Complete mystery with bitwise operators...

jpooton

I'm hoping for some ideas here, as I'm feeling truely stumped at this point...

In a nutshell I have 4 PHP enabled web servers. One of these is a new machine running Fedora Core 3 and PHP 4.3.10. (The others are Fedora Core 2 and PHP 4.3.10). I'm seeing different output from the new FC3 machine for bitwise operations, and have no idea why. For example:

echo (-3281063054 & 3);

gives "2" on all three of my FC2 machines.

it gives "0" on my new FC3 machine.

I'm looking for any suggestions on how to proceed. Any further tests I can do to help debug this? Is anyone running FC3 that could test my statement above on their setup? Any help would be greatly appreciated...

drew010

seeing as php mostly uses your os's local c files see if its a php problem or an fc3 problem

#include <stdio.h>

int main() {
  printf("%d", -3281063054 & 3);

  return 0;
}

try to compile that with gcc bit.c -o bit

see what output you get on both machines when you run it. i did get 2 on my machine, but its redhat 9.

i was thinking originally it was a problem with the - sign on that number, but regardless, the answer is still the same be it positive or negative.

when i compiled i got a message
bit.c:4: warning: decimal constant is so large that it is unsigned

hopefully this will shed some light.

jpooton

Originally posted by drew010
seeing as php mostly uses your os's local c files see if its a php problem or an fc3 problem

when i compiled i got a message
bit.c:4: warning: decimal constant is so large that it is unsigned

hopefully this will shed some light.

Thank you for your reply! I just ran your test here. When compiling I got:

 [root@thunder php-4.3.11]# gcc james-bittest.c -o james-bittest
james-bittest.c: In function `main':
james-bittest.c:4: warning: this decimal constant is unsigned only in ISO C90

When running the executable it returned a "2". So I guess this is starting to point the finger more at PHP? :bemused:

drew010

yeah thats pretty strange, i checked it just in case with php but i got 2 as well. i looked on the manual page and saw no interesting warnings about it.
you may want to go to bugs.php.net and search for something similar and if you find nothing submit a new ticket.

jpooton

Thanks.. you read my mind. I've been browsing all the bitwise bug reports there and it's sort of facinating. There are lots of reports of problems with bitwise operations, but they all get marked at "bogus" or "closed" with some sort of reference that it will never be fixed and use GMP instead 😉

One bug report had this for an answer:

"-2147483649 is to big for a PHP int so it is actually treated as a
float, and the behavior of & on floats is undefined"

and while I do see my value is seen as a float, it works properly on all my other PHP machines, so something has changed.

In any event, I tried the GMP function and that gives the correct answer. Only problem is that I'd have to rewrite this entire encryption library that's already working on all the other machines. (at that point it might be faster to just nuke the box and go back to one of those established working configs).

Anyways, thanks for your help.

drew010

yeah thats pretty interesting. maybe the change has to do with how values are passed back and forth between the php script and the zend engine which is powered by c.
My thought is if you are the only one using the library you may as well try to go back to fc2 since we know it works there, but if you distribute the library, its probably safer and more reliable to use the GMP functions since you wont know who will be running what configurations.

Good luck