wsk/deerflow-code
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: unify formal Word export

Browse Source
This commit is contained in:
吉鑫升 committed 2026-08-26 16:02:23 +08:00
1 parent d35d075669
commit 0980eea975
16 files changed
+1356 -609
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

frontend-web/src/core/api/markdown-docx-export.ts
+10 -102
View File
@@ -1,114 +1,22 @@
import { apiFetch } from "@/core/api/fetch-client";
import { getBackendBaseURL } from "@/core/config";
import { OptType, writeYLog } from "@/core/log/y-log";
import { downloadMarkdownAsDocx } from "@/core/scheduled-tasks/markdown-docx";
function filenameFromContentDisposition(header: string | null): string | null {
if (!header) return null;
const star = /filename\*=(?:UTF-8'')?([^;]+)/i.exec(header);
if (star) {
const raw = star[1].replace(/["']/g, "").trim();
try {
return decodeURIComponent(raw);
} catch {
return raw;
}
}
const plain = /filename="?([^";\n]+)"?/i.exec(header);
return plain ? plain[1].trim() : null;
}
function sanitizeFallbackFilename(title: string | undefined): string {
const base = (title?.trim() || "document").replace(/[<>:"/\\|?*\x00-\x1f]+/g, "_");
return base.toLowerCase().endsWith(".docx") ? base : `${base}.docx`;
}
async function downloadBlobResponse(response: Response, fallbackFilename: string) {
if (!response.ok) {
const error = await response
.json()
.catch(() => ({ detail: "导出 Word 失败" }));
const detail = error.detail;
if (typeof detail === "string") throw new Error(detail);
throw new Error("导出 Word 失败");
}
const blob = await response.blob();
const filename =
filenameFromContentDisposition(response.headers.get("Content-Disposition")) ??
fallbackFilename;
const url = URL.createObjectURL(blob);
const anchor = document.createElement("a");
anchor.href = url;
anchor.download = filename;
anchor.rel = "noopener";
document.body.appendChild(anchor);
anchor.click();
anchor.remove();
window.setTimeout(() => URL.revokeObjectURL(url), 30_000);
}
/**
* 通用 Markdown → Word 导出（Gateway `writing` 路由，与 `/api/writing/rewrite` 同级）。
* Unified Markdown → Word export.
*
* POST /api/writing/export/docx
* Body: { title?: string, markdown: string }
*
* 返回 `application/vnd.openxmlformats-officedocument.wordprocessingml.document`，
* 文件名在 `Content-Disposition` 中。
* All download surfaces use the same required Gateway endpoint. The Gateway
* applies the formal OOXML profile and embeds its bundled title font; there is
* deliberately no client-font fallback.
*/
export async function downloadMarkdownDocxFromBackend(options: {
title?: string;
markdown: string;
}): Promise<void> {
const markdown = options.markdown.trim();
if (!markdown) {
throw new Error("文档内容为空");
}
if (!markdown) throw new Error("文档内容为空");
const fallbackFilename = sanitizeFallbackFilename(options.title);
try {
const response = await apiFetch(`${getBackendBaseURL()}/api/writing/export/docx`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
title: options.title?.trim() || undefined,
markdown,
}),
});
// Older/offline gateways mount /api/writing but do not provide export/docx;
// FastAPI then answers 404 or 405. The browser converter is feature-equivalent.
if (response.status === 404 || response.status === 405) {
await downloadMarkdownAsDocx(markdown, options.title?.trim() || "document", options.title);
writeYLog({
optType: OptType.export,
optDetail: "CM助手-全量-导出-Word",
url: `${getBackendBaseURL()}/api/writing/export/docx`,
reqParam: { title: options.title?.trim() || undefined },
});
return;
}
await downloadBlobResponse(response, fallbackFilename);
writeYLog({
optType: OptType.export,
optDetail: "CM助手-全量-导出-Word",
url: `${getBackendBaseURL()}/api/writing/export/docx`,
reqParam: { title: options.title?.trim() || undefined },
});
} catch (err) {
if (err instanceof TypeError) {
await downloadMarkdownAsDocx(markdown, options.title?.trim() || "document", options.title);
writeYLog({
optType: OptType.export,
optDetail: "CM助手-全量-导出-Word",
url: `${getBackendBaseURL()}/api/writing/export/docx`,
reqParam: { title: options.title?.trim() || undefined },
});
return;
}
throw err;
}
await downloadMarkdownAsDocx(
markdown,
options.title?.trim() || "document",
options.title,
);
}
frontend-web/src/core/scheduled-tasks/markdown-docx.ts
+45 -467
View File
@@ -1,485 +1,63 @@
/**
* Convert markdown text to a Word (.docx) Blob in the browser.
