Recognize non-BMP punctuation & symbols (#297)

Closes #296.

lib/inlines.js

@@ -127,9 +127,10 @@ var match = function(re) {
 
 // Returns the code for the character at the current subject position, or -1
 // there are no more characters.
+// This function must be non-BMP aware because the Unicode category of its result is used.
 var peek = function() {
     if (this.pos < this.subject.length) {
-        return this.subject.charCodeAt(this.pos);
+        return this.subject.codePointAt(this.pos);
     } else {
         return -1;
     }
@@ -270,7 +271,7 @@ var scanDelims = function(cc) {
         return null;
     }
 
-    char_before = startpos === 0 ? "\n" : this.subject.charAt(startpos - 1);
+    char_before = previousChar(this.subject, startpos);
 
     cc_after = this.peek();
     if (cc_after === -1) {
@@ -304,6 +305,25 @@ var scanDelims = function(cc) {
     }
     this.pos = startpos;
     return { numdelims: numdelims, can_open: can_open, can_close: can_close };
+
+    function previousChar(str, pos) {
+        if (pos === 0) {
+            return "\n";
+        }
+        var previous_cc = str.charCodeAt(pos - 1);
+        // not low surrogate (BMP)
+        if ((previous_cc & 0xfc00) !== 0xdc00) {
+            return str.charAt(pos - 1);
+        }
+        // returns NaN if out of range
+        var two_previous_cc = str.charCodeAt(pos - 2);
+        // NaN & 0xfc00 = 0
+        // checks if 2 previous char is high surrogate
+        if ((two_previous_cc & 0xfc00) !== 0xd800) {
+            return previous_char;
+        }
+        return str.slice(pos - 2, pos);
+    }
 };
 
 // Handle a delimiter marker for emphasis or a quote.

test/regression.txt

@@ -546,3 +546,19 @@ foo <!-- test --> more -->
 <p>　全角スペース (U+3000) 全形空白　</p>
 <p>ZWNBSP (U+FEFF) ZWNBSP</p>
 ````````````````````````````````
+
+#296
+```````````````````````````````` example
+a**a∇**a
+
+a**∇a**a
+
+a**a𝜵**a
+
+a**𝜵a**a
+.
+<p>a**a∇**a</p>
+<p>a**∇a**a</p>
+<p>a**a𝜵**a</p>
+<p>a**𝜵a**a</p>
+````````````````````````````````