Inserted string validation as separate task in compiler.

frog/leg/ssa/codegen.dart

Index: frog/leg/ssa/codegen.dart
diff --git a/frog/leg/ssa/codegen.dart b/frog/leg/ssa/codegen.dart
index 6a02fc44aea093949b1baf9a6e32d95ebc35efe4..5920a6a3e83edb06face6a85e932d0922d2896c4 100644
--- a/frog/leg/ssa/codegen.dart
+++ b/frog/leg/ssa/codegen.dart
@@ -489,6 +489,102 @@ class SsaCodeGenerator implements HVisitor {
     buffer.add(')');
   }
 
+  /**
+   * Write the contents of the quoted string to a [StringBuffer] in
+   * a form that is valid as JavaScript string literal content.
+   * The string is assumed quoted by [quote] characters.
+   */
+  static void writeEscapedString(QuotedString string,
+                                 StringBuffer buffer,
+                                 int quote,
+                                 void cancel(String reason)) {
+    bool raw = string.quoting.raw;
+    Iterator<int> iter = string.iterator();

[ahe, 2012/01/26 19:24:28]

Do you think performance would suffer if this was named "iterator" :-)

[Lasse Reichstein Nielsen, 2012/01/27 11:39:04]

Absoleffinglutely. But I'll rename it anyway, just out of spite!

+    while (iter.hasNext()) {
+      int code = iter.next();
+      if (code === quote) {
+        // We need to add a backslash before quotes, both in normal
+        // and in raw strings.
+        buffer.add(@'\');
+        buffer.add(code === $SQ ? "'" : '"');
+      } else if (code === $LF) {
+        // Newlines in strings only occour in multiline strings.

[ahe, 2012/01/26 19:24:28]

occour -> occur

[Lasse Reichstein Nielsen, 2012/01/27 11:39:04]

Done.

+        // They need to be written using escapes in JS..

[ahe, 2012/01/26 19:24:28]

Extra .

[Lasse Reichstein Nielsen, 2012/01/27 11:39:04]

Done.

+        assert(string.quoting.multiline);
+        buffer.add(@'\n');
+      } else if (code === $CR) {
+        assert(string.quoting.multiline);
+        buffer.add(@'\r');
+      } else if (code === $LS) {
+        // This Unicode line terminator and $PS are invalid in JS string
+        // literals.
+        buffer.add(@'\u2028');
+      } else if (code === $PS) {
+        buffer.add(@'\u2029');
+      } else if (code !== $BACKSLASH) {
+        buffer.add(new String.fromCharCodes([code]));
+      } else if (raw) {
+        buffer.add(@'\\');
+      } else {

[ahe, 2012/01/26 19:24:28]

For documentation, how about adding:

assert(code === $BACKSLASH); 

[Lasse Reichstein Nielsen, 2012/01/27 11:39:04]

Done.

+        code = iter.next();
+        switch (code) {
+          case $u:
+            buffer.add(@'\u');
+            code = iter.next();
+            if (code == $OPEN_CURLY_BRACKET) {
+              int value = 0;
+              code = iter.next();
+              do {
+                value = value * 16 + hexDigitValue(code);
+                code = iter.next();
+              } while (code !== $CLOSE_CURLY_BRACKET);
+              if (code > 0xffff) {
+                cancel("Unhandled non-BMP character: " +
+                       "U+${code.toRadixString(16)}");
+              }
+              for (int i = 12; i >= 0; i -= 4) {
+                buffer.add(((value >> i) & 0xf).toRadixString(16));
+              }
+            } else {
+              buffer.add(new String.fromCharCodes([code]));
+              // Remaining three hex digits will be copied verbatim.
+            }
+            break;
+          case $x:
+            buffer.add(@'\x');
+            // The two hex digits will be copied verbatim.
+            break;
+          // Character escapes that identical in meaning in JS.

[karlklose, 2012/01/27 09:11:59]

that *are* identical

[Lasse Reichstein Nielsen, 2012/01/27 11:39:04]

Good catch.

[ahe, 2012/01/27 12:12:37]

You forgot to change it :-)

+          case $b: buffer.add(@'\b'); break;
+          case $f: buffer.add(@'\f'); break;
+          case $n: buffer.add(@'\n'); break;
+          case $r: buffer.add(@'\r'); break;
+          case $t: buffer.add(@'\t'); break;
+          case $v: buffer.add(@'\v'); break;
+          // Identity escapes that must be escaped in JS strings.
+          case $BACKSLASH: buffer.add(@'\\'); break;
+          case $LF: buffer.add(@'\n'); break;
+          case $CR: buffer.add(@'\r'); break;
+          case $LS: buffer.add(@'\u2028'); break;
+          case $PS: buffer.add(@'\u2029'); break;
+          // Quotes may or may not need the escape.
+          case $SQ:
+          case $DQ:
+            // Only escape quotes if they match the generated string quotes.
+            if (code == quote) buffer.add(@'\');
+            buffer.add(code === $SQ ? "'" : '"');
+            break;
+          default:
+            // All other escaped characters are identity escapes,
+            // and don't need a backslash in JS.
+            buffer.add(new String.fromCharCodes([code]));
+            break;
+        }
+      }
+    }
+  }
+
+
   visitLiteral(HLiteral node) {
     if (node.isLiteralNull()) {
       buffer.add("(void 0)");
@@ -496,12 +592,13 @@ class SsaCodeGenerator implements HVisitor {
       buffer.add('(${node.value})');
     } else if (node.isLiteralString()) {
       QuotedString string = node.value;
-      String quote = string.quoteChar;
+      StringQuoting quoting = string.quoting;
+      String quote = quoting.quoteChar;
       buffer.add(quote);
-      string.writeEscaped(buffer, string.quoteCharCode,
-                          (String reason) {
-                               compiler.cancel(reason, instruction: node);
-                          });
+      writeEscapedString(string, buffer, quoting.quote,
+                         (String reason) {
+                           compiler.cancel(reason, instruction:node);
+                         });
       buffer.add(quote);
     } else {
       buffer.add(node.value);