* Unified server-side Markdown -> Word export.
*
* Designed for scheduled-task reports that often contain Chinese text,
* markdown tables, and emojis. Compared to Word's stock heading styles we
* deliberately:
* - Force every run color to pure black (`#000000`) so headings, links,
* and inline code never render as Word's default blue / hyperlink theme.
* - Render markdown tables as real OOXML tables (full-width, auto-wrapping)
* instead of falling back to plain text rows that overflow the page.
* - Strip variation selectors, zero-width joiners, and other invisible
* characters that often break Word renderers when emojis leak into the
* report.
*
* Block grammar handled line-by-line:
* - ATX heading: `^(#{1,6})\s+(.*)$`
* - Fenced code: lines between ``` ``` are emitted verbatim in a mono font
* - Blockquote: `^>\s?(.*)$`
* - Ordered list: `^\s*(\d+)\.\s+(.*)$`
* - Unordered list: `^\s*[-*+]\s+(.*)$`
* - Horizontal rule: `^(---|\*\*\*|___)\s*$`
* - Pipe table: contiguous lines starting with `|` plus a separator row
* - Blank line: paragraph break
* - Anything else: paragraph (with inline formatting)
*
* Inline grammar handled:
* `code`, **bold**, __bold__, *italic*, _italic_, [text](url)
* Every Word download surface calls the Gateway so the document receives the
* same formal OOXML styles and the server-bundled 方正小标宋简体 title font.
* The user's browser and desktop Word installation are never used for font
* detection or document generation.
*/
import {
AlignmentType,
BorderStyle,
Document,
HeadingLevel,
Packer,
Paragraph,
Table,
TableCell,
TableRow,
TextRun,
WidthType,
type IRunOptions,
} from "docx";
import { apiFetch } from "@/core/api/fetch-client";
import { getBackendBaseURL } from "@/core/config";
import { OptType, writeYLog } from "@/core/log/y-log";
const BLACK = "000000";
const BODY_FONT = { ascii: "Calibri", eastAsia: "Microsoft YaHei" };
const MONO_FONT = { ascii: "Consolas", eastAsia: "Microsoft YaHei" };
const DOCX_MEDIA_TYPE =
"application/vnd.openxmlformats-officedocument.wordprocessingml.document";
const HEADING_SIZES_HALF_PT: Record<number, number> = {
// Word "size" property is half-points. 32 = 16pt.
1: 36,
2: 32,
3: 28,
4: 26,
5: 24,
6: 22,
};
const BODY_SIZE_HALF_PT = 22; // 11pt
type InlineSegment = {
text: string;
bold?: boolean;
italic?: boolean;
code?: boolean;
};
/**
* Drop characters that Word renders inconsistently or that frequently smuggle
* emoji glitches into corporate reports.
*
* Removes:
* - Variation selectors (FE00–FE0F, E0100–E01EF)
* - Zero-width joiner / non-joiner / space (200C, 200D, FEFF, 200B)
* - Soft hyphen (00AD)
* - Object replacement character (FFFC)
*/
function sanitizeText(text: string): string {
return text
.replace(/[︀-️]/g, "")
.replace(/[​-‍]/g, "")
.replace(/­/g, "")
.replace(//g, "")
.replace(/[\u{E0100}-\u{E01EF}]/gu, "");
function sanitizeFilename(value: string): string {
const name = value
.replace(/\.(md|markdown|docx)$/i, "")
.replace(/[<>:"/\\|?*\x00-\x1f]+/g, "_")
.trim();
return name || "document";
}
/**
* Tokenise a single line of markdown into formatted segments.
*
* Recognised inline syntax:
* `code`, **bold**, __bold__, *italic*, _italic_, [text](url)
* Markdown links are flattened to their text content — we deliberately do not
* emit hyperlink runs because they pick up Word's blue theme color.
