public class Main
{ public static void main( final java.lang.String[] args )
 { for( double v = java.lang.Math.PI * 1e4; v >= 1e-5; v /= 10 )
   java.lang.System.out.printf( "%.2g%n", v ); }}

3,1e+04
3,1e+03
3,1e+02
31
3,1
0,31
0,031
0,0031
0,00031
3,1e-05

 The documentation for »%g« on

http://download.java.net/jdk7/docs/api/java/util/Formatter.html#syntax

 seems to be insufficient:

     »'g', 'G'

     floating point

     The result is formatted using computerized scientific
     notation or decimal format, depending on the precision and
     the value after rounding.«

 The documentation for »%g« in

ISO/IEC 9899:1999 (E)

 is more detailed:

     »g,G

     A double argument representing a floating-point number is
     converted in style f or e (or in style F or E in the case
     of a G conversion specifier), with the precision
     specifying the number of significant digits. If the
     precision is zero, it is taken as 1.

     The style used depends on the value converted; style e (or
     E) is used only if the exponent resulting from such a
     conversion is less than -4 or greater than or equal to the
     precision.«

 Either Sun Microsystems, Inc. expects the programmer to look
 into ISO/IEC 9899:1999 (E) for »%g«, or there is a more
 detailed description by Sun Microsystems, Inc., which I just
 have not found yet.