On 6/24/2013 12:56 PM, Brian Burkhalter wrote:
This issue
http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6178739
exposes what could be either a minor specification or implementation flaw
depending on one's interpretation. The test case is
public class TestFormat {
public static void main(String argv[]) {
System.out.printf("%1.4f\n", 56789.456789);
System.out.printf("%0.4f\n", 56789.456789); //
java.util.MissingFormatWidthException
}
}
The format specification applicable to the "%f" conversion is
%[argument_index$][flag][width][.precision]conversion
The problem comes in because, according to the specification,
1) zero (0) is a legal flag which indicates that the output should be
zero-padded, and
2) width is a non-negative value indicating the minimum number of characters to
be printed.
The outcome is that the format string which provokes the exception above could be seen to
be ambiguous: it could be interpreted to mean either that the output is zero-padded or
that the minimum number of characters to print is zero. If one however uses the format
string "%00.4f\n" to indicate zero padding and zero width it provokes a
java.util.DuplicateFormatFlagsException.
By way of comparison, this C code
printf("%1.4f\n", 56789.456789F);
printf("%0.4f\n", 56789.456789F);
prints this output
56789.4570
56789.4570
and providing the format string "%00.4f" is a compilation error.
This result appears to assume a default width of zero for the zero-padded case
which seems to be reasonable behavior, but it is precluded for the Formatter by
the javadoc for the zero-padding flag:
"Requires the output to be padded with leading zeros to the minimum field width
following any sign or radix indicator except when converting NaN or infinity. If the
width is not provided, then a MissingFormatWidthException will be thrown."
Comments?
As a point of comparison, how does the C specification of printf and how
do C implementations handle this combination?
-Joe