*/
function parseInlineSegments(text: string): InlineSegment[] {
const cleaned = sanitizeText(text);
// Strip markdown link syntax first so the regex below doesn't trip on `[...]`.
const withoutLinks = cleaned.replace(/\[([^\]]+)\]\([^)]*\)/g, "$1");
const tokens: InlineSegment[] = [];
const pattern =
/(`+)([^`]+)\1|\*\*([^*]+)\*\*|__([^_]+)__|\*([^*]+)\*|_([^_]+)_/g;
let cursor = 0;
let match: RegExpExecArray | null;
while ((match = pattern.exec(withoutLinks)) !== null) {
if (match.index > cursor) {
tokens.push({ text: withoutLinks.slice(cursor, match.index) });
}
if (match[2] !== undefined) tokens.push({ text: match[2], code: true });
else if (match[3] !== undefined) tokens.push({ text: match[3], bold: true });
else if (match[4] !== undefined) tokens.push({ text: match[4], bold: true });
else if (match[5] !== undefined)
tokens.push({ text: match[5], italic: true });
else if (match[6] !== undefined)
tokens.push({ text: match[6], italic: true });
cursor = pattern.lastIndex;
}
if (cursor < withoutLinks.length) {
tokens.push({ text: withoutLinks.slice(cursor) });
}
return tokens.length > 0 ? tokens : [{ text: withoutLinks }];
async function responseError(response: Response): Promise<Error> {
const payload = await response
.json()
.catch(() => ({ detail: `生成 Word 失败（HTTP ${response.status}）` }));
const detail = payload?.detail;
return new Error(typeof detail === "string" ? detail : "生成 Word 失败");
}
function segmentsToRuns(
segments: InlineSegment[],
extra?: Partial<IRunOptions>,
): TextRun[] {
return segments.map((seg) => {
const opts: IRunOptions = {
text: seg.text,
bold: seg.bold,
italics: seg.italic,
color: BLACK,
font: seg.code ? MONO_FONT : BODY_FONT,
size: extra?.size ?? BODY_SIZE_HALF_PT,
...extra,
};
return new TextRun(opts);
});
}
function makeParagraphFromInline(text: string): Paragraph {
return new Paragraph({
spacing: { after: 120 },
children: segmentsToRuns(parseInlineSegments(text)),
});
}
function makeHeadingParagraph(level: number, text: string): Paragraph {
const clamped = Math.max(1, Math.min(level, 6));
return new Paragraph({
// We attach a heading level so the table-of-contents still works, but the
// run-level color / font / size below override Word's blue theme.
heading:
({
1: HeadingLevel.HEADING_1,
2: HeadingLevel.HEADING_2,
3: HeadingLevel.HEADING_3,
4: HeadingLevel.HEADING_4,
5: HeadingLevel.HEADING_5,
6: HeadingLevel.HEADING_6,
}[clamped] as (typeof HeadingLevel)[keyof typeof HeadingLevel]) ?? undefined,
spacing: { before: 200, after: 120 },
children: segmentsToRuns(parseInlineSegments(text), {
bold: true,
color: BLACK,
size: HEADING_SIZES_HALF_PT[clamped] ?? BODY_SIZE_HALF_PT,
}),
});
}
function makeListItem(
text: string,
ordered: boolean,
level: number,
): Paragraph {
return new Paragraph({
spacing: { after: 80 },
children: segmentsToRuns(parseInlineSegments(text)),
numbering: ordered
? { reference: "ordered-list", level: Math.min(level, 8) }
: undefined,
bullet: ordered ? undefined : { level: Math.min(level, 8) },
});
}
function makeCodeParagraph(line: string): Paragraph {
return new Paragraph({
spacing: { after: 0 },
children: [
new TextRun({
text: sanitizeText(line) || " ",
font: MONO_FONT,
color: BLACK,
size: BODY_SIZE_HALF_PT,
}),
],
});
}
function makeQuoteParagraph(text: string): Paragraph {
return new Paragraph({
indent: { left: 720 },
spacing: { after: 120 },
children: segmentsToRuns(parseInlineSegments(text), { italics: true }),
});
}
function makeHorizontalRule(): Paragraph {
return new Paragraph({
alignment: AlignmentType.CENTER,
spacing: { before: 120, after: 120 },
children: [
new TextRun({
text: "────────",
color: BLACK,
font: BODY_FONT,
size: BODY_SIZE_HALF_PT,
}),
],
});
}
const CELL_BORDER = {
style: BorderStyle.SINGLE,
size: 4,
color: "BFBFBF",
};
function makeTableCell(
text: string,
options: { header?: boolean } = {},
): TableCell {
const segments = parseInlineSegments(text);
return new TableCell({
margins: { top: 80, bottom: 80, left: 100, right: 100 },
children: [
new Paragraph({
spacing: { after: 0 },
children: segmentsToRuns(segments, {
bold: options.header || undefined,
size: BODY_SIZE_HALF_PT,
}),
}),
],
});
}
function splitTableRow(line: string): string[] {
// Strip leading/trailing pipe and any escaped pipes (rare in agent output).
