[SOLVED] Problems with NOT ~ bitwise operator

Skyer · 09-23-2011, 10:09 AM

Hello,
I am trying to figure out this little problem with C++'s NOT bitwise operator. I guess example with code will explain my situation best:

Code:

...
bool x=1,y=0;
y=~x;
cout<<"y="<<y<<"~x"<<~x<<endl;
...

Produces output like this:

Code:

y=1~x-2

I expected y to be 0, as inversion of "1" bit is "0" bit. Also, I don't clearly understand why y is different when ~x.

Thanks for your time,

Skyer

Proud · 09-23-2011, 10:24 AM

As for 1 -> -2

Quote:

For two's complement signed integers, the bitwise complement of a number is equal to the negative of that number minus 1

Skyer · 09-23-2011, 10:32 AM

Thanks for your answer.
I feel embarrassed, I already read that - just didn't see the part.

Could you please provide me with the an example on how to just "toggle bits" on and off?

Thanks,

Skyer

SigTerm · 09-23-2011, 10:35 AM

Do not treat "bool" as integer. "bool" have only two acceptable values - "true" and "false". "~" is meant to be used on integers only. For boolean variables, use "!".

Code:

bool y = true, x = true;
y = !x;
unsigned int x1 = 10, y1 = 20;
y1 = ~x1;

Skyer · 09-23-2011, 10:51 AM

Quote:

Do not treat "bool" as integer. "bool" have only two acceptable values - "true" and "false".

I thought that setting x to 1 is equivalent as declaring it true.

Quote:

"~" is meant to be used on integers only.

I was assuming I can use ~ in that manners, thank you for clearing this for me.

I am going to try it out and report back.

EDIT: Alright, it works. Thanks for help.

Skyer

SigTerm · 09-23-2011, 11:59 AM

Quote:

Originally Posted by Skyer

I thought that setting x to 1 is equivalent as declaring it true.

Not exactly. As far as I know, in this case compiler converts 1 to bool, and assigns result to variable. You can't assign "1" to boolean, since "1" is an integer.

Quote:

Originally Posted by Skyer

I was assuming I can use ~ in that manners, thank you for clearing this for me.

IF I understand it correctly, when you use ~ with bool, following happens:

bool is converted to int.
the result of ~ is calculated.
result of ~ is converted to bool

~true results in ~int(true) which is 0xfffffffe (i.e. ~1) for integer types. Which converts to "true", because it is not zero.

Code:

#include <stdio.h>

int main(int argc, char** argv){
	bool a = true, b = !a, c = ~a;
	int i = ~a;
	printf("a: %d, b:%d, c:%d, i:%d", a, b, c, i);
	return 0;
}

Please note that:

You may need to check C++ standard to check if this is a defined behavior and that it actually works the way I described (I'm not 100% sure).
Compiler may print a warning about bool to int conversion. (well, MS compiler warns you about it, g++ doesn't care)

MTK358 · 09-23-2011, 12:20 PM

Quote:

Originally Posted by SigTerm

Not exactly. As far as I know, in this case compiler converts 1 to bool, and assigns result to variable. You can't assign "1" to boolean, since "1" is an integer.

I always that that bool is just another name for an integer, and that the only real point of it is to make it more explicit that you're storing a true or false value rather than a number.

Ramurd · 09-23-2011, 03:41 PM

Quote:

Originally Posted by MTK358

I always that that bool is just another name for an integer, and that the only real point of it is to make it more explicit that you're storing a true or false value rather than a number.

And there was me thinking that bool is a char (short int?) type where 0 = false and everything else = true; That's what I recall from my kerningham & richie book.

~ is a bitwise inverter; so if 1 is 0001 then ~1 is 1110; given that MSB is used for the signedness; you get a binary -110, which is -6 in decimal in a 4-bit signed value; so the size of the type matters for inversion.

You can use inversion nicely if you want to toggle values:

Code:

#define nothing      0
#define worldexec    1
#define worldwrite   2
#define worldread    4
#define groupexec    8
#define groupwrite  16
#define groupread   32
#define userexec    64
#define userwrite  128
#define userread   256
#define all        511

int filemask=userread|userexec|userwrite;

/* toggle userexec off */
filemask=(filemask & ~userexec);

MTK358 · 09-23-2011, 03:54 PM

Quote:

Originally Posted by Ramurd

And there was me thinking that bool is a char (short int?) type where 0 = false and everything else = true; That's what I recall from my kerningham & richie book.

What do you mean? that's pretty much what I thought. I just didn't know what size of integer it is (I would assume that it's a char not to waste RAM).

Proud · 09-23-2011, 04:05 PM

Quote:

Originally Posted by Ramurd

given that MSB is used for the signedness

And have negative 0? Two's complement is common for a reason.

Binary -110 aka 1110 is then -2, as per the OP's output and my earlier quote, not -6.
In fact 1...any number of 1s...10 is always -2 in this scheme. The same goes goes for any other negative number, see the sign extension section. Yes the MSB relates to the sign, but padding with 1s not 0s means you don't just take the rest of the bits and calculate their positive representation as though MSB were 0, then negate that.
I believe one's compliment has the behaviour you describe.

Your flags example is dealing with positive/unsigned values that are all exact powers of 2 (using a single 1 in a field of 0s). This with bitwise AND doesn't seem to relate to signedness.

SigTerm · 09-23-2011, 05:02 PM

Quote:

Originally Posted by Ramurd

And there was me thinking that bool is a char (short int?) type where 0 = false and everything else = true; That's what I recall from my kerningham & richie book.

Size of boolean is implementation defined, and although it converts to int AND many types convert to bool, I haven't found anything about the way compiler should store bool, true and false values.

Regarding your example.

Code:

#define nothing      0
#define worldexec    1
#define worldwrite   2
#define worldread    4
#define groupexec    8
#define groupwrite  16
#define groupread   32
#define userexec    64
#define userwrite  128
#define userread   256

For things like these C++ has bitfields and unions.

Code:

#include <stdio.h>

struct Filemask1{
	int worldexec	:1;
	int worldwrite  :1;
	int worldread   :1;
	int groupexec   :1;
	int groupwrite  :1;
	int groupread   :1;
	int userexec    :1;
	int userwrite   :1;
	int userread    :1;
};

struct Filemask2{
	union{
		struct{
			unsigned int worldexec	 :1;
			unsigned int worldwrite  :1;
			unsigned int worldread   :1;
			unsigned int groupexec   :1;
			unsigned int groupwrite  :1;
			unsigned int groupread   :1;
			unsigned int userexec    :1;
			unsigned int userwrite   :1;
			unsigned int userread    :1;
			unsigned int rsvd        :23;
		};
		unsigned int mask;
	};
};
int main(int argc, char** argv){
	Filemask2 mask;
	mask.mask = 0;
	printf("%d\n", mask.mask);
	mask.userread = 1;
	printf("%d\n", mask.mask);	
	return 0;
}