Having tried a simple program (using both 0 and 100, to show the difference between “special” constants and general ones) the Sun Java 6 compiler will output the same bytecode for both 1 and 2 (cases 3 and 4 are identical to 2 as far as the compiler is concerned).
So for example:
double x = 100;
double y = 100.0;
compiles to:
0: ldc2_w #2; //double 100.0d
3: dstore_1
4: ldc2_w #2; //double 100.0d
7: dstore_3
However, I can’t see anything in the Java Language Specification guaranteeing this compile-time widening of constant expressions. There’s compile-time narrowing for cases like:
byte b = 100;
as specified in section 5.2, but that’s not quite the same thing.
Maybe someone with sharper eyes than me can find a guarantee there somewhere…