const trimmed = line.replace(/^\s*\|/, "").replace(/\|\s*$/, "");
return trimmed.split("|").map((cell) => cell.trim());
}
function buildTable(headerLine: string, bodyLines: string[]): Table {
const headerCells = splitTableRow(headerLine);
const colCount = headerCells.length;
const rows: TableRow[] = [];
rows.push(
new TableRow({
tableHeader: true,
children: headerCells.map((cell) =>
makeTableCell(cell, { header: true }),
),
}),
);
for (const row of bodyLines) {
const cells = splitTableRow(row);
// Pad/truncate to match the header column count so Word renders evenly.
const normalized = Array.from(
{ length: colCount },
(_, i) => cells[i] ?? "",
);
rows.push(
new TableRow({
children: normalized.map((c) => makeTableCell(c)),
}),
);
}
return new Table({
width: { size: 100, type: WidthType.PERCENTAGE },
rows,
borders: {
top: CELL_BORDER,
bottom: CELL_BORDER,
left: CELL_BORDER,
right: CELL_BORDER,
insideHorizontal: CELL_BORDER,
insideVertical: CELL_BORDER,
},
});
}
function isTableSeparator(line: string): boolean {
// e.g. |---|:---:|---:| with optional spaces
return /^\s*\|?(\s*:?-{3,}:?\s*\|)+\s*:?-{3,}:?\s*\|?\s*$/.test(line);
}
function looksLikeTableRow(line: string): boolean {
return /^\s*\|.*\|\s*$/.test(line) || /^\s*\|.+/.test(line);
}
/**
* Convert markdown text into an array of docx block elements (Paragraph | Table).
*/
function markdownToBlocks(markdown: string): Array<Paragraph | Table> {
const lines = markdown.replace(/\r\n/g, "\n").split("\n");
const blocks: Array<Paragraph | Table> = [];
let inFence = false;
let paragraphBuffer: string[] = [];
const flushParagraph = () => {
if (paragraphBuffer.length === 0) return;
blocks.push(makeParagraphFromInline(paragraphBuffer.join(" ")));
paragraphBuffer = [];
};
for (let i = 0; i < lines.length; i++) {
const rawLine = lines[i] ?? "";
const line = rawLine.replace(/\t/g, " ");
if (line.trim().startsWith("```")) {
flushParagraph();
inFence = !inFence;
continue;
}
if (inFence) {
blocks.push(makeCodeParagraph(line));
continue;
}
if (!line.trim()) {
flushParagraph();
continue;
}
const heading = /^(#{1,6})\s+(.*)$/.exec(line);
if (heading) {
flushParagraph();
blocks.push(
makeHeadingParagraph(heading[1]!.length, heading[2]!.trim()),
);
continue;
}
if (/^(?:---|\*\*\*|___)\s*$/.test(line)) {
flushParagraph();
blocks.push(makeHorizontalRule());
continue;
}
const blockquote = /^>\s?(.*)$/.exec(line);
if (blockquote) {
flushParagraph();
blocks.push(makeQuoteParagraph(blockquote[1] ?? ""));
continue;
}
// Table detection — a header row followed by a separator row.
if (
looksLikeTableRow(line) &&
i + 1 < lines.length &&
isTableSeparator(lines[i + 1] ?? "")
) {
flushParagraph();
const headerLine = line;
i += 2; // skip header + separator
const body: string[] = [];
while (i < lines.length && looksLikeTableRow(lines[i] ?? "")) {
body.push(lines[i] ?? "");
i++;
}
// Re-step back one so the outer loop's i++ lands on the next unprocessed
// line. Without this, blank lines after the table get swallowed.
