@devries-cameron20 / t3code Mirror · read-only / commits / e330593141
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, 14:45 UTC

fix(web): bare URLs no longer swallow Arabic commas and question marks (#36779)

Co-authored-by: Julius Marminge <julius0216@outlook.com>

nasser committed Nov 14, 2026 at 17:53 UTC e3305931412d82f8cf09474b23c7e5409ee360b7
2 files changed +32 -23
packages/shared/src/markdownPipeline.test.ts
+9 -1
@@ -139,6 +139,14 @@ it.each([
139 ["[記事](https://example.com/a,b)", "[記事](https://example.com/a,b)"],
140 ["[http://example.com/a,b](http://example.com/a,b)", link("http://example.com/a,b")],
141 ["`https://example.com/a。次`", "`https://example.com/a。次`"],
142 + [
143 + "منشور على https://example.com/report، ونسخته",
144 + `منشور على ${link("https://example.com/report")}، ونسخته`,
145 + ],
146 + [
147 + "هل يعمل **http://localhost:3000**؟ نعم؛ www.example.com۔",
148 + `هل يعمل **${link("http://localhost:3000")}**؟ نعم؛ ${link("www.example.com")}۔`,
149 + ],
150 [
151 "https://x.com/a_ https://x.com/b_",
152 `${link("https://x.com/a_")} ${link("https://x.com/b_")}`,
@@ -169,7 +177,7 @@ it.each([
177 }
178 });
179
172 -it("ends bare URLs at CJK punctuation with source positions", () => {
180 +it("ends bare URLs at CJK and Arabic punctuation with source positions", () => {
181 const processor = unified().use(remarkParse).use(CHAT_MARKDOWN_REMARK_PLUGINS);
182 const parse = (markdown: string) => processor.runSync(processor.parse(markdown), markdown);
183 const at = (line: number, column: number, offset: number) => ({ line, column, offset });
packages/shared/src/markdownPipeline.ts
+23 -22
@@ -246,7 +246,7 @@ const CHAT_MARKDOWN_SANITIZE_SCHEMA = {
246
247 export const CHAT_MARKDOWN_REMARK_PLUGINS: PluggableList = [
248 remarkGfm,
249 - remarkEndAutolinksAtCjkPunctuation,
249 + remarkEndAutolinksAtScriptPunctuation,
250 remarkKeepAutolinkSuffixes,
251 remarkLinkAbsoluteFilePaths,
252 remarkKeepWindowsPaths,
@@ -261,7 +261,7 @@ export const CHAT_MARKDOWN_REMARK_PLUGINS: PluggableList = [
261
262 export const CHAT_MARKDOWN_REMARK_PLUGINS_WITH_BREAKS: PluggableList = [
263 remarkGfm,
264 - remarkEndAutolinksAtCjkPunctuation,
264 + remarkEndAutolinksAtScriptPunctuation,
265 remarkKeepAutolinkSuffixes,
266 remarkLinkAbsoluteFilePaths,
267 remarkKeepWindowsPaths,
@@ -471,20 +471,21 @@ function remarkKeepAutolinkSuffixes(): Transformer<Root> {
471 };
472 }
473
474 -// Punctuation and symbols from CJK Symbols and Punctuation and from Halfwidth and Fullwidth Forms.
475 -const CJK_PUNCTUATION = /(?=[ -〿＀-￯])[\p{P}\p{S}]/u;
474 +// Punctuation and symbols from CJK Symbols and Punctuation and from Halfwidth and Fullwidth Forms,
475 +// plus the Arabic-script comma, semicolon, question mark, and full stop.
476 +const SCRIPT_PUNCTUATION = /(?=[ -〿＀-￯،؛؟۔])[\p{P}\p{S}]/u;
477
478 /**
478 - * GFM runs a bare URL on until whitespace, so Chinese and Japanese punctuation joins it, along with
479 - * the rest of the sentence and any `**` that should have closed around it. A URL that is ASCII up to
480 - * such a mark ends there. Unicode earlier in the URL means the mark may belong to it, so it stays.
479 + * GFM runs a bare URL on until whitespace, so Chinese, Japanese, and Arabic punctuation joins it,
480 + * along with the rest of the sentence and any `**` that should have closed around it. A URL that is
481 + * ASCII up to such a mark ends there. Unicode earlier in the URL means the mark may belong to it, so it stays.
482 */
482 -function cjkPunctuationIndex(link: Link): number {
483 +function scriptPunctuationIndex(link: Link): number {
484 const label = link.children.length === 1 ? link.children[0] : undefined;
485 if (label?.type !== "text" || !/^https?:\/\//i.test(link.url)) return -1;
486 if (link.url !== label.value && link.url !== `http://${label.value}`) return -1;
487 const index = label.value.search(/[^!-~]/u);
487 - return index > 0 && CJK_PUNCTUATION.test(label.value.charAt(index)) ? index : -1;
488 + return index > 0 && SCRIPT_PUNCTUATION.test(label.value.charAt(index)) ? index : -1;
