Skip to content

Your lightweight open-source personal agent.

fix(webui): render CJK bold labels before Latin text (#6099)

chengyongru59b89a834c18
diff --git a/webui/bun.lock b/webui/bun.lock
index b74767da7..7d6064d49 100644
--- a/webui/bun.lock
+++ b/webui/bun.lock
@@ -19,6 +19,7 @@
"i18next": "^26.0.6",
"lucide-react": "^0.469.0",
"micromark-factory-space": "^2.0.1",
+ "micromark-util-decode-string": "^2.0.1",
"qrcode": "^1.5.4",
"react": "^18.3.1",
"react-dom": "^18.3.1",
@@ -556,7 +557,7 @@
"chai": ["[email protected]", "", { "dependencies": { "assertion-error": "^2.0.1", "check-error": "^2.1.1", "deep-eql": "^5.0.1", "loupe": "^3.1.0", "pathval": "^2.0.0" } }, "sha512-4zNhdJD/iOjSH0A05ea+Ke6MU5mmpQcbQsSOkgdaUMJ9zTlDTD/GYlwohmIE2u0gaxHYiVHEn1Fw9mZ/ktJWgw=="],
- "character-entities": ["[email protected]", "", {}, "sha512-iBMyeEHxfVnIakwOuDXpVkc54HijNgCyQB2w0VfGQThle6NXn50zU6V/u+LDhxHcDUPojn6Kpga3PTAD8W1bQw=="],
+ "character-entities": ["[email protected]", "", {}, "sha512-shx7oQ0Awen/BRIdkjkvz54PnEEI/EjwXDSIZp86/KKdbafHh1Df/RYGBhn4hbe2+uKC9FnT5UCEdyPz3ai9hQ=="],
"character-entities-html4": ["[email protected]", "", {}, "sha512-1v7fgQRj6hnSwFpq1Eu0ynr/CDEw0rXo2B61qXrLNdHZmPKgb7fqS1a2JwF0rISo9q77jDI8VMEHoApn8qDoZA=="],
@@ -1424,8 +1425,6 @@
"d3-sankey/d3-shape": ["[email protected]", "", { "dependencies": { "d3-path": "1" } }, "sha512-EUkvKjqPFUAZyOlhY5gzCxCeI0Aep04LwIRpsZ/mLFelJiUfnK56jo5JMDSE7yyP2kLSb6LtF+S5chMk7uqPqw=="],
- "decode-named-character-reference/character-entities": ["[email protected]", "", {}, "sha512-shx7oQ0Awen/BRIdkjkvz54PnEEI/EjwXDSIZp86/KKdbafHh1Df/RYGBhn4hbe2+uKC9FnT5UCEdyPz3ai9hQ=="],
-
"fast-glob/glob-parent": ["[email protected]", "", { "dependencies": { "is-glob": "^4.0.1" } }, "sha512-AOIgSQCepiJYwP3ARnGx+5VnTu2HBYdzbGP45eLw1vr3zB3vZLeyed1sC9hnbcOc9/SrMyM5RPQrkGz4aS9Zow=="],
"glob/minimatch": ["[email protected]", "", { "dependencies": { "brace-expansion": "^2.0.2" } }, "sha512-OBwBN9AL4dqmETlpS2zasx+vTeWclWzkblfZk7KTA5j3jeOONz/tRCnZomUyvNg83wL5Zv9Ss6HMJXAgL8R2Yg=="],
@@ -1450,6 +1449,8 @@
"mermaid/marked": ["[email protected]", "", { "bin": { "marked": "bin/marked.js" } }, "sha512-TI3V8YYWvkVf3KJe1dRkpnjs68JUPyEa5vjKrp1XEEJUAOaQc+Qj+L1qWbPd0SJuAdQkFU0h73sXXqwDYxsiDA=="],
+ "parse-entities/character-entities": ["[email protected]", "", {}, "sha512-iBMyeEHxfVnIakwOuDXpVkc54HijNgCyQB2w0VfGQThle6NXn50zU6V/u+LDhxHcDUPojn6Kpga3PTAD8W1bQw=="],
+
"path-scurry/lru-cache": ["[email protected]", "", {}, "sha512-JNAzZcXrCt42VGLuYz0zfAzDfAvJWW6AfYlDBQyDV5DClI2m5sAmK+OIO7s59XfsRsWHp02jAJrRadPRGTt6SQ=="],
"postcss-nested/postcss-selector-parser": ["[email protected]", "", { "dependencies": { "cssesc": "^3.0.0", "util-deprecate": "^1.0.2" } }, "sha512-Q8qQfPiZ+THO/3ZrOrO0cJJKfpYCagtMUkXbnEfmgUjwXg6z/WBeOyS9APBBPCTSiDV+s4SwQGu8yFsiMRIudg=="],
diff --git a/webui/package.json b/webui/package.json
index af8ed1a1e..2820525bc 100644
--- a/webui/package.json
+++ b/webui/package.json
@@ -27,6 +27,7 @@
"i18next": "^26.0.6",
"lucide-react": "^0.469.0",
"micromark-factory-space": "^2.0.1",
+ "micromark-util-decode-string": "^2.0.1",
"qrcode": "^1.5.4",
"react": "^18.3.1",
"react-dom": "^18.3.1",
diff --git a/webui/src/components/MarkdownTextRenderer.tsx b/webui/src/components/MarkdownTextRenderer.tsx
index 1541e1953..772ae0a7a 100644
--- a/webui/src/components/MarkdownTextRenderer.tsx
