add UseCounters for NonOctalDecimalIntegerLiteral in strict mode

src/parsing/scanner.cc

Index: src/parsing/scanner.cc
diff --git a/src/parsing/scanner.cc b/src/parsing/scanner.cc
index faec88b8a4bfb5ff37bf8dad58fefc32cd8ff03f..e8e77b9ed6956f63822015c35c979a4afe646610 100644
--- a/src/parsing/scanner.cc
+++ b/src/parsing/scanner.cc
@@ -976,9 +976,11 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
   DCHECK(IsDecimalDigit(c0_));  // the first digit of the number or the fraction
 
   enum { DECIMAL, HEX, OCTAL, IMPLICIT_OCTAL, BINARY } kind = DECIMAL;
+  bool decimal_that_looks_like_octal = false;
 
   LiteralScope literal(this);
   bool at_start = !seen_period;
+  int start_pos = source_pos();  // For reporting octal positions.
   if (seen_period) {
     // we have already seen a decimal point of the float
     AddLiteralChar('.');
@@ -987,7 +989,6 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
   } else {
     // if the first character is '0' we must check for octals and hex
     if (c0_ == '0') {
-      int start_pos = source_pos();  // For reporting octal positions.

[caitp (gmail), 2016/05/05 18:25:02]

This change seems not to be needed, since it's irrelevant when the number
doesn't begin with 0

[jwolfe, 2016/05/05 18:42:02]

I can leave the local variable uninitialized until this line. I just need to
declare it out where the later code can see it.

I would have loved to have dealt with `octal_pos_` inside the `c0_ == '0'`
block, but I didn't know where the end of the token was going to be yet.

       AddLiteralCharAdvance();
 
       // either 0, 0exxx, 0Exxx, 0.xxx, a hex number, a binary number or
@@ -1030,6 +1031,7 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
           if (c0_ == '8' || c0_ == '9') {
             at_start = false;
             kind = DECIMAL;
+            decimal_that_looks_like_octal = true;

[caitp (gmail), 2016/05/05 18:21:08]

'8' and '9' are decidedly not octal digits --- why are we saying this looks
octal? Just confused about what problem this solves

             break;
           }
           if (c0_  < '0' || '7'  < c0_) {
@@ -1039,6 +1041,8 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
           }
           AddLiteralCharAdvance();
         }
+      } else if (c0_ == '8' || c0_ == '9') {
+        decimal_that_looks_like_octal = true;
       }
     }
 
@@ -1060,6 +1064,8 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
           literal.Complete();
           HandleLeadSurrogate();
 
+          if (decimal_that_looks_like_octal)

[caitp (gmail), 2016/05/05 18:25:02]

`decimal_that_looks_like_octal` is always false unless the `c0_ == '0'` branch
is taken on line 991 --- these should be safe to get rid of, AFAIK?

[jwolfe, 2016/05/05 18:42:01]

I was surprised by needing two different checks in this function. This check is
for `09`, and the one on line 1107 is for `019`. I don't know why `019` isn't
reported as a SMI token by this function. Perhaps it's a mistake?

+            octal_pos_ = Location(start_pos, source_pos());
           return Token::SMI;
         }
         HandleLeadSurrogate();
@@ -1098,6 +1104,8 @@ Token::Value Scanner::ScanNumber(bool seen_period) {
 
   literal.Complete();
 
+  if (decimal_that_looks_like_octal)
+    octal_pos_ = Location(start_pos, source_pos());

[caitp (gmail), 2016/05/06 21:08:03]

So, if we want to have a different error message for this, we could save it as
`decimal_leading_zero_pos_` or something, and  pick which one to report in
CheckOctalLiteral()

   return Token::NUMBER;
 }