What is the difference between -1 and ~0

Question

The title really says it all: what is the difference between minus one and tilda (ones-complement) zero?

The question came up during a discussion of the best way to specify a bit mask in which all bits are set. Which of the following is better?

int func(int value, int mask = -1) {
    return (value & mask);
}

or

int func(int value, int mask = ~0) {
    return (value & mask);
}

Are there any other uses where it would be the other way around?

Update: There has been a similar discussion on this topic over at stackoverflow.com/q/809227/34509 which I missed during my prior research. Thanks to Johannes Schaub for pointing it out.

Admittedly I was thinking C++ while writing the examples, but the question originally came up while using java. — zennehoy, Jul 18 '11 at 14:18
If by "best" you mean "most readable", then they're both awful. What's wrong with 0xFFFFFFFF? — Edward Thomson, Jul 18 '11 at 14:21
Ok, I guess I should have avoided default arguments for the example to work in java. But the same concept should apply for java too, right? — zennehoy, Jul 18 '11 at 14:22
0xFFFFFFFF assumes that ints are 32 bits wide, which is not necessarily the case. — zennehoy, Jul 18 '11 at 14:23
@Edward: I would agree... though of course you are then dependent on the size of `int` to determine the number of `F`... — Matthieu M., Jul 18 '11 at 14:23
0xFFFFFFF also assumes that the reader can recognise the difference between 8 and 7 repetitions of a character at a glance, to confirm that the value they're reading really is what they're expecting to see. So not all that readable, for me, even if we're assuming a 32 bit int. My version would be `unsigned int mask = -1`. — Steve Jessop, Jul 18 '11 at 17:41
@Johannes: Thanks for that link, a very interesting read! I'm surprised it didn't come up while searching for an existing answer to this topic - I was confounded that this hadn't been asked before. — zennehoy, Jul 19 '11 at 14:55

score 8 · Accepted Answer · answered Jul 18 '11 at 14:16

8

The first variant relies on 2's complement representation of negative numbers, which isn't necessarily used. 1's complement can be used too... or other encoding. My vote is for the second approach

answered Jul 18 '11 at 14:16

Armen Tsirunyan

130,161
59
324
434

Ok, that makes sense. Though honestly, what architecture still uses anything other than 2's complement? And I believe java actually guarantees it? – zennehoy Jul 18 '11 at 14:26
The JVM does guarantee the int type to be a 32 bit two's complement number. – Edward Thomson Jul 18 '11 at 16:48
I think that both rely on two's complement (if you use `int`). We had this question previously http://stackoverflow.com/q/809227/34509 . In particular, the `~0` could end up being the value `0`, in which case it could end up being all bits zero when stored. Only on two's complement, the resulting bit pattern of `~0` will identify a unique value (-1) instead of a single value having two possible representations (negative/positive zero). So I think the only portable way is to use `unsigned int` and assign `-1` to it. – Johannes Schaub - litb Jul 19 '11 at 14:11

score 3 · Answer 2 · answered Jul 18 '11 at 14:15

3

The second example is more clear as to what you're trying to test for.

answered Jul 18 '11 at 14:15

Bryan Boettcher

4,412
1
28
49

score 1 · Answer 3 · answered Jul 18 '11 at 14:17

1

Both are same. Except that, -1 doesn go well with unsigned int without warning.

answered Jul 18 '11 at 14:17

iammilind

68,093
33
169
336

What is the difference between -1 and ~0

3 Answers3

Linked