@devries-cameron20 / t3code Mirror · read-only / commits / 787c59c802
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 23:30 UTC

fix(web): Word previews no longer cut ordinary documents short (#32040)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Julius Marminge <julius0216@outlook.com>

yaacovcorcos committed Nov 11, 2026 at 21:44 UTC 787c59c802ff58274b06a2cdfb1a369bf7c04571
3 files changed +224 -40
apps/web/src/components/files/AttachmentFilePreview.browser.test.tsx
+23 -17
@@ -482,6 +482,7 @@ describe("bounded binary previews", () => {
482 });
483 });
484 import { OFFICE_PREVIEW_MAX_COLUMNS, type OfficeWorkerResult } from "./officePreviewProtocol";
485 +import { WORD_LIMITS } from "./officeArchive";
486
487 describe("large text preview recovery", () => {
488 it("retains the workspace viewer across URL renewals and reloads revisions and manual refreshes", async () => {
@@ -1190,7 +1191,7 @@ it.each([
1191 : scenario === "runs"
1192 ? `<w:p>${"<w:r><w:t>x</w:t></w:r>".repeat(50_001)}</w:p>`
1193 : scenario === "tables"
1193 - ? "<w:tbl/>".repeat(50_001)
1194 + ? "<w:tbl/>".repeat(2 * WORD_LIMITS.bodyElements)
1195 : scenario === "XML bytes"
1196 ? `<w:p><w:r><w:t>${"x".repeat(2 * 1024 * 1024)}</w:t></w:r></w:p>`
1197 : "<w:p/>";
@@ -1203,7 +1204,7 @@ it.each([
1204 if (scenario === "auxiliary XML" || scenario === "renamed XML")
1205 archive.file(
1206 scenario === "renamed XML" ? "word/footnotes.dat" : "word/footnotes.xml",
1206 - `<w:footnotes>${"<w:p/>".repeat(50_001)}</w:footnotes>`,
1207 + `<w:footnotes>${"<w:p/>".repeat(WORD_LIMITS.auxiliaryMarkup + 1)}</w:footnotes>`,
1208 );
1209 const file = await archive.generateAsync({ type: "blob", compression: "DEFLATE" });
1210 const src = URL.createObjectURL(file);
@@ -1254,8 +1255,8 @@ it.each([
1255 );
1256 }
1257 expect(markup.length).toBeLessThan(body.length);
1257 - expect(markup.length).toBeLessThanOrEqual(3 * 1024 * 1024);
1258 - expect(markup.match(/<(?![!?/])/g)!.length).toBeLessThanOrEqual(100_000);
1258 + expect(markup.length).toBeLessThanOrEqual(WORD_LIMITS.bodyBytes);
1259 + expect(markup.match(/<(?![!?/])/g)!.length).toBeLessThanOrEqual(WORD_LIMITS.bodyTags);
1260 }
1261 } finally {
1262 worker.terminate();
@@ -1286,15 +1287,15 @@ it.each([
1287 );
1288 archive.file(
1289 "word/_rels/document.xml.rels",
1289 - `<Relationships>${scenario === "content type" ? "" : `<Relationship Id="style" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/styles" Target="${path.slice(5)}"/>`}${scenario === "metadata type" ? "<unused/>".repeat(50_001) : ""}</Relationships>`,
1290 + `<Relationships>${scenario === "content type" ? "" : `<Relationship Id="style" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/styles" Target="${path.slice(5)}"/>`}${scenario === "metadata type" ? "<unused/>".repeat(WORD_LIMITS.auxiliaryMarkup + 1) : ""}</Relationships>`,
1291 );
1292 archive.file(
1293 path,
1294 scenario === "relationship bytes"
1294 - ? `<styles>${"x".repeat(1024 * 1024)}</styles>`
1295 + ? `<styles>${"x".repeat(WORD_LIMITS.auxiliaryPartBytes)}</styles>`
1296 : scenario === "deep XML"
1297 ? `${"<styles>".repeat(65)}${"</styles>".repeat(65)}`
1297 - : `<styles>${"<style/>".repeat(50_001)}</styles>`,
1298 + : `<styles>${"<style/>".repeat(WORD_LIMITS.auxiliaryMarkup + 1)}</styles>`,
1299 );
1300 const bytes = await archive.generateAsync({ type: "arraybuffer", compression: "DEFLATE" });
1301 const worker = new Worker(new URL("./officePreview.worker.ts", import.meta.url), {
@@ -1348,11 +1349,12 @@ async function expandedWordFixture(role: (typeof EXPANDED_WORD_ROLES)[number]) {
1349 '<Relationships><Relationship Id="doc" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/officeDocument" Target="word/document.xml"/></Relationships>',
1350 );
1351 const text = (value: string) => `<w:p><w:r><w:t>${value}</w:t></w:r></w:p>`;
1351 - // Each copy of the 2,000-paragraph payload costs 8,000 markup units, so 20
1352 - // references (or pages) exceed the 100,000-unit expansion budget.