+++ b/webui/src/components/MarkdownTextRenderer.tsx
@@ -8,6 +8,7 @@ import {
} from "react";
import { useTranslation } from "react-i18next";
import { Check, Globe2 } from "lucide-react";
+import { decodeString } from "micromark-util-decode-string";
import remarkBreaks from "remark-breaks";
import remarkGfm from "remark-gfm";
import remarkMath from "remark-math";
@@ -54,6 +55,10 @@ type MarkdownAstNode = {
type: string;
value?: string;
children?: MarkdownAstNode[];
+ position?: {
+ start: { offset: number };
+ end: { offset: number };
+ };
data?: {
hName?: string;
};
@@ -245,48 +250,58 @@ function remarkSafeHtmlSubset() {
};
}
-// Recover a common model-output edge case that CommonMark leaves as literal
-// text: `**结论。**如果`, with no separator after the closing delimiter.
-const CJK_AFTER_STRONG =
- /(?<!\\)\*\*([^*\r\n]+?)(?<!\\)\*\*(?=[\p{Script=Han}\p{Script=Hiragana}\p{Script=Katakana}\p{Script=Hangul}])/gu;
+// Model output can put CJK punctuation or whitespace before a closing `**`
+// followed immediately by a word, which CommonMark leaves as literal text.
+const STRONG_BEFORE_WORD = /(?<!\\)\*\*([^\s*][^*\r\n]*?)(?<!\\)\*\*(?=[\p{L}\p{N}])/gu;
+const CJK_CHARACTER = /[\p{Script=Han}\p{Script=Hiragana}\p{Script=Katakana}\p{Script=Hangul}]/u;
-function normalizeCjkStrongBoundaries(node: MarkdownAstNode): void {
+function normalizeCjkStrongBoundaries(node: MarkdownAstNode, source: string): void {
if (!node.children) return;
node.children = node.children.flatMap((child) => {
- if (child.type !== "text" || !child.value?.includes("**")) {
- normalizeCjkStrongBoundaries(child);
+ if (child.type !== "text" || !child.value?.includes("**") || !child.position) {
+ normalizeCjkStrongBoundaries(child, source);
return [child];
}
+ // Match source delimiters before decoding escapes and character references.
+ // Paragraph continuation indentation is not part of the parsed text value.
+ const textSource = source.slice(child.position.start.offset, child.position.end.offset)
+ .replace(/(\r\n?|\n)[\t ]+/g, "$1");
+
const replacement: MarkdownAstNode[] = [];
let cursor = 0;
- for (const match of child.value.matchAll(CJK_AFTER_STRONG)) {
+ for (const match of textSource.matchAll(STRONG_BEFORE_WORD)) {
const start = match.index;
- if (start > cursor) replacement.push(safeText(child.value.slice(cursor, start)));
+ const end = start + match[0].length;
+ const content = decodeString(match[1]);
+ if (!CJK_CHARACTER.test(content + textSource.slice(end, end + 1))) continue;
+ const label = content.trimEnd();
+ if (start > cursor) replacement.push(safeText(decodeString(textSource.slice(cursor, start))));
replacement.push({
type: "strong",
- children: [safeText(match[1])],
+ children: [safeText(label)],
});
- cursor = start + match[0].length;
+ if (label.length < content.length) replacement.push(safeText(content.slice(label.length)));
+ cursor = end;
}
if (cursor === 0) return [child];
- if (cursor < child.value.length) replacement.push(safeText(child.value.slice(cursor)));
+ if (cursor < textSource.length) replacement.push(safeText(decodeString(textSource.slice(cursor))));
return replacement;
});
}
function remarkCjkStrongBoundaries() {
- return (tree: MarkdownAstNode) => {
- normalizeCjkStrongBoundaries(tree);
+ return (tree: MarkdownAstNode, file: { toString(): string }) => {
+ normalizeCjkStrongBoundaries(tree, file.toString());
};
