deepseek-harness/patches/micromark-extension-gfm-autolink-literal@2.1.0.patch

diff --git a/dev/lib/syntax.js b/dev/lib/syntax.js
index 904e27200ae4980183beda525d33feef70c2e7f7..1e05880bc1e7df0e05137811cd182f10872985df 100644
--- a/dev/lib/syntax.js
+++ b/dev/lib/syntax.js
@@ -16,6 +16,7 @@ const wwwPrefix = {tokenize: tokenizeWwwPrefix, partial: true}
 const domain = {tokenize: tokenizeDomain, partial: true}
 const path = {tokenize: tokenizePath, partial: true}
 const trail = {tokenize: tokenizeTrail, partial: true}
+const astralPunctuation = {tokenize: tokenizeAstralPunctuation, partial: true}
 const emailDomainDotTrail = {
   tokenize: tokenizeEmailDomainDotTrail,
   partial: true
@@ -659,10 +660,15 @@ function tokenizePath(effects, ok) {
       return effects.check(trail, ok, pathAtPunctuation)(code)
     }
 
+    if (highSurrogate(code)) {
+      return effects.check(astralPunctuation, ok, pathAtPunctuation)(code)
+    }
+
     if (
       code === codes.eof ||
       markdownLineEndingOrSpace(code) ||
-      unicodeWhitespace(code)
+      unicodeWhitespace(code) ||
+      nonAsciiPunctuation(code)
     ) {
       return ok(code)
     }
@@ -762,11 +768,16 @@ function tokenizeTrail(effects, ok, nok) {
       // So is whitespace.
       code === codes.eof ||
       markdownLineEndingOrSpace(code) ||
-      unicodeWhitespace(code)
+      unicodeWhitespace(code) ||
+      nonAsciiPunctuation(code)
     ) {
       return ok(code)
     }
 
+    if (highSurrogate(code)) {
+      return effects.check(astralPunctuation, ok, nok)(code)
+    }
+
     return nok(code)
   }
 
@@ -932,6 +943,56 @@ function previousEmail(code) {
   return !(code === codes.slash || gfmAtext(code))
 }
 
+/**
+ * Non-ASCII punctuation or symbol, such as `。` or `，`, which ends a literal
+ * in CJK prose.
+ *
+ * @param {Code} code
+ * @returns {boolean}
+ */
+function nonAsciiPunctuation(code) {
+  return code !== codes.eof && code > 127 && unicodePunctuation(code)
+}
+
+/**
+ * Whether `code` is a UTF-16 high surrogate, which starts a character outside
+ * the BMP such as `😀`.
+ *
+ * @param {Code} code
+ * @returns {boolean}
+ */
+function highSurrogate(code) {
+  return code !== codes.eof && code >= 0xd800 && code <= 0xdbff
+}
+
+/**
+ * Check whether a surrogate pair is punctuation or a symbol.
+ *
+ * @this {TokenizeContext}
+ * @type {Tokenizer}
+ */
+function tokenizeAstralPunctuation(effects, ok, nok) {
+  /** @type {number} */
+  let high
+
+  return start
+
+  /** @type {State} */
+  function start(code) {
+    high = /** @type {number} */ (code)
+    effects.consume(code)
+    return low
+  }
+
+  /** @type {State} */
+  function low(code) {
+    return code !== codes.eof && code >= 0xdc00 && code <= 0xdfff &&
+      /[\p{P}\p{S}]/u.test(String.fromCharCode(high, code))
+      ? ok(code)
+      : nok(code)
+  }
+}
+
 /**
  * @param {Code} code
  * @returns {boolean}
diff --git a/lib/syntax.js b/lib/syntax.js
index dffbaf15146b34d3b3f634286629ccb148866f4c..60b1d5d43811b31592728b4539c5f4d43174dd35 100644
--- a/lib/syntax.js
+++ b/lib/syntax.js
@@ -19,6 +19,10 @@ const trail = {
   tokenize: tokenizeTrail,
   partial: true
 };
+const astralPunctuation = {
+  tokenize: tokenizeAstralPunctuation,
+  partial: true
+};
 const emailDomainDotTrail = {
   tokenize: tokenizeEmailDomainDotTrail,
   partial: true
@@ -584,7 +588,10 @@ function tokenizePath(effects, ok) {
     if (code === 33 || code === 34 || code === 38 || code === 39 || code === 41 || code === 42 || code === 44 || code === 46 || code === 58 || code === 59 || code === 60 || code === 63 || code === 93 || code === 95 || code === 126) {
       return effects.check(trail, ok, pathAtPunctuation)(code);
     }
-    if (code === null || markdownLineEndingOrSpace(code) || unicodeWhitespace(code)) {
+    if (highSurrogate(code)) {
+      return effects.check(astralPunctuation, ok, pathAtPunctuation)(code);
+    }
+    if (code === null || markdownLineEndingOrSpace(code) || unicodeWhitespace(code) || nonAsciiPunctuation(code)) {
       return ok(code);
     }
     effects.consume(code);
@@ -665,9 +672,12 @@ function tokenizeTrail(effects, ok, nok) {
     // `<` is an end.
     code === 60 ||
     // So is whitespace.
-    code === null || markdownLineEndingOrSpace(code) || unicodeWhitespace(code)) {
+    code === null || markdownLineEndingOrSpace(code) || unicodeWhitespace(code) || nonAsciiPunctuation(code)) {
       return ok(code);
     }
+    if (highSurrogate(code)) {
+      return effects.check(astralPunctuation, ok, nok)(code);
+    }
     return nok(code);
   }
 
@@ -816,6 +826,52 @@ function previousEmail(code) {
   return !(code === 47 || gfmAtext(code));
 }
 
+/**
+ * Non-ASCII punctuation or symbol, such as `。` or `，`, which ends a literal
+ * in CJK prose.
+ *
+ * @param {Code} code
+ * @returns {boolean}
+ */
+function nonAsciiPunctuation(code) {
+  return code !== null && code > 127 && unicodePunctuation(code);
+}
+
+/**
+ * Whether `code` is a UTF-16 high surrogate, which starts a character outside
+ * the BMP such as `😀`.
+ *
+ * @param {Code} code
+ * @returns {boolean}
+ */
+function highSurrogate(code) {
+  return code !== null && code >= 0xd800 && code <= 0xdbff;
+}
+
+/**
+ * Check whether a surrogate pair is punctuation or a symbol.
+ *
+ * @this {TokenizeContext}
+ * @type {Tokenizer}
+ */
+function tokenizeAstralPunctuation(effects, ok, nok) {
+  /** @type {number} */
+  let high;
+  return start;
+
+  /** @type {State} */
+  function start(code) {
+    high = /** @type {number} */code;
+    effects.consume(code);
+    return low;
+  }
+
+  /** @type {State} */
+  function low(code) {
+    return code !== null && code >= 0xdc00 && code <= 0xdfff && /[\p{P}\p{S}]/u.test(String.fromCharCode(high, code)) ? ok(code) : nok(code);
+  }
+}
+
 /**
  * @param {Code} code
  * @returns {boolean}