fix(markdown): autolink read-only URLs in the parse tree, not raw text

Read-only markdown surfaces (comments, descriptions, chat) pre-linkified
bare URLs by rewriting the raw source to [url](url) before parsing. Because
linkify-it treats `*` as a valid URL character, a bare URL followed by a
bold close — `**PR:https://…/5081**` — had the trailing `**` swallowed into
the match and rewritten as [url**](url**). That consumed the emphasis closer
(the bold never closed; the leading `**` rendered as literal asterisks) and
corrupted the href with a trailing `**` (MUL-4242).

Let remark-gfm autolink URLs in the parse tree instead, where emphasis is
already resolved so an adjacent delimiter can never be absorbed. The custom
string pass now runs in a `urls: false` mode on read-only surfaces and only
linkifies file paths (which gfm never does). A small remark plugin
(remark-cjk-autolink) re-applies the existing CJK URL boundary to gfm's
autolink literals so `https://x/a。后面` still stops at 。.

The Tiptap editor path is unchanged (`urls: true`): @tiptap/markdown does not
autolink bare URLs, so it still needs the string pass.

Note: read-only URL autolinking now follows GFM semantics (scheme, www., or
email required); bare fuzzy domains like `NBA.com` render as plain text on
read-only surfaces, matching CommonMark/GFM.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
Naiyuan Qing
2026-07-08 17:09:06 +08:00
parent 947a54b674
commit 9df40b7b13
8 changed files with 208 additions and 15 deletions

View File

@@ -12,6 +12,7 @@ import { CODE_LIGATURE_CLASS } from '@multica/ui/lib/code-style'
import { CodeBlock, InlineCode } from './CodeBlock'
import { isAllowedFileCardHref, preprocessFileCards } from './file-cards'
import { preprocessLinks } from './linkify'
import { remarkCjkAutolink } from './remark-cjk-autolink'
import { preprocessMentionShortcodes } from './mentions'
import 'katex/dist/katex.min.css'
import './markdown.css'
@@ -449,11 +450,15 @@ export function Markdown({
[mode, onUrlClick, onFileClick, renderMention, renderImage, renderFileCard]
)
// Preprocess: convert mention shortcodes, raw URLs, and file cards to renderable content
// Preprocess: convert mention shortcodes, file paths, and file cards to
// renderable content. URLs are intentionally left bare (urls: false) so
// remark-gfm autolinks them in the parse tree — a bare URL can then no longer
// swallow an adjacent markdown delimiter like a closing `**` (MUL-4242).
// remarkCjkAutolink re-applies the CJK boundary on gfm's autolink nodes.
const processedContent = React.useMemo(
() => {
let result = preprocessMentionShortcodes(children)
result = preprocessLinks(result)
result = preprocessLinks(result, { urls: false })
result = preprocessFileCards(result, cdnDomain ?? '')
return result
},
@@ -467,6 +472,7 @@ export function Markdown({
[remarkMath, { singleDollarTextMath: false }],
remarkBreaks,
[remarkGfm, { singleTilde: false }],
remarkCjkAutolink,
]}
rehypePlugins={[rehypeRaw, [rehypeSanitize, sanitizeSchema], rehypeKatex]}
urlTransform={urlTransform}

View File

@@ -1,7 +1,8 @@
export { Markdown, MemoizedMarkdown, type MarkdownProps, type RenderMode } from './Markdown'
export { CodeBlock, InlineCode, type CodeBlockProps } from './CodeBlock'
export { StreamingMarkdown, type StreamingMarkdownProps } from './StreamingMarkdown'
export { preprocessLinks, detectLinks, hasLinks } from './linkify'
export { preprocessLinks, detectLinks, hasLinks, CJK_URL_TERMINATOR_REGEX } from './linkify'
export { remarkCjkAutolink } from './remark-cjk-autolink'
export { preprocessMentionShortcodes } from './mentions'
export {
preprocessFileCards,