}
const remarkPlugins: NonNullable<StreamdownProps["remarkPlugins"]> = [
+ remarkCjkStrongBoundaries,
remarkBreaks,
remarkGfm,
[remarkMath, { singleDollarTextMath: false }],
remarkTexMath,
- remarkCjkStrongBoundaries,
remarkSafeHtmlSubset,
];
type MathPlugin = typeof import("@/lib/markdown-math").default;
diff --git a/webui/src/tests/markdown-text-renderer.test.tsx b/webui/src/tests/markdown-text-renderer.test.tsx
index b4b135a42..913c0504c 100644
--- a/webui/src/tests/markdown-text-renderer.test.tsx
+++ b/webui/src/tests/markdown-text-renderer.test.tsx
@@ -601,6 +601,53 @@ describe("MarkdownTextRenderer", () => {
expect(screen.getByText(/如果你之后不想再用/)).toBeInTheDocument();
});
+ it.each([
+ ["**边界说明:**issue 只有截图,没有原始消息。", "issue 只有截图,没有原始消息。"],
+ ["**边界说明: **issue 只有截图,没有原始消息。", " issue 只有截图,没有原始消息。"],
+ ])("renders CJK bold labels before Latin text through stream completion: %s", (source, following) => {
+ const { container, rerender } = render(<MarkdownTextRenderer>{source}</MarkdownTextRenderer>);
+ expect(container.querySelector("strong")).toHaveTextContent("边界说明:");
+ expect(container.textContent).toBe(`边界说明:${following}`);
+
+ for (let end = 1; end <= source.length; end += 1) {
+ rerender(<MarkdownTextRenderer streaming preserveStreamingLayout>{source.slice(0, end)}</MarkdownTextRenderer>);
+ }
+ expect(container.querySelector("strong")).toHaveTextContent("边界说明:");
+ rerender(<MarkdownTextRenderer preserveStreamingLayout>{source}</MarkdownTextRenderer>);
+ expect(container.querySelector("strong")).toHaveTextContent("边界说明:");
+ expect(container.textContent).toBe(`边界说明:${following}`);
+ });
+
+ it("keeps literal bold markers in code and escaped text", () => {
+ const { container } = render(
+ <MarkdownTextRenderer highlightCode={false}>
+ {"`**边界说明:**issue`\n\n```text\n**边界说明: **issue\n```\n\n\\*\\*边界说明:\\*\\*issue"}
+ </MarkdownTextRenderer>,
+ );
+ expect(container.querySelector("strong")).toBeNull();
+ expect(container).toHaveTextContent("**边界说明:**issue");
+ expect(container).toHaveTextContent("**边界说明: **issue");
+ expect(container.querySelector("p:last-child")?.textContent).toBe("**边界说明:**issue");
+ });
+
+ it.each([
+ ["**结论:**继续 &amp; 其他说明。", "结论:继续 & 其他说明。"],
+ ["**边界说明:**issue &amp; details", "边界说明:issue & details"],
+ ["先写转义标点 \\#,然后 **结论:**继续说明。", "先写转义标点 #,然后 结论:继续说明。"],
+ ["\\*\\*字面符号:\\*\\*issue 和 **结论:**继续说明。", "**字面符号:**issue 和 结论:继续说明。"],
+ ["&#42;&#42;字面符号:&#42;&#42;issue 和 **结论:**继续说明。", "**字面符号:**issue 和 结论:继续说明。"],
+ ["- 第一行\n **结论:**继续 &amp; 其他说明。", "第一行\n结论:继续 & 其他说明。"],
+ ])("recovers CJK bold labels without reinterpreting decoded text: %s", (source, text) => {
+ const { container, rerender } = render(<MarkdownTextRenderer>{source}</MarkdownTextRenderer>);
+ expect(container.querySelectorAll("strong")).toHaveLength(1);
+ expect(container.textContent?.trim()).toBe(text);
+ rerender(<MarkdownTextRenderer streaming preserveStreamingLayout>{source}</MarkdownTextRenderer>);
+ expect(container.querySelectorAll("strong")).toHaveLength(1);
+ rerender(<MarkdownTextRenderer preserveStreamingLayout>{source}</MarkdownTextRenderer>);
+ expect(container.querySelectorAll("strong")).toHaveLength(1);
+ expect(container.textContent?.trim()).toBe(text);
+ });
+
it("adds line numbers to multiline fenced code without changing inline code", () => {
render(
<MarkdownTextRenderer highlightCode={false}>