Issue #1732 - Implement JSON superset proposal.

This change makes unicode line and paragraph separators allowed in JS strings,
removing the only quirk of the JS string definition that made JSON not a valid
subset of JS.

Resolves #1732

js/src/builtin/Eval.cpp

@@ -141,35 +141,29 @@ template <typename CharT>
 static bool
 EvalStringMightBeJSON(const mozilla::Range<const CharT> chars)
 {
-    // If the eval string starts with '(' or '[' and ends with ')' or ']', it may be JSON.
-    // Try the JSON parser first because it's much faster.  If the eval string
-    // isn't JSON, JSON parsing will probably fail quickly, so little time
-    // will be lost.
+    // If the eval string starts with '(' or '[' and ends with ')' or ']', it
+    // may be JSON. Try the JSON parser first because it's much faster.  If
+    // the eval string isn't JSON, JSON parsing will probably fail quickly, so
+    // little time will be lost.
     size_t length = chars.length();
-    if (length > 2 &&
-        ((chars[0] == '[' && chars[length - 1] == ']') ||
-         (chars[0] == '(' && chars[length - 1] == ')')))
-    {
-        // Remarkably, JavaScript syntax is not a superset of JSON syntax:
-        // strings in JavaScript cannot contain the Unicode line and paragraph
-        // terminator characters U+2028 and U+2029, but strings in JSON can.
-        // Rather than force the JSON parser to handle this quirk when used by
-        // eval, we simply don't use the JSON parser when either character
-        // appears in the provided string.  See bug 657367.
-        if (sizeof(CharT) > 1) {
-            for (RangedPtr<const CharT> cp = chars.begin() + 1, end = chars.end() - 1;
-                 cp < end;
-                 cp++)
-            {
-                char16_t c = *cp;
-                if (c == 0x2028 || c == 0x2029)
-                    return false;
-            }
-        }
+    if (length < 2)
+        return false;
 
-        return true;
-    }
-    return false;
+    // It used to be that strings in JavaScript forbid U+2028 LINE SEPARATOR
+    // and U+2029 PARAGRAPH SEPARATOR, so something like
+    //
+    //   eval("['" + "\u2028" + "']");
+    //
+    // i.e. an array containing a string with a line separator in it, *would*
+    // be JSON but *would not* be valid JavaScript.  Handing such a string to
+    // the JSON parser would then fail to recognize a syntax error.  As of
+    // <https://tc39.github.io/proposal-json-superset/> JavaScript strings may
+    // contain these two code points, so it's safe to JSON-parse eval strings
+    // that contain them.
+
+    CharT first = chars[0], last = chars[length - 1];
+    return (first == '[' && last == ']') ||
+           (first == '(' && last == ')');
 }
 
 template <typename CharT>

js/src/frontend/TokenStream.cpp

@@ -603,6 +603,10 @@ TokenStream::TokenBuf::findEOLMax(size_t start, size_t max)
         if (n >= max)
             break;
         n++;
+        
+        // This stops at U+2028 LINE SEPARATOR or U+2029 PARAGRAPH SEPARATOR in
+        // string and template literals.  These code points do affect line and
+        // column coordinates, even as they encode their literal values.
         if (TokenBuf::isRawEOLChar(*p++))
             break;
     }
@@ -2129,8 +2133,9 @@ TokenStream::getStringOrTemplateToken(int untilChar, Token** tp)
                 }
                 break;
             }
-        } else if (TokenBuf::isRawEOLChar(c)) {
+        } else if (c == '\r' || c == '\n') {
             if (!parsingTemplate) {
+                // String literals don't allow ASCII line breaks.
                 ungetCharIgnoreEOL(c);
                 error(JSMSG_UNTERMINATED_STRING);
                 return false;
@@ -2138,10 +2143,18 @@ TokenStream::getStringOrTemplateToken(int untilChar, Token** tp)
             if (c == '\r') {
                 c = '\n';
                 if (userbuf.peekRawChar() == '\n')
+                    // Treat CRLF as a single line break.
                     skipCharsIgnoreEOL(1);
             }
             updateLineInfoForEOL();
             updateFlagsForEOL();
+        } else if (c == LINE_SEPARATOR || c == PARA_SEPARATOR) {
+            // U+2028 LINE SEPARATOR and U+2029 PARAGRAPH SEPARATOR encode
+            // their literal values in template literals and (as of the 
+            // JSON superset proposal) string literals, but they still count
+            // as line terminators when computing line/column coordinates.
+            updateLineInfoForEOL();
+            updateFlagsForEOL();
         } else if (parsingTemplate && c == '$') {
             if ((nc = getCharIgnoreEOL()) == '{')
                 break;

js/src/jit-test/tests/latin1/json.js

@@ -55,13 +55,10 @@ function testEvalHackNotJSON() {
     arr = eval("[]; var z; [1, 2, 3, \"abc\u1200\"]");
     assertEq(JSON.stringify(arr), '[1,2,3,"abc\u1200"]');
 
-    try {
-	eval("[1, 2, 3, \"abc\u2028\"]");
-	throw new Error("U+2028 shouldn't eval");
-    } catch (e) {
-	assertEq(e instanceof SyntaxError, true,
-		 "should have thrown a SyntaxError, instead got " + e);
-    }
+    // JSON superset
+    var arr = eval("[1, 2, 3, \"abc\u2028\"]");
+    assertEq(arr.length, 4);
+    assertEq(arr[3], "abc\u2028");
 }
 testEvalHackNotJSON();
 

js/src/tests/ecma_5/eval/line-terminator-paragraph-terminator.js

@@ -3,7 +3,9 @@
 
 //-----------------------------------------------------------------------------
 var BUGNUMBER = 657367;
-var summary = "eval must not parse strings containing U+2028 or U+2029";
+var summary =
+  "eval via the JSON parser should parse strings containing U+2028/U+2029 " +
+  "(as of <https://tc39.github.io/proposal-json-superset/>, that is)";
 
 print(BUGNUMBER + ": " + summary);
 
@@ -11,59 +13,8 @@ print(BUGNUMBER + ": " + summary);
  * BEGIN TEST *
  **************/
 
-function esc(s)
-{
-  return s.split("").map(function(v)
-                         {
-                           var code =
-                             ("000" + v.charCodeAt(0).toString(16)).slice(-4);
-                           return "\\u" + code;
-                         }).join("");
-}
-
-try
-{
-  var r = eval('"\u2028"');
-  throw new Error("\"\\u2028\" didn't throw, returned " + esc(r));
-}
-catch (e)
-{
-  assertEq(e instanceof SyntaxError, true,
-           "U+2028 is not a valid string character");
-}
-
-try
-{
-  var r = eval('("\u2028")');
-  throw new Error("(\"\\u2028\") didn't throw, returned " + esc(r));
-}
-catch (e)
-{
-  assertEq(e instanceof SyntaxError, true,
-           "U+2028 is not a valid string character");
-}
-
-try
-{
-  var r = eval('"\u2029"');
-  throw new Error("\"\\u2029\" didn't throw, returned " + esc(r));
-}
-catch (e)
-{
-  assertEq(e instanceof SyntaxError, true,
-           "U+2029 is not a valid string character");
-}
-
-try
-{
-  var r = eval('("\u2029")');
-  throw new Error("(\"\\u2029\") didn't throw, returned " + esc(r));
-}
-catch (e)
-{
-  assertEq(e instanceof SyntaxError, true,
-           "U+2029 is not a valid string character");
-}
+assertEq(eval('("\u2028")'), "\u2028");
+assertEq(eval('("\u2029")'), "\u2029");
 
 /******************************************************************************/