1352 + // Each copy of the 2,000-paragraph payload costs 6,000 markup units, so enough
1353 + // references (or pages) exceed the expansion budget.
1354 + const copies = Math.ceil(WORD_LIMITS.expandedMarkup / 6_000) + 1;
1355 archive.file(
1356 "word/document.xml",
1355 - `<w:document xmlns:w="${ns}" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"><w:body>${text("body retained")}${Array.from({ length: 20 }, () => `<w:p><w:r><w:footnoteReference w:id="1"/><w:endnoteReference w:id="1"/><w:commentReference w:id="1"/><w:br w:type="page"/></w:r></w:p>`).join("")}<w:sectPr><w:headerReference w:type="default" r:id="aux"/><w:footerReference w:type="default" r:id="aux"/></w:sectPr></w:body></w:document>`,
1357 + `<w:document xmlns:w="${ns}" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"><w:body>${text("body retained")}${Array.from({ length: copies }, () => `<w:p><w:r><w:footnoteReference w:id="1"/><w:endnoteReference w:id="1"/><w:commentReference w:id="1"/><w:br w:type="page"/></w:r></w:p>`).join("")}<w:sectPr><w:headerReference w:type="default" r:id="aux"/><w:footerReference w:type="default" r:id="aux"/></w:sectPr></w:body></w:document>`,
1358 );
1359 archive.file(
1360 "word/_rels/document.xml.rels",
@@ -1360,23 +1362,25 @@ async function expandedWordFixture(role: (typeof EXPANDED_WORD_ROLES)[number]) {
1362 );
1363 const xml =
1364 role === "style chain" || role === "style count" || role === "style visits"
1363 - ? `<w:styles xmlns:w="${ns}">${Array.from({ length: role === "style count" ? 3_000 : 1_000 }, (_, i) => `<w:style w:styleId="s${i}" w:type="paragraph">${i ? `<w:basedOn w:val="s${role === "style visits" && i >= 30 ? 29 : i - 1}"/>` : ""}<w:pPr><w:spacing w:after="120"/></w:pPr></w:style>`).join("")}</w:styles>`
1365 + ? `<w:styles xmlns:w="${ns}">${Array.from({ length: role === "style count" ? WORD_LIMITS.styles + 1 : role === "style visits" ? Math.ceil(WORD_LIMITS.styleVisits / 30) + 1 : 1_000 }, (_, i) => `<w:style w:styleId="s${i}" w:type="paragraph">${i ? `<w:basedOn w:val="s${role === "style visits" && i >= 30 ? 29 : i - 1}"/>` : ""}<w:pPr><w:spacing w:after="120"/></w:pPr></w:style>`).join("")}</w:styles>`
1366 : role === "styles"
1367 ? `<w:styles xmlns:w="${ns}">${Array.from({ length: 100 }, (_, i) => `<w:style w:styleId="${"s".repeat(2_000)}${i}" w:type="paragraph"><w:basedOn w:val="s0"/><w:pPr><w:spacing w:after="120"/></w:pPr></w:style>`).join("")}</w:styles>`
1368 : role.includes("numbering")
1367 - ? `<w:numbering xmlns:w="${ns}">${role === "picture numbering" ? '<w:numPicBullet w:numPicBulletId="1"/>' : ""}<w:abstractNum w:abstractNumId="1">${Array.from({ length: 9 }, (_, i) => `<w:lvl w:ilvl="${i}"><w:lvlText w:val="%1."/></w:lvl>`).join("")}</w:abstractNum>${Array.from({ length: 100 }, (_, i) => `<w:num w:numId="${i}"><w:abstractNumId w:val="1"/></w:num>`).join("")}</w:numbering>`