489 }
490
491 /** GFM's text pass links bare URLs without positions; its tokenizer's links span their URL. */
@@ -502,14 +503,14 @@ function isBareLink(link: Link, source: string): boolean {
503 }
504
505 /** Source offsets of the marks that tokenized bare URLs run past. */
505 -function cjkPunctuationOffsets(node: RootContent, source: string): number[] {
506 +function scriptPunctuationOffsets(node: RootContent, source: string): number[] {
507 if (node.type === "link") {
507 - const index = cjkPunctuationIndex(node);
508 + const index = scriptPunctuationIndex(node);
509 const start = node.position?.start.offset;
510 return index > 0 && start !== undefined && isBareLink(node, source) ? [start + index] : [];
511 }
512 return "children" in node && node.type !== "linkReference"
512 - ? (node.children as RootContent[]).flatMap((child) => cjkPunctuationOffsets(child, source))
513 + ? (node.children as RootContent[]).flatMap((child) => scriptPunctuationOffsets(child, source))
514 : [];
515 }
516
@@ -537,7 +538,7 @@ function countBelow(sorted: readonly number[], value: number): number {
538 * dropped and positions mapped back to the source. Returns false and leaves the block alone when its
539 * text would parse differently on its own, as a quote's continuation lines do.
540 */
540 -function reparseAtCjkPunctuation(
541 +function reparseAtScriptPunctuation(
542 processor: Processor,
543 block: Paragraph | Heading | TableCell,
544 source: string,
@@ -570,7 +571,7 @@ function reparseAtCjkPunctuation(
571 .join("\t");
572 const [paragraph, ...rest] = (processor.parse(marked) as Root).children;
573 if (paragraph?.type !== "paragraph" || rest.length > 0) return false;
573 - const more = cjkPunctuationOffsets(paragraph, marked);
574 + const more = scriptPunctuationOffsets(paragraph, marked);
575 if (more.length > 0) {
576 marks.push(...more.map(unmarked));
577 continue;
@@ -613,16 +614,16 @@ function reparseAtCjkPunctuation(
614 }
615
616 /** Ends the bare URLs a reparse could not, and those GFM's text pass linked after emphasis. */
616 -function splitAtCjkPunctuation(node: RootContent, source: string) {
617 +function splitAtScriptPunctuation(node: RootContent, source: string) {
618 if (!("children" in node) || node.type === "link" || node.type === "linkReference") return;
619 const children = node.children as RootContent[];
620 for (let index = 0; index < children.length; index++) {
621 const child = children[index]!;
622 if (child.type !== "link") {
622 - splitAtCjkPunctuation(child, source);
623 + splitAtScriptPunctuation(child, source);
624 continue;
625 }
625 - const at = isBareLink(child, source) ? cjkPunctuationIndex(child) : -1;
626 + const at = isBareLink(child, source) ? scriptPunctuationIndex(child) : -1;
627 const label = child.children[0];
628 if (at < 0 || label?.type !== "text") continue;
629 // GFM drops this punctuation from the end of a URL; remarkKeepAutolinkSuffixes restores `_` and `~`.
@@ -646,15 +647,15 @@ function splitAtCjkPunctuation(node: RootContent, source: string) {
647 }
648 }
649
649 -function remarkEndAutolinksAtCjkPunctuation(this: Processor): Transformer<Root> {
650 +function remarkEndAutolinksAtScriptPunctuation(this: Processor): Transformer<Root> {
651 return (tree, file) => {
652 const source = String(file);
652 - if (!CJK_PUNCTUATION.test(source)) return;
653 + if (!SCRIPT_PUNCTUATION.test(source)) return;
654 const visit = (node: RootContent) => {
655 if (node.type === "paragraph" || node.type === "heading" || node.type === "tableCell") {
655 - const offsets = cjkPunctuationOffsets(node, source);
656 - if (offsets.length > 0) reparseAtCjkPunctuation(this, node, source, offsets);
657 - splitAtCjkPunctuation(node, source);
656 + const offsets = scriptPunctuationOffsets(node, source);
657 + if (offsets.length > 0) reparseAtScriptPunctuation(this, node, source, offsets);
658 + splitAtScriptPunctuation(node, source);
659 } else if ("children" in node) {
660 (node.children as RootContent[]).forEach(visit);
661 }