View File

@@ -35,7 +35,10 @@ const BARE_FILENAME_REGEX = new RegExp(`^[\\w.-]+\\.(?:${FILE_EXTENSIONS})$`, 'i
// character up to the next whitespace swallowed into the href. We truncate the
// detected URL at the first occurrence of any of these characters. Character
// set mirrors the fix applied in mattermost/marked#22.
const CJK_URL_TERMINATOR_REGEX =
//
// Exported so the read-only render pipeline can apply the same boundary to URLs
// that remark-gfm autolinks in the parse tree (see remark-cjk-autolink.ts).
export const CJK_URL_TERMINATOR_REGEX =
/[----~、。「-】]/
interface DetectedLink {
@@ -270,13 +273,18 @@ function collectLinkifyMatches(text: string, offset: number, out: DetectedLink[]
}
/**
* Detect all links (URLs, emails, file paths) in text
* Detect all links (URLs, emails, file paths) in text.
*
* `includeUrls` gates the URL/email pass. Read-only markdown renderers pass
* `false` and let remark-gfm autolink URLs in the parse tree instead, which
* cannot corrupt adjacent markdown (e.g. a trailing `**`). File paths, which
* remark-gfm does not linkify, are always detected. See preprocessLinks.
*/
export function detectLinks(text: string): DetectedLink[] {
export function detectLinks(text: string, includeUrls = true): DetectedLink[] {
const links: DetectedLink[] = []
// 1. Detect URLs and emails with linkify-it, applying CJK boundary handling.
collectLinkifyMatches(text, 0, links)
if (includeUrls) collectLinkifyMatches(text, 0, links)
// 2. Detect file paths with custom regex
// Reset regex state
@@ -310,10 +318,20 @@ export function detectLinks(text: string): DetectedLink[] {
}
/**
* Preprocess text to convert raw URLs and file paths into markdown links
* Skips code blocks and already-linked content
* Preprocess text to convert raw URLs and file paths into markdown links.
* Skips code blocks and already-linked content.
*
* `opts.urls` (default `true`) controls the URL/email pass. The Tiptap editor
* keeps it on because @tiptap/markdown does not autolink bare URLs. Read-only
* react-markdown renderers pass `false`: they let remark-gfm autolink URLs in
* the parse tree, where a bare URL can no longer swallow an adjacent markdown
* delimiter — the string pass here can't tell `https://x**` (URL + bold close)
* from a URL that legitimately ends in `*`, so it corrupted both (MUL-4242).
* File paths (which remark-gfm never linkifies) are converted in both modes.
*/
export function preprocessLinks(text: string): string {
export function preprocessLinks(text: string, opts?: { urls?: boolean }): string {
const includeUrls = opts?.urls ?? true
// Quick check - if no potential links, return early
if (!linkify.pretest(text) && !/[~/.]\//.test(text)) {
return text
@@ -321,7 +339,7 @@ export function preprocessLinks(text: string): string {
const codeRanges = findCodeRanges(text)
const markdownLinkRanges = findMarkdownLinkRanges(text)
const links = detectLinks(text)
const links = detectLinks(text, includeUrls)
if (links.length === 0) return text