1369 + ? `<w:numbering xmlns:w="${ns}">${role === "picture numbering" ? '<w:numPicBullet w:numPicBulletId="1"/>' : ""}<w:abstractNum w:abstractNumId="1">${Array.from({ length: 9 }, (_, i) => `<w:lvl w:ilvl="${i}"><w:lvlText w:val="%1."/></w:lvl>`).join("")}</w:abstractNum>${Array.from({ length: WORD_LIMITS.numberingReferences + 1 }, (_, i) => `<w:num w:numId="${i}"><w:abstractNumId w:val="1"/></w:num>`).join("")}</w:numbering>`
1370 : role === "header" || role === "footer"
1371 ? `<w:${role === "header" ? "hdr" : "ftr"} xmlns:w="${ns}">${text("omitted payload").repeat(2_000)}</w:${role === "header" ? "hdr" : "ftr"}>`
1372 : `<w:${role} xmlns:w="${ns}"><w:${role.slice(0, -1)} w:id="1">${text("omitted payload").repeat(2_000)}</w:${role.slice(0, -1)}></w:${role}>`;
1373 archive.file("word/media/aux.xml", xml);
1374 if (role === "note roots") {
1375 + // Each note's XML repeats the 32 KB root, which the expansion budget charges twice.
1376 + const notes = Math.ceil(WORD_LIMITS.expandedBytes / 64_000) + 1;
1377 archive.file(
1378 "word/document.xml",
1375 - `<w:document xmlns:w="${ns}"><w:body>${text("body retained")}${Array.from({ length: 200 }, (_, id) => `<w:p><w:r><w:footnoteReference w:id="${id}"/></w:r></w:p>`).join("")}</w:body></w:document>`,
1379 + `<w:document xmlns:w="${ns}"><w:body>${text("body retained")}${Array.from({ length: notes }, (_, id) => `<w:p><w:r><w:footnoteReference w:id="${id}"/></w:r></w:p>`).join("")}</w:body></w:document>`,
1380 );
1381 archive.file(
1382 "word/media/aux.xml",
1379 - `<w:footnotes xmlns:w="${ns}" padding="${"x".repeat(32_000)}">${Array.from({ length: 200 }, (_, id) => `<w:footnote w:id="${id}">${text("omitted payload")}</w:footnote>`).join("")}</w:footnotes>`,
1383 + `<w:footnotes xmlns:w="${ns}" padding="${"x".repeat(32_000)}">${Array.from({ length: notes }, (_, id) => `<w:footnote w:id="${id}">${text("omitted payload")}</w:footnote>`).join("")}</w:footnotes>`,
1384 );
1385 }
1386 return archive.generateAsync({ type: "blob", compression: "DEFLATE" });
@@ -1401,7 +1405,7 @@ it.each(EXPANDED_WORD_ROLES)(
1405 (sum, [chunk]) => sum + (typeof chunk === "string" ? chunk.length : 0),
1406 0,
1407 ),
1404 - ).toBeLessThan(4 * 1024 * 1024);
1408 + ).toBeLessThan(WORD_LIMITS.expandedBytes + 4 * 1024 * 1024);
1409 const bounded = await JSZip.loadAsync(prepared.bytes);
1410 const markup = await bounded.file("word/document.xml")!.async("string");
1411 expect(markup).toContain("body retained");
@@ -1622,7 +1626,8 @@ it.each([false, true])(
1626 },
1627 );
1628
1625 -it.each([1, 145])(
1629 +// A body just over the read limit once each paragraph repeats its text this often.
