I'm preparing a fix which restricts the range of BigInteger so that
bitLength() always
return correct result.
I can't choose between two variants of range of valid BigInteger.
A) Two's complement range
[-2^Integer.MAX_VALUE, 2^Integer.MAX_VALUE-1]
This is exactly those numbers with bitLength() <= Integer.MAX_VALUE.
B) Symmetrical range
[-2^Integer.MAX_VALUE+1, 2^Integer.MAX_VALUE-1]
In variant A)
- x.neg() and x.abs() will throw ArithmeticException on minimum
BigInteger value, because the exact result is out of range;
- bit operations or(), and(), xor(), andNot() will always return valid
result.
In variant B)
- x.neg() and x.abs() will never throw ArithmeticException;
- and(), andNot(), xor() may throw ArithmeticException for certain
arguments.
Which variant do you prefer ?
-Dima
/
/>>>>/ d) All BigInteger constructors ensure that the bit length is no
/>>>>/ larger than 2^31 - 1;
/>>>/
/>>>/ d) is arguably the most correct approach to address the problem.
/>>>/ However, I think the practical consequences of this flaw are low.
/>>/ Nevertheless, may I use this unimportant flaw as a lesson of Java bug
/>>/ fixing ?
/>>/
/>
No fix for the bug has yet been offered.
You're independently rediscovered a bug that has been in the platform
for about 12 years.
-Joe
