In java,
There are integral types(char
/short
/int
/long
/byte
)
There are floating types(f
The JLS lists
19 specific conversions on primitive types are called the widening primitive conversions:
Everything else needs an explicit cast. Narrowing is a little more complex:
double
to float
uses standard IEEE 754 rounding.(byte)0xfff == (byte)-1
;long
, the value is converted by rounding towards zero.long
, the value is first converted to int
by rounding towards zero. Then the resulting int
is converted to the target type using integer conversion.Examples:
int doubleToInt = (int)aDoubleValue;
yields Integer.MAX_VALUE
as per rounding rules.
byte doubleToByte = (byte)aDoubleValue;
first converts to int
, yielding Integer.MAX_VALUE
and then converts that to byte
. Integer.MAX_VALUE
is 0x7fffffff
, hence the byte value 0xff
which is -1
.
short doubleToShort = (short)aDoubleValue;
same again: converts to int
, yielding Integer.MAX_VALUE
. 0x7fffffff
to short
yields 0xffff
, i.e. -1
.
The tricky thing is actually the to-char
conversion. char
is a single, 16-bit unicode character, hence char doubleToChar = (char)aDoubleValue
gives you '\uffff'
by the now familiar rules.
As can be seen there is a difference between floating point and integer narrowing operations. The floating point operations do actual rounding, while the integer operations perform bitwise clamping.
The integer semantics are probably inherited from C. At least the first step of the float-to-integral narrowing ops are also what you expected. The second narrowing steps, from double/float to short, byte and char may seem a little surprising, but if you really cast float to short, you should probably double check that you know what you are doing anyway.