1630 +it.each([1, Math.ceil(WORD_LIMITS.bodyBytes / 5_000 / 4)])(
1631 "counts shared Word settings once with %s text repetitions",
1632 async (repetitions) => {
1633 const { default: JSZip } = await import("jszip");
@@ -1658,7 +1663,8 @@ it.each([1, 145])(
1663 const document = await bounded.file("word/document.xml")!.async("string");
1664 const retainedSettings = await bounded.file("word/settings.xml")!.async("string");
1665 expect(document.match(/<w:p>/g)!.length).toBeGreaterThan(100);
1661 - expect(document.length + retainedSettings.length).toBeLessThanOrEqual(3 * 1024 * 1024);
1666 + expect(document.length).toBeLessThanOrEqual(WORD_LIMITS.bodyBytes);
1667 + expect(retainedSettings).toBe(settings);
1668 } finally {
1669 worker.terminate();
1670 }
apps/web/src/components/files/officeArchive.test.ts new
+136
@@ -0,0 +1,136 @@
1 +import JSZip from "jszip";
2 +import { describe, expect, it } from "vite-plus/test";
3 +
4 +import { prepareOffice } from "./officeArchive";
5 +
6 +const W = 'xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"';
7 +const V = 'xmlns:v="urn:schemas-microsoft-com:vml"';
8 +const R = 'xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships"';
9 +const relationship = (type: string, target: string) =>
10 + `<Relationship Id="${type}" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/${type}" Target="${target}"/>`;
11 +const paragraph = (text: string, properties = "") =>
12 + `<w:p>${properties}<w:r><w:t xml:space="preserve">${text}</w:t></w:r></w:p>`;
13 +
14 +async function wordFile(
15 + body: string,
16 + parts: Partial<Record<"styles" | "numbering" | "header" | "footer" | "footnotes", string>> = {},
17 + section = "",
18 +) {
19 + const zip = new JSZip();
20 + zip.file(
21 + "[Content_Types].xml",
22 + `<Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types"><Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/><Default Extension="xml" ContentType="application/xml"/><Override PartName="/word/document.xml" ContentType="application/vnd.openxmlformats-officedocument.wordprocessingml.document.main+xml"/></Types>`,
23 + );
24 + zip.file(
25 + "_rels/.rels",
26 + `<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">${relationship("officeDocument", "word/document.xml")}</Relationships>`,
27 + );
28 + zip.file(
29 + "word/_rels/document.xml.rels",
30 + `<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">${Object.keys(
31 + parts,
32 + )
33 + .map((part) => relationship(part, `${part}.xml`))
34 + .join("")}</Relationships>`,
35 + );
36 + for (const [part, xml] of Object.entries(parts)) zip.file(`word/${part}.xml`, xml);
37 + zip.file(
38 + "word/document.xml",
39 + `<w:document ${W} ${V} ${R}><w:body>${body}<w:sectPr>${section}<w:pgSz w:w="12240" w:h="15840"/></w:sectPr></w:body></w:document>`,
40 + );
41 + return zip.generateAsync({ type: "arraybuffer" });
42 +}
43 +
44 +async function prepared(bytes: ArrayBuffer) {
45 + const result = await prepareOffice(bytes, "docx");
46 + const zip = await JSZip.loadAsync(result.bytes);
47 + return { truncated: result.truncated, part: (path: string) => zip.file(path)?.async("string") };
48 +}
49 +
50 +// regression: the budgets sized for a main-thread renderer cut ordinary documents short.
