PHP bug (or my bad code)?

[Bob Purdon]

Consider this code:

<?
$mask_n = (pow(2,32) - (pow(2,32-24)-1));
printf("%x\n", $mask_n);
?>

On PHP 4.1.2 it prints:

ffffff00

On PHP 4.3.10 is prints:

ffffff01


Both versions of PHP are the standard Debian packaged versions (4.1.2
from Woody, 4.3.10 from Sarge)

tecra# php -v
PHP 4.3.10-9 (cli) (built: Mar 6 2005 17:45:14)
Copyright (c) 1997-2004 The PHP Group
Zend Engine v1.3.0, Copyright (c) 1998-2004 Zend Technologies

admin# php -v
4.1.2

Results are (predictably) the same under both the CLI version and under
the Apache module.


Is this a bug in PHP, or is there something broken in my code?

[Hans van Kranenburg]

Re: PHP bug (or my bad code)?

Bob Purdon wrote:[color=blue]
> Consider this code:
>
> <?
> $mask_n = (pow(2,32) - (pow(2,32-24)-1));
> printf("%x\n", $mask_n);
> ?>
>
> On PHP 4.1.2 it prints:
> ffffff00
>
> On PHP 4.3.10 is prints:
> ffffff01
>
> Both versions of PHP are the standard Debian packaged versions (4.1.2
> from Woody, 4.3.10 from Sarge)[/color]

Some other code:

<pre>
<?
$mask_n = (pow(2,32) - (pow(2,32-24)-1));
var_dump($mask_ n);
printf("%x\n", $mask_n);

$mask_n2 = 4294967041;
var_dump($mask_ n2);
printf("%x\n", $mask_n2);

echo "so... what's the difference ... " . ($mask_n - $mask_n2);

?>
</pre>

On woody 4.1.2:

float(429496704 1)
ffffff00
float(429496704 1)
ffffff01
so... what's the difference ... -3.3378601074219 E-06

On sarge 4.3.10:

float(429496704 1)
ffffff01
float(429496704 1)
ffffff01
so... what's the difference ... 0

What's going on?!

Hans van Kranenburg

--
"He who asks a question is a fool for five minutes;
he who does not ask a question remains a fool forever"

[Dana Cartwright]

Re: PHP bug (or my bad code)?

"Hans van Kranenburg" <f00b4r@xs4all. nl> wrote in message
news:4246a1d3$0 $143$e4fe514c@n ews.xs4all.nl.. .[color=blue]
> Bob Purdon wrote:[color=green]
>> Consider this code:
>>
>> <?
>> $mask_n = (pow(2,32) - (pow(2,32-24)-1));
>> printf("%x\n", $mask_n);
>> ?>
>>
>> On PHP 4.1.2 it prints:
>> ffffff00
>>
>> On PHP 4.3.10 is prints:
>> ffffff01
>>
>> Both versions of PHP are the standard Debian packaged versions (4.1.2
>> from Woody, 4.3.10 from Sarge)[/color]
>
> Some other code:
>
> <pre>
> <?
> $mask_n = (pow(2,32) - (pow(2,32-24)-1));
> var_dump($mask_ n);
> printf("%x\n", $mask_n);
>
> $mask_n2 = 4294967041;
> var_dump($mask_ n2);
> printf("%x\n", $mask_n2);
>
> echo "so... what's the difference ... " . ($mask_n - $mask_n2);
>
> ?>
> </pre>
>
> On woody 4.1.2:
>
> float(429496704 1)
> ffffff00
> float(429496704 1)
> ffffff01
> so... what's the difference ... -3.3378601074219 E-06
>
> On sarge 4.3.10:
>
> float(429496704 1)
> ffffff01
> float(429496704 1)
> ffffff01
> so... what's the difference ... 0
>
> What's going on?!
>
> Hans van Kranenburg[/color]

I imagine the relevant statement from the manual is:

"If possible, this function will return an integer"

Your expression would be OK if indeed it were calculated in integers. But
since 2 raised to the power 32 cannot be represented in 32 bits (it misses
by 1), you're thrown into the realm of floating point, and in that world,
you can't assume all bits will be as you think they will be, only that the
numeric values will be sufficiently close to the exact result. And in
floating point, the *order* in which you do arithmetic matters, quite a lot.

As to your code, the most straight-forward way to create bit masks (which is
what I'm thinking you are trying to do) is like this:

$mask = 0xFFFFFFFF << (32 - 8)

which will reliably give you:

0xFFFFFF00

Further, if this were my software, I'd simply write:

$mask = 0xFFFFFF00;

because that says, as clearly as can be said, "I want a 32-bit mask with the
high 24 bits set and the low 8 bits clear".

If I had a variable that specified the number of low zero bits I needed, I'd
write:

$mask = 0xFFFFFFFF << (32 - $low_zero_bits) ;

because that says, as clearly as can be said, "I want a mask that's 32 bits
wide and is everywhere 1's except for the $low_zero_bits which are zero".

Finally, if you want this to work cross-platform, you should define a
constant that's the number of bits in an integer, and work from that,
remembering to never shift by more than that many bits, since Intel
disagrees with most of the rest of the world as to the meaning of
over-shifting (shifting more than 32 on a 32-bit machine, for example).

[Bob Purdon]

Re: PHP bug (or my bad code)?

> As to your code, the most straight-forward way to create bit masks (which is[color=blue]
> what I'm thinking you are trying to do) is like this:
>
> $mask = 0xFFFFFFFF << (32 - 8)[/color]

Yeah, indeed a much better way.
[color=blue]
> Further, if this were my software, I'd simply write:
>
> $mask = 0xFFFFFF00;[/color]

Agree, except the mask varies, so:
[color=blue]
> $mask = 0xFFFFFFFF << (32 - $low_zero_bits) ;[/color]

Is the way it needs to go...

Seems I've been cracking this nut the hard way and overlooking the
obvious :-)

Thanks all.