View File

@@ -0,0 +1,79 @@
import { CJK_URL_TERMINATOR_REGEX } from './linkify'
/**
* remark-cjk-autolink — trim CJK punctuation that remark-gfm's autolink literal
* swallowed into a URL.
*
* Read-only renderers let remark-gfm autolink bare URLs in the parse tree, so an
* adjacent markdown delimiter (e.g. a closing `**`) is never absorbed into the
* href — that was MUL-4242. gfm's autolink literal, however, shares linkify-it's
* CJK weakness: `https://x/a。后面` extends the link across the ideographic full
* stop and the run after it. preprocessLinks used to trim this before parsing;
* since URLs are no longer preprocessed in read-only mode, we re-apply the same
* boundary on the parsed tree.
*
* Only autolink *literals* are touched — links whose href was derived from the
* visible text (`https://…`, `www.…` → `http://…`, `a@b` → `mailto:a@b`).
* Explicit `[label](url)` links keep whatever destination the author wrote, even
* when it contains CJK punctuation.
*/
interface MdNode {
type: string
url?: string
value?: string
children?: MdNode[]
}
// The scheme prefix remark-gfm prepends to an autolink literal's href. Returns
// null when `url` was not derived from `text`, i.e. an explicit link — leave it.
function autolinkSchemePrefix(url: string, text: string): string | null {
if (url === text) return ''
for (const prefix of ['http://', 'https://', 'mailto:']) {
if (url === prefix + text) return prefix
}
return null
}
// If `node` is an autolink literal whose text runs past a CJK terminator, return
// [trimmed link, trailing text] to splice in its place; otherwise null.
function splitCjkAutolink(node: MdNode): MdNode[] | null {
if (node.type !== 'link' || !node.url || node.children?.length !== 1) return null
const child = node.children[0]
if (!child || child.type !== 'text' || typeof child.value !== 'string') return null
const text = child.value
if (autolinkSchemePrefix(node.url, text) === null) return null
const cut = text.search(CJK_URL_TERMINATOR_REGEX)
if (cut <= 0) return null // no CJK punctuation, or the text starts with it
const rest = text.slice(cut)
const trimmed: MdNode = {
...node,
url: node.url.slice(0, node.url.length - rest.length),
children: [{ type: 'text', value: text.slice(0, cut) }],
}
return [trimmed, { type: 'text', value: rest }]
}
function transform(node: MdNode): void {
const children = node.children
if (!children) return
for (let i = 0; i < children.length; i++) {
const split = splitCjkAutolink(children[i]!)
if (split) {
children.splice(i, 1, ...split)
i += split.length - 1 // skip the appended trailing-text node
} else {
transform(children[i]!)
}
}
}
/** unified/remark plugin. Attach after remark-gfm. */
export function remarkCjkAutolink() {
return (tree: unknown): void => {
transform(tree as MdNode)
}
}

View File

@@ -616,3 +616,54 @@ describe("ReadonlyContent slash command rendering", () => {
expect(container.querySelector("a")).not.toBeNull();
});
});
describe("ReadonlyContent bare URL autolinking (MUL-4242)", () => {
// A bare URL wrapped in bold used to be pre-linkified into [url**](url**),
// which swallowed the closing `**`: the bold never closed (leading `**`
// showed as literal asterisks) and the href was corrupted with a trailing
// `**`. URLs are now autolinked by remark-gfm in the parse tree — after
// emphasis is resolved — so an adjacent delimiter can no longer be absorbed.
it("renders a bold-wrapped bare URL as bold plus a clean link", () => {
const url = "https://github.com/multica-ai/multica/pull/5081";
const { container } = render(<ReadonlyContent content={`**PR${url}**`} />);
const strong = container.querySelector("strong");
expect(strong).not.toBeNull();
const anchor = strong!.querySelector("a");
expect(anchor?.getAttribute("href")).toBe(url);
// No literal asterisks leak into the text, no trailing `**` in the href.
expect(container.textContent).not.toContain("**");
expect(anchor?.getAttribute("href")).not.toContain("*");
});
it("still autolinks a plain bare URL", () => {
const { container } = render(
<ReadonlyContent content={"see https://example.com/foo here"} />,
);
expect(container.querySelector('a[href="https://example.com/foo"]')).not.toBeNull();
});
it("stops an autolinked URL at CJK punctuation instead of swallowing it", () => {
const { container } = render(
<ReadonlyContent content={"见 https://example.com/foo。后面还有字"} />,
);
const anchor = container.querySelector("a");
expect(anchor?.getAttribute("href")).toBe("https://example.com/foo");
expect(anchor?.textContent).toBe("https://example.com/foo");
// The CJK tail stays outside the link.
expect(container.textContent).toContain("。后面还有字");
});
it("leaves an explicit link's destination untouched even when it ends in CJK", () => {
const { container } = render(
<ReadonlyContent content={"[看](https://example.com/x。)后文"} />,
);
const anchor = container.querySelector("a");
// react-markdown percent-encodes the CJK char; the point is it is NOT
// trimmed off the way an autolink literal would be.
expect(decodeURIComponent(anchor?.getAttribute("href") ?? "")).toBe(
"https://example.com/x。",
);
expect(anchor?.textContent).toBe("看");
});
});