51 +describe("Word preview limits", () => {
52 + it("keeps a document whose table follows a long one", async () => {
53 + const rows = Array.from(
54 + { length: 500 },
55 + (_, row) =>
56 + `<w:tr>${Array.from({ length: 5 }, (_, cell) => `<w:tc>${paragraph(`r${row}c${cell}`)}</w:tc>`).join("")}</w:tr>`,
57 + ).join("");
58 + const result = await prepared(
59 + await wordFile(`<w:tbl>${rows}</w:tbl>${paragraph("After the table")}`),
60 + );
61 + expect(result.truncated).toBe(false);
62 + expect(await result.part("word/document.xml")).toContain("After the table");
63 + });
64 +
65 + it("keeps the list and style definitions of a long document", async () => {
66 + const levels = Array.from(
67 + { length: 9 },
68 + (_, level) =>
69 + `<w:lvl w:ilvl="${level}"><w:numFmt w:val="decimal"/><w:lvlText w:val="%${level + 1}."/></w:lvl>`,
70 + ).join("");
71 + const numbering = `<w:numbering ${W}>${Array.from({ length: 60 }, (_, list) => `<w:abstractNum w:abstractNumId="${list}">${levels}</w:abstractNum>`).join("")}${Array.from({ length: 60 }, (_, list) => `<w:num w:numId="${list + 1}"><w:abstractNumId w:val="${list}"/></w:num>`).join("")}</w:numbering>`;
72 + const styles = `<w:styles ${W}>${Array.from(
73 + { length: 1_500 },
74 + (_, style) =>
75 + `<w:style w:type="paragraph" w:styleId="S${style}">${style % 10 ? `<w:basedOn w:val="S${style - 1}"/>` : ""}<w:pPr><w:spacing w:after="${style}"/></w:pPr><w:rPr><w:sz w:val="22"/></w:rPr></w:style>`,
76 + ).join("")}</w:styles>`;
77 + const result = await prepared(
78 + await wordFile(
79 + Array.from({ length: 6_000 }, (_, index) =>
80 + paragraph(
81 + `Item ${index}`,
82 + `<w:pPr><w:pStyle w:val="S${index % 1_500}"/><w:numPr><w:ilvl w:val="0"/><w:numId w:val="${(index % 60) + 1}"/></w:numPr></w:pPr>`,
83 + ),
84 + ).join(""),
85 + { styles, numbering },
86 + ),
87 + );
88 + expect(result.truncated).toBe(false);
89 + expect(await result.part("word/numbering.xml")).toBe(numbering);
90 + expect(await result.part("word/styles.xml")).toBe(styles);
91 + expect(await result.part("word/document.xml")).toContain("Item 5999");
92 + });
93 +
94 + const text = "x".repeat(140_000);
95 + it.each([
96 + ["on its own", `<w:p><w:r><w:t>${text}</w:t></w:r></w:p>`],
97 + [
98 + "after an empty paragraph in a text box",
99 + `<w:p><w:r><w:pict><v:shape><v:textbox><w:txbxContent><w:p/></w:txbxContent></v:textbox></v:shape></w:pict></w:r><w:r><w:t>${text}</w:t></w:r></w:p>`,
100 + ],
101 + ["written as CDATA", `<w:p><w:r><w:t><![CDATA[${text}]]></w:t></w:r></w:p>`],
102 + ])("stops before a paragraph too long to lay out %s", async (_, long) => {
103 + const result = await prepared(
104 + await wordFile(paragraph("Kept") + long + paragraph("Not reached")),
105 + );
106 + expect(result.truncated).toBe(true);
107 + const document = await result.part("word/document.xml");
108 + expect(document).toContain("Kept");
109 + expect(document).not.toContain("Not reached");
110 + });
111 +
112 + it("counts only the text a paragraph lays out", async () => {
113 + const formatted = `<w:p>${" ".repeat(140_000)}<w:r><w:instrText>${"x".repeat(140_000)}</w:instrText></w:r><w:r><w:t>${"x".repeat(40_000)}</w:t></w:r></w:p>`;
114 + const result = await prepared(await wordFile(formatted + paragraph("After")));
115 + expect(result.truncated).toBe(false);
116 + expect(await result.part("word/document.xml")).toContain("After");
117 + });
118 +
119 + it("keeps a long header", async () => {
120 + const header = `<w:hdr ${W}>${Array.from({ length: 3_000 }, (_, line) => paragraph(`Header line ${line}`)).join("")}</w:hdr>`;
121 + const footer = `<w:ftr ${W}>${paragraph("Footer")}</w:ftr>`;
122 + const footnotes = `<w:footnotes ${W}><w:footnote w:id="1">${paragraph("Note")}</w:footnote></w:footnotes>`;
123 + const result = await prepared(
124 + await wordFile(
125 + `<w:p><w:r><w:t>Body</w:t></w:r><w:r><w:footnoteReference w:id="1"/></w:r></w:p>`,
126 + { header, footer, footnotes },
127 + `<w:headerReference w:type="default" r:id="header"/><w:footerReference w:type="default" r:id="footer"/>`,
128 + ),
129 + );
130 + expect(result.truncated).toBe(false);
131 + expect(await result.part("word/header.xml")).toBe(header);
132 + expect(await result.part("word/footer.xml")).toBe(footer);
133 + expect(await result.part("word/footnotes.xml")).toBe(footnotes);
134 + expect(await result.part("word/document.xml")).toContain("Body");
135 + });
136 +});
apps/web/src/components/files/officeArchive.ts
+65 -23
@@ -10,6 +10,31 @@ import {
10 type OfficeWorkerRequest,
11 } from "./officePreviewProtocol";
12
13 +// Word limits, measured against the @betteroffice/docx engine, which lays out in its own worker.