i--;
blocks.push(buildTable(headerLine, body));
// Word's table rendering eats following spacing; inject a tiny gap.
blocks.push(
new Paragraph({ spacing: { after: 0 }, children: [new TextRun("")] }),
);
continue;
}
const orderedItem = /^(\s*)(\d+)\.\s+(.*)$/.exec(line);
if (orderedItem) {
flushParagraph();
const indent = Math.floor(orderedItem[1]!.length / 2);
blocks.push(makeListItem(orderedItem[3]!, true, indent));
continue;
}
const unorderedItem = /^(\s*)[-*+]\s+(.*)$/.exec(line);
if (unorderedItem) {
flushParagraph();
const indent = Math.floor(unorderedItem[1]!.length / 2);
blocks.push(makeListItem(unorderedItem[2]!, false, indent));
continue;
}
paragraphBuffer.push(line.trim());
}
flushParagraph();
return blocks;
}
/**
* Convert a markdown string into a .docx Blob suitable for download.
*/
/** Ask the Gateway to build a DOCX with its embedded licensed font resource. */
export async function markdownToDocxBlob(
markdown: string,
title?: string,
): Promise<Blob> {
const children: Array<Paragraph | Table> = [];
if (title?.trim()) {
children.push(
new Paragraph({
spacing: { before: 0, after: 240 },
children: [
new TextRun({
text: sanitizeText(title.trim()),
bold: true,
color: BLACK,
font: BODY_FONT,
size: 40, // 20pt
}),
],
}),
);
}
children.push(...markdownToBlocks(markdown));
const source = markdown?.trim();
if (!source) throw new Error("文档内容为空");
const doc = new Document({
styles: {
default: {
document: {
run: {
color: BLACK,
font: BODY_FONT,
size: BODY_SIZE_HALF_PT,
},
},
},
},
numbering: {
config: [
{
reference: "ordered-list",
levels: [
{
level: 0,
format: "decimal",
text: "%1.",
alignment: AlignmentType.START,
},
{
level: 1,
format: "decimal",
text: "%2.",
alignment: AlignmentType.START,
},
{
level: 2,
format: "decimal",
text: "%3.",
alignment: AlignmentType.START,
},
],
},
],
},
sections: [{ children }],
const endpoint = `${getBackendBaseURL()}/api/writing/export/docx`;
const response = await apiFetch(endpoint, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
title: title?.trim() || undefined,
markdown: source,
}),
});
return Packer.toBlob(doc);
if (!response.ok) throw await responseError(response);
const blob = await response.blob();
if (blob.size === 0) throw new Error("服务器返回了空的 Word 文件");
if (blob.type && blob.type !== DOCX_MEDIA_TYPE) {
throw new Error("服务器返回的文件类型不是 Word 文档");
}
return blob;
}
/**
* Trigger a browser download of the markdown source converted to .docx.
*/
/** Generate and immediately download a formal DOCX. */
export async function downloadMarkdownAsDocx(
markdown: string,
fileName: string,
@@ -488,15 +66,15 @@ export async function downloadMarkdownAsDocx(
const blob = await markdownToDocxBlob(markdown, title);
writeYLog({
optType: OptType.export,
optDetail: "CM助手-全量-定时任务-导出",
url: "/scheduled-tasks/export/docx",
optDetail: "CM助手-Word导出",
url: "/api/writing/export/docx",
reqParam: { fileName, title },
});
const safeName = fileName.replace(/\.(md|markdown)$/i, "") || "document";
const url = URL.createObjectURL(blob);
const anchor = document.createElement("a");
anchor.href = url;
anchor.download = `${safeName}.docx`;
anchor.download = `${sanitizeFilename(fileName)}.docx`;
anchor.rel = "noopener";
document.body.appendChild(anchor);
anchor.click();
frontend-web/src/core/threads/export.ts
+1 -1
View File
@@ -144,7 +144,7 @@ export function exportThreadAsJSON(thread: AgentThread, messages: Message[]) {
/**
* Export the conversation as a Word document (.docx). The whole thread is
* serialised to Markdown first (same formatter the `.md` export uses), and
* then converted to a docx Blob in the browser via `markdownToDocxBlob`.
* then sent to the Gateway's shared formal DOCX generator.
*/
export async function exportThreadAsWord(
thread: AgentThread,