Json.parse decodes \uXXXX to UTF-8 - a surrogate pair joined, a lone surrogate or bad hex as U+FFFD - and \t \r \b \f, where it dropped the backslash and kept the hex as text; examples/lang/json_unicode checks it byte by byte

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

runtime/native/value.ludic

@@ -616,6 +616,16 @@ function jp_string_esc(p: JP) -> string {
       if p.i < p.n {
         let e = p.s[p.i]
         if e == 'n' { out += "\n" }
+        else if e == 't' { out += "\t" }
+        else if e == 'r' { out += "\r" }
+        else if e == 'b' or e == 'f' or e == 'u' {
+          var code = 8
+          if e == 'f' { code = 12 }
+          if e == 'u' { code = jp_code(p) }            # leaves p.i on the escape's last hex digit
+          let u = jp_utf8(code)
+          out += u
+          free(u)
+        }
         else { out += p.s[p.i..p.i + 1] }      # \" \\ \/ -> the literal char
         p.i += 1
       }
@@ -627,6 +637,53 @@ function jp_string_esc(p: JP) -> string {
   return out
 }
 
+# \uXXXX, p.i on the 'u': the code point, a surrogate pair joined (\ud83d\ude00 is one emoji); a
+# lone or broken surrogate, or bad hex, is U+FFFD. Python's json.dump writes non-ASCII this way.
+function jp_code(p: JP) -> int {
+  let hi = jp_hex4(p, p.i + 1)
+  if hi < 0 { return 65533 }                        # p.i stays on the 'u': what follows is read as text
+  p.i += 4
+  if hi >= 55296 and hi <= 56319 {                  # a high surrogate: a low one must follow
+    if p.i + 6 < p.n and p.s[p.i + 1] == '\\' and p.s[p.i + 2] == 'u' {
+      let lo = jp_hex4(p, p.i + 3)
+      if lo >= 56320 and lo <= 57343 {
+        p.i += 6
+        return 65536 + ((hi - 55296) << 10) + (lo - 56320)
+      }
+    }
+    return 65533
+  }
+  if hi >= 56320 and hi <= 57343 { return 65533 }  # a low surrogate on its own
+  return hi
+}
+# four hex digits at `at`, or -1
+function jp_hex4(p: JP, at: int) -> int {
+  if at + 4 > p.n { return -1 }
+  var v = 0
+  for k in 0 .. 4 {
+    let c = p.s[at + k]
+    var d = -1
+    if c >= '0' and c <= '9' { d = c - 48 }
+    else if c >= 'a' and c <= 'f' { d = c - 87 }
+    else if c >= 'A' and c <= 'F' { d = c - 55 }
+    if d < 0 { return -1 }
+    v = v * 16 + d
+  }
+  return v
+}
+# one code point as a UTF-8 string
+function jp_utf8(c: int) -> string {
+  let out = bytes(5)
+  var n = 0
+  if c < 128 { out[0] = c;  n = 1 }
+  else if c < 2048 { out[0] = 192 | (c >> 6);  out[1] = 128 | (c & 63);  n = 2 }
+  else if c < 65536 { out[0] = 224 | (c >> 12);  out[1] = 128 | ((c >> 6) & 63);  out[2] = 128 | (c & 63);  n = 3 }
+  else { out[0] = 240 | (c >> 18);  out[1] = 128 | ((c >> 12) & 63);  out[2] = 128 | ((c >> 6) & 63);  out[3] = 128 | (c & 63);  n = 4 }
+  out[n] = 0
+  let t: string = out
+  return t
+}
+
 # read a number; a '.' makes it a fixed node, otherwise an int node.
 function jp_number(p: JP) -> Val {
   var neg = 0