14 +export const WORD_LIMITS = {
15 + /** document.xml read and kept, about 700 pages of prose; the engine opens it in about 1 s. */
16 + bodyBytes: 4 * 1024 * 1024,
17 + bodyElements: 50_000,
18 + bodyTags: 1_000_000,
19 + /** A table is laid out in one go: 100,000 tags is a 4,000-row table, about 2 s. */
20 + elementTags: 100_000,
21 + /** Line breaking grows with the square of a paragraph's text: 40K characters take about 2 s. */
22 + paragraphText: 128 * 1024,
23 + styles: 8_192,
24 + styleBytes: 16 * 1024 * 1024,
25 + styleUnits: 400_000,
26 + styleVisits: 131_072,
27 + stylePropertyVisits: 2_000_000,
28 + numberingReferences: 20_000,
29 + numberingBytes: 32 * 1024 * 1024,
30 + numberingMarkup: 400_000,
31 + auxiliaryPartBytes: 8 * 1024 * 1024,
32 + auxiliaryMarkup: 1_000_000,
33 + /** Body, styles, lists and notes, with headers and footers counted once per page. */
34 + expandedBytes: 64 * 1024 * 1024,
35 + expandedMarkup: 4_000_000,
36 +};
37 +
38 function xmlParser() {
39 const parser = new SaxesParser({ xmlns: true });
40 parser.on("doctype", () => {
@@ -143,27 +168,41 @@ function wordDocuments(
168 let elementTags = 0;
169 let batchTags = 0;
170 let elements = 0;
171 + // The text each open paragraph lays out (its w:t runs); a nested paragraph counts apart.
172 + const paragraphText: number[] = [];
173 + let inText = false;
174 let truncated = incomplete;
175 const limit = Symbol("Word preview limit");
176 + const countText = (text: string) => {
177 + if (!inText || paragraphText.length === 0) return;
178 + const length = paragraphText.at(-1)! + text.length;
179 + if (length > WORD_LIMITS.paragraphText) throw limit;
180 + paragraphText[paragraphText.length - 1] = length;
181 + };
182 + parser.on("text", countText);
183 + parser.on("cdata", countText);
184 parser.on("opentag", (tag) => {
185 names.push(tag.name);
186 + if (prefix && tag.local === "p") paragraphText.push(0);
187 + if (tag.local === "t") inText = true;
188 if (names.length === 2 && tag.local === "body") {
189 prefix = xml.slice(0, parser.position).replace(/\/>$/, ">");
190 suffix = `</${tag.name}></${names[0]}>`;
191 boundary = batchStart = parser.position;
192 } else if (prefix && names.length > 2) {
155 - if (names.length === 3 && ++elements > 5_000) throw limit;
193 + if (names.length === 3 && ++elements > WORD_LIMITS.bodyElements) throw limit;
194 tags++;
195 elementTags++;
196 batchTags++;
159 - if (tags > 100_000 || elementTags > 2_000 || parser.position - boundary > 128 * 1024)
160 - throw limit;
197 + if (tags > WORD_LIMITS.bodyTags || elementTags > WORD_LIMITS.elementTags) throw limit;
198 }
199 if (names.length > 64) throw limit;
200 });
201 parser.on("closetag", (tag) => {
202 + if (prefix && tag.local === "p") paragraphText.pop();
203 + if (tag.local === "t") inText = false;
204 if (prefix && names.length === 3 && tag.local !== "sectPr") {
166 - if (parser.position - prefix.length > 3 * 1024 * 1024) throw limit;
205 + if (parser.position - prefix.length > WORD_LIMITS.bodyBytes) throw limit;
206 boundary = parser.position;
207 elementTags = 0;
208 if (batchTags >= 500) {
@@ -177,7 +216,6 @@ function wordDocuments(
216 try {
217 for (let offset = 0; offset < xml.length; offset += 16 * 1024) {
218 parser.write(xml.slice(offset, offset + 16 * 1024));
180 - if (prefix && parser.position - boundary > 128 * 1024) throw limit;
219 }
220 if (!incomplete) parser.close();
221 } catch (cause) {
@@ -195,7 +233,7 @@ function wordDocuments(
233 const document = prefix + batch.content + (section?.xml ?? "") + suffix;
234 length += batch.content.length;
235 markup += batch.content.match(/<(?![!?/])/g)?.length ?? 0;
198 - if (length > 3 * 1024 * 1024 || markup > 100_000) {
236 + if (length > WORD_LIMITS.bodyBytes || markup > WORD_LIMITS.bodyTags) {
237 truncated = true;
238 break;
239 }
@@ -617,9 +655,6 @@ function wordPartTypes() {
655
656 // Bound inherited style and numbering expansion before the engine loads it.
