Add some support for the code-point code-unit distinction.

sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart

Index: sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart
diff --git a/sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart b/sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart
index ad7b2b055f31b680a7b176a195af55da1fb5f454..5e7575bd1a987f9def335be6cb086396d34699da 100644
--- a/sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart
+++ b/sdk/lib/_internal/compiler/implementation/js_backend/constant_emitter.dart
@@ -84,6 +84,26 @@ class ConstantEmitter implements ConstantVisitor {
     buffer.add("false");
   }
 
+  void addHexEscape(CodeBuffer buffer, int code) {
+    assert(code < 0x10000);
+    if (code < 0x20) {
+      buffer.add(r'\x');
+      if (code < 0x10) buffer.add('0');
+      buffer.add(code.toRadixString(16));
+    } else {
+      if (code < 0x100) {
+        assert(code >= 0x80);
+        buffer.add(r'\x');
+      } else {
+        buffer.add(r'\u');
+        if (code < 0x1000) {
+          buffer.add('0');
+        }
+      }
+      buffer.add(code.toRadixString(16));
+    }
+  }
+
   /**
    * Write the contents of the quoted string to a [CodeBuffer] in
    * a form that is valid as JavaScript string literal content.
@@ -104,34 +124,25 @@ class ConstantEmitter implements ConstantVisitor {
       } else if (identical(code, $LS)) {
         // This Unicode line terminator and $PS are invalid in JS string
         // literals.
-        buffer.add(r'\u2028');
+        addHexEscape(buffer, 0x2028);
       } else if (identical(code, $PS)) {
-        buffer.add(r'\u2029');
+        addHexEscape(buffer, 0x2029);
       } else if (identical(code, $BACKSLASH)) {
         buffer.add(r'\\');
       } else {
-        if (code > 0xffff) {
-          compiler.reportError(
-              diagnosticNode,
-              'Unhandled non-BMP character: U+${code.toRadixString(16)}');
-        }
-        // TODO(lrn): Consider whether all codes above 0x7f really need to
-        // be escaped. We build a Dart string here, so it should be a literal
-        // stage that converts it to, e.g., UTF-8 for a JS interpreter.
         if (code < 0x20) {
-          buffer.add(r'\x');
-          if (code < 0x10) buffer.add('0');
-          buffer.add(code.toRadixString(16));
+          addHexEscape(buffer, code);
         } else if (code >= 0x80) {
-          if (code < 0x100) {
-            buffer.add(r'\x');
+          // TODO(lrn): Consider whether all codes above 0x7f really need to
+          // be escaped. We build a Dart string here, so it should be a literal
+          // stage that converts it to, e.g., UTF-8 for a JS interpreter.
+          if (code >= 0x10000) {

[floitsch, 2012/11/08 15:28:21]

Why not move the surrogate-code into addHexEscape?

[erikcorry, 2012/11/15 13:28:25]

This part has all been rewritten with the new JSON outputting code.

+            // Surrogate pair.
+            addHexEscape(buffer, 0xd800 + (((code - 0x10000) >> 10) & 0x3ff));
+            addHexEscape(buffer, 0xdc00 + (code & 0x3ff));
           } else {
-            buffer.add(r'\u');
-            if (code < 0x1000) {
-              buffer.add('0');
-            }
+            addHexEscape(buffer, code);
           }
-          buffer.add(code.toRadixString(16));
         } else {
           buffer.addCharCode(code);
         }