As a general comment, I find a seven-digit bug id in isolation
extremely uninformative in terms of letting me know whether or not an
issue is of interest to me. For the messages I send to the list, I
always try to include the synopsis of the bugs in question in at least
one of the subject line and the body of the message.
-Joe
On 1/20/2012 10:19 AM, Brandon Passanisi wrote:
Resending again...
Hello core-libs. I was wondering of somebody could be please review
the following fix for #6469160 and #7088271. The changes in the
webrev fix both bugs. Information is below:
Webrev URL: http://cr.openjdk.java.net/~bpassani/6469160_7088271/1/
<http://cr.openjdk.java.net/%7Ebpassani/6469160_7088271/1/>
Bug #6469160:
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6469160
Bug #7088271:
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7088271
Both bugs uncover the current behavior where using a 0 or 1 precision
value with a float zero causes an ArrayIndexOutOfBoundsException. The
current code in FormattedFloatingDecimal.java interprets the float
zero as "0.0" in the case where precision is 0 or 1 and returns the
length of it's characters as 3. Later in Formatter.addZeros(), the
character array "0.0" is passed in, but a new array of only 1
character is allocated. When an System.arraycopy() is performed, the
ArrayIndexOutOfBoundsException occurs. In fact, when run with "-esa"
an AssertionError occurs at "assert (outPrec <= prec);" on line 3393
of Formatter.java. The fix is for FormatedFloatingDecimal.java to
interpret the float zero as a single "0" because of the precision
being set to 0 or 1.
Since java has been throwing exceptions in these cases, I consulted
with the output of C's printf to make sure that the outputted strings
are the same. I updated the Formatter's Basic-X template of tests
with a little over 20 test format strings that were causing exceptions
without the change and the output of each is compared with the output
from C's printf with the same format string. And, I ran all of the
Basic-X tests to make sure there weren't any regressions in behavior.
Thanks.