657 // Picture bullets can duplicate embedded images, so omit those definitions.
620 -const WORD_STYLE_MAX_BYTES = 1024 * 1024;
621 -const WORD_STYLE_MAX_UNITS = 20_000;
622 -
658 function wordXmlCost(xml: string, roles: Set<string> | undefined) {
659 const parser = xmlParser();
660 let depth = 0;
@@ -648,7 +683,7 @@ function wordXmlCost(xml: string, roles: Set<string> | undefined) {
683 Object.values(tag.attributes).find((a) => a.local === name)?.value ?? "";
684 names.push(tag.local === "tblStylePr" ? `${tag.local}:${attr("type")}` : tag.local);
685 if (roles?.has("styles") && (tag.local === "style" || tag.local === "docDefaults")) {
651 - if (++styleCount > 1_024 || attr("styleId").length > 256) throw limit;
686 + if (++styleCount > WORD_LIMITS.styles || attr("styleId").length > 256) throw limit;
687 style = { id: attr("styleId"), properties: new Map(), base: "", linked: "" };
688 }
689 if (style) {
@@ -659,7 +694,7 @@ function wordXmlCost(xml: string, roles: Set<string> | undefined) {
694 const size = new TextEncoder().encode(
695 xml.slice(xml.lastIndexOf("<", parser.position - 1), parser.position),
696 ).length;
662 - if (++ownUnits > WORD_STYLE_MAX_UNITS) throw limit;
697 + if (++ownUnits > WORD_LIMITS.styleUnits) throw limit;
698 style.properties.set(key, size);
699 }
700 }
@@ -671,7 +706,7 @@ function wordXmlCost(xml: string, roles: Set<string> | undefined) {
706 markup: markup - 1,
707 };
708 if (tag.local === "num") {
674 - if (references.length >= 2_000) throw limit;
709 + if (references.length >= WORD_LIMITS.numberingReferences) throw limit;
710 reference = { id: "", selector: attr("numId").length + 32 };
711 references.push(reference);
712 }
@@ -731,16 +766,16 @@ function wordXmlCost(xml: string, roles: Set<string> | undefined) {
766 ]) {
767 const seen = new Set<string>();
768 for (let node = root; node; node = node.parent) {
734 - if (++visited > 8_192) return exceeded;
769 + if (++visited > WORD_LIMITS.styleVisits) return exceeded;
770 for (const [key, size] of node.properties) {
736 - if (++propertyVisits > 100_000) return exceeded;
771 + if (++propertyVisits > WORD_LIMITS.stylePropertyVisits) return exceeded;
772 if (seen.has(key)) continue;
773 // Latent styles never enter properties. Charge each own property
774 // once per resolved style, including linked and default selectors.