View File

@@ -41,7 +41,7 @@ import { IssueMentionCard } from "../issues/components/issue-mention-card";
import { ProjectChip } from "../projects/components/project-chip";
import { useLinkHover, LinkHoverCard } from "./link-hover-card";
import { openLink, isMentionHref } from "./utils/link-handler";
import { isAllowedFileCardHref } from "@multica/ui/markdown";
import { isAllowedFileCardHref, remarkCjkAutolink } from "@multica/ui/markdown";
import { preprocessMarkdown } from "./utils/preprocess";
import { highlightToHtml } from "./utils/highlight-markdown";
import { MermaidDiagram } from "./mermaid-diagram";
@@ -432,8 +432,11 @@ export const ReadonlyContent = memo(function ReadonlyContent({
className,
attachments,
}: ReadonlyContentProps) {
// urls: false — let remark-gfm autolink bare URLs in the parse tree so an
// adjacent markdown delimiter (e.g. a closing `**`) is never swallowed into
// the href (MUL-4242). remarkCjkAutolink below re-applies the CJK boundary.
const processed = useMemo(
() => highlightToHtml(preprocessMarkdown(content)),
() => highlightToHtml(preprocessMarkdown(content, { urls: false })),
[content],
);
const wrapperRef = useRef<HTMLDivElement>(null);
@@ -458,6 +461,7 @@ export const ReadonlyContent = memo(function ReadonlyContent({
[remarkMath, { singleDollarTextMath: false }],
remarkBreaks,
[remarkGfm, { singleTilde: false }],
remarkCjkAutolink,
]}
rehypePlugins={[rehypeRaw, [rehypeSanitize, sanitizeSchema], rehypeKatex]}
urlTransform={urlTransform}

View File

@@ -136,3 +136,33 @@ describe("preprocessLinks — bare filenames are not auto-linked as URLs", () =>
);
});
});
// Read-only react-markdown renderers pass { urls: false } and let remark-gfm
// autolink URLs in the parse tree instead, so a bare URL can no longer swallow
// an adjacent markdown delimiter like a closing ** (MUL-4242). File paths, which
// remark-gfm never linkifies, are still converted.
describe("preprocessLinks — urls:false (read-only mode)", () => {
it("leaves bare URLs untouched so remark-gfm can autolink them", () => {
expect(preprocessLinks("see https://example.com/x here", { urls: false })).toBe(
"see https://example.com/x here",
);
});
it("does not rewrite a bold-wrapped URL into [url**](url**) (the root cause)", () => {
expect(preprocessLinks("**PRhttps://example.com/x**", { urls: false })).toBe(
"**PRhttps://example.com/x**",
);
});
it("still linkifies explicit ./ file paths", () => {
expect(preprocessLinks("see ./src/main.go here", { urls: false })).toBe(
"see [./src/main.go](./src/main.go) here",
);
});
it("default mode still linkifies URLs (editor path unchanged)", () => {
expect(preprocessLinks("see https://example.com/x here")).toBe(
"see [https://example.com/x](https://example.com/x) here",
);
});
});

View File

@@ -14,12 +14,16 @@ import { configStore } from "@multica/core/config";
* 2. Raw URLs → markdown links via linkify-it (so they render as clickable Link nodes)
* 3. File card syntax (new !file[name](url) + legacy [name](cdnUrl)) → HTML div for
* fileCard node parsing
*
* `opts.urls` (default `true`) forwards to preprocessLinks. The Tiptap editor
* needs it on; read-only react-markdown surfaces pass `false` and let remark-gfm
* autolink URLs in the parse tree instead (MUL-4242). See preprocessLinks.
*/
export function preprocessMarkdown(markdown: string): string {
export function preprocessMarkdown(markdown: string, opts?: { urls?: boolean }): string {
if (!markdown) return "";
const cdnDomain = configStore.getState().cdnDomain;
const step1 = preprocessMentionShortcodes(markdown);
const step2 = preprocessLinks(step1);
const step2 = preprocessLinks(step1, opts);
const step3 = preprocessFileCards(step2, cdnDomain);
return step3;
}