860,28 @@ export class PdfEditDocument {
         };
         return record.geometry;
     }
+    #paragraphScan(scan, target) {
+        const paragraph = target.paragraph;
+        if (paragraph.memberIds.includes(paragraph.id))
+            return scan;
+        const members = new Map(paragraph.members.map((member) => [member.elementId, member]));
+        return {
+            ...scan,
+            elements: [
+                ...scan.elements.filter((element) => !members.has(element.id)),
+                paragraphElement(paragraph),
+            ],
+            offsets: scan.offsets.map((position) => {
+                const member = position && members.get(position.elementId);
+                return member && position
+                    ? {
+                        elementId: paragraph.id,
+                        offset: Math.min(member.end, member.start + position.offset),
+                    }
+                    : position;
+            }),
+        };
+    }
     /** Loads a page with its text page and the character-to-element mapping. */
     #scanText(pageIndex, use) {
         return this.#withPage(pageIndex, (page, textPage, geometry) => {

package/dist/edit/pdf/engine/elements.d.ts

@@ -2,6 +2,7 @@ import type { PageRect } from "../../types.js";
 import type { PdfElement, PdfTextStyle } from "../types.js";
 import { type PageGeometry } from "./geometry.js";
 import type { Pdfium } from "./pdfium.js";
+import type { ParagraphSpec } from "./paragraph.js";
 /** FPDFPageObj_GetType values. */
 export declare const OBJECT_TEXT = 1;
 export declare const OBJECT_PATH = 2;
@@ -11,10 +12,13 @@ export declare const MARK_NAME = "WebDoc";
 export declare const MARK_PARAM = "webdoc";
 /** What a `WebDoc` mark says about the objects it tags. */
 export interface MarkParams {
-    readonly kind: "textBox" | "table";
+    readonly kind: "textBox" | "table" | "paragraph";
     readonly id: string;
     readonly [key: string]: unknown;
 }
+export declare function paragraphSpecOf(mark: unknown): ParagraphSpec | undefined;
+/** Paragraph marks store the upright frame; the page may subsequently rotate. */
+export declare function paragraphBounds(spec: ParagraphSpec, geometry: PageGeometry): PageRect;
 /** One PDFium page object with what the model knows about it. */
 export interface ObjectRecord {
     /** Element id of a plain object; objects of a marked group share their group id. */