775 const copies = id ? 1 : 2;
776 bytes += (size * 2 + id.length + 32) * copies;
777 expandedMarkup += copies;
743 - if (bytes > WORD_STYLE_MAX_BYTES || expandedMarkup > WORD_STYLE_MAX_UNITS)
778 + if (bytes > WORD_LIMITS.styleBytes || expandedMarkup > WORD_LIMITS.styleUnits)
779 return exceeded;
780 seen.add(key);
781 }
@@ -756,7 +791,8 @@ function wordXmlCost(xml: string, roles: Set<string> | undefined) {
791 if (!cost) continue;
792 bytes += cost.bytes * 2 + cost.markup * ref.selector;
793 markup += cost.markup;
759 - if (bytes > 128 * 1024 || markup > 2_000) return exceeded;
794 + if (bytes > WORD_LIMITS.numberingBytes || markup > WORD_LIMITS.numberingMarkup)
795 + return exceeded;
796 }
797 }
798 return { bytes, markup, pictures };
@@ -851,7 +887,7 @@ async function wordExpansion(
887 let expanded = memo.get(key);
888 if (!expanded) {
889 expandedXmlBytes += node.xml.length * 2;
854 - if (expandedXmlBytes > 3 * 1024 * 1024) {
890 + if (expandedXmlBytes > WORD_LIMITS.expandedBytes) {
891 cost.bytes = Infinity;
892 break;
893 }
@@ -901,7 +937,9 @@ async function wordExpansion(
937 markup += cost.markup;
938 }
939 const overBudget =
904 - bytes > 3 * 1024 * 1024 || markup > 100_000 || media > OFFICE_PREVIEW_MAX_BYTES;
940 + bytes > WORD_LIMITS.expandedBytes ||
941 + markup > WORD_LIMITS.expandedMarkup ||
942 + media > OFFICE_PREVIEW_MAX_BYTES;
943 const truncated = (used.size > 0 && overBudget) || bytes === Infinity;
944 const omitted = new Set([...optional].filter((path) => truncated || !used.has(path)));
945 for (const path of omitted) archive.remove(path);
@@ -1014,7 +1052,10 @@ export async function boundedArchive(input: Extract<OfficeWorkerRequest, { type:
1052 // stay small before the engine parses it.
1053 if (input.kind === "docx" && !document && !media) {
1054 for (const byte of chunk) if (byte === 0x3c) wordAuxiliaryMarkup++;
1017 - if (length > 1024 * 1024 || wordAuxiliaryMarkup > 50_000) {
1055 + if (
1056 + length > WORD_LIMITS.auxiliaryPartBytes ||
1057 + wordAuxiliaryMarkup > WORD_LIMITS.auxiliaryMarkup
1058 + ) {
1059 stream.pause();
1060 reject(
1061 new Error("Preview limited: this Word document has too much auxiliary markup."),
@@ -1058,8 +1099,9 @@ export async function boundedArchive(input: Extract<OfficeWorkerRequest, { type:
1099 if (
1100 (roles?.has("numbering") && cost.pictures) ||
1101 (roles?.has("styles") &&
1061 - (cost.bytes > WORD_STYLE_MAX_BYTES || cost.markup > WORD_STYLE_MAX_UNITS)) ||
1062 - (roles?.has("numbering") && (cost.bytes > 128 * 1024 || cost.markup > 2_000))
1102 + (cost.bytes > WORD_LIMITS.styleBytes || cost.markup > WORD_LIMITS.styleUnits)) ||
1103 + (roles?.has("numbering") &&
1104 + (cost.bytes > WORD_LIMITS.numberingBytes || cost.markup > WORD_LIMITS.numberingMarkup))
1105 ) {
1106 wordInputTruncated = true;
1107 continue;
@@ -1098,8 +1140,8 @@ export async function boundedArchive(input: Extract<OfficeWorkerRequest, { type:
1140 const images = extra.media;
1141 const batchImages = new Set(extra.images);
1142 if (
1101 - bytes + cost.bytes > 3 * 1024 * 1024 ||
1102 - markup + cost.markup > 100_000 ||
1143 + bytes + cost.bytes > WORD_LIMITS.expandedBytes ||
1144 + markup + cost.markup > WORD_LIMITS.expandedMarkup ||
1145 mediaBytes + images > OFFICE_PREVIEW_MAX_BYTES
1146 )
1147 break;