Widening conversion bugs ?

Question

I dont understand .. (from A Programmer's Guide to Java Certification 3rd Edition - mughal pg 172)
a) this work
Integer iRef3 = (short)10; // constant in range: casting by narrowing to short,
// widening to int, then boxing
b) final short x = 3;
Integer y = x;
this work but not this.
short x = 3;
Integer y = x;
The 'final' means constant expression which is used for implicit narrowing conversion. The 2nd one is not narrowing conversion so why the 2nd one didnt work?
c) Exactly as a) but instead of Integer .. i use Long. // This doesnt work
Long iRef3 = (short)10;
c) This also doesnt work unless i put the final on short
short s = 10; // narrowing (a)
Integer iRef1 = s; // short not assignable to Integer -- error here (b)
Since implicit conversion succeded for (a) and you dont need constant expression for widening conversion (b) .. why it doesnt work on (b).
I thought the variable s would be widen to int and then boxing into Integer.
Any help is appreciated
Edited by: yapkm01 on Jul 16, 2009 9:32 PM

Answer 1

A type conversion from long to float is
commonly called a widening conversion and thus legal.
I understand that the range of float variables
is wider than that of long variables.
However, a long variable can hold about 4
billion (2^64) distinct integer values. How could
these possibly all be represented by a float
variable that uses only 32 bits - and thus can only
represent 2^32 distinct values. The mantissa of a
float variable actually uses even less bits -
only 23.
So, while any value of a long variable will
fall within the range of a float variable,
occasionally approximation should occur and
information get lost.
Why is then usually said that widening conversion
doesn't lose information?Mathematically, a float's bigger. You do lose precision, though.
What's your point?