jkesselm 02/02/05 06:19:55
Modified: java/src/org/apache/xpath/objects XStringForFSB.java
Log:
Bugzilla 5346: Responding to the re-open; more direct and reliable
edge-case test. This does lose a few low bits on many-digit values,
so it is not the best long-term solution.
Revision Changes Path
1.12 +12 -8
xml-xalan/java/src/org/apache/xpath/objects/XStringForFSB.java
Index: XStringForFSB.java
===================================================================
RCS file:
/home/cvs/xml-xalan/java/src/org/apache/xpath/objects/XStringForFSB.java,v
retrieving revision 1.11
retrieving revision 1.12
diff -u -r1.11 -r1.12
--- XStringForFSB.java 4 Feb 2002 19:03:41 -0000 1.11
+++ XStringForFSB.java 5 Feb 2002 14:19:55 -0000 1.12
@@ -68,6 +68,7 @@
*/
public class XStringForFSB extends XString
{
+ static long MAX_NO_OVERFLOW_RISK=(Long.MAX_VALUE-'9')/10;
/** The start position in the fsb. */
int m_start;
@@ -1034,6 +1035,9 @@
break;
case '0': // NOT Unicode isDigit(); ASCII digits
_only_
+ if(digitType==0 && digitsFound[0]==0) // Ignore leading zeros
+ break;
+ // Else fall through into normal digit handling.
case '1':
case '2':
case '3':
@@ -1046,17 +1050,17 @@
// We have a potential overflow issue that we need
// to guard against. See Bugzilla 5346.
//
- // %REVIEW% Is it possible that we should be accepting _some_
- // of the bits of the new digit, but not all of them?
- long newResult = longResult * 10 + (c - '0');
- if(newResult>=0)
+ // %REVIEW% MUST BE RECONSIDERED since this may not be
+ // fully compliant with the IEEE floating-point spec. (On the
+ // other hand, it's wicked fast compared to the usual solution,
+ // and works fine for all but a very few edge cases.)
+ if(longResult>MAX_NO_OVERFLOW_RISK)
+ ++overflow;
+ else
{
- longResult=newResult;
+ longResult=longResult*10 + (c - '0');
++digitsFound[digitType]; // Remember scaling
}
- else
- ++overflow;
-
break;
default:
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
