Why is the number 16 converted to float(6.1026988574311E_320) by PHP using Zend_Amf

Question

The Zend_Amf specification states that a Number type returned from flash will map to a float in PHP. Fine. But why does the number 16 get returned as 6.1026988574311E_320 ? PHP version is 5.2.9 running on OS X.

I have tried forcing a cast to integer in PHP (the above value gets rounded to 0) and also from Actionscript using int(16) - the latter comes through as NULL. How can ensure that Flash returns an integer via AMF and that PHP can deal with it?

Answer 1

You have a classic endian problem. Looks like either Zend or flash is doing the wrong thing with the endianness of this double. Here is the a program that prints the double (and its hex value). It then reverses the endianness and prints it again.

#include <stdio.h>
#include <stdint.h>

int main(void)
{
  double d = 16;
  uint32_t *i = (uint32_t *)(&d);
  uint8_t *c = (uint8_t *)(&d);
  size_t j;

  printf("%08x %08x %lg\n", i[1], i[0], d);

  for(j = 0; j < sizeof(double) / 2; j++)
  {
        uint8_t tmp;

        tmp = c[j];
        c[j] = c[sizeof(double) - j - 1];
        c[sizeof(double) - j - 1] = tmp;
    }

  printf("%08x %08x %lg\n", i[1], i[0], d);

  return 0;
}

On an Intel (little endian processor), you get this output

40300000 00000000 16
00000000 00003040 6.1027e-320

Are you perhaps running this on a PPC Mac (big endian)? Seems like one of your tools is not doing the right thing on your architecture. File a bug with the vendors.

As a hacky workaround, I suggest turning your number into a string and then converting it back to a double on the other end.

Answer 2

I don't know entirely what is going wrong in your situation, but I think I may be able to partially illuminate what is happening to your code. I ran the following quick hack to test a theory:

void hexdump_double(double dbl)
{
    assert(8 == sizeof(dbl));
    printf("double: %02X %02X %02X %02X %02X %02X %02X %02X (%lg)\n",  
           ((char *)&(dbl))[0],
           ((char *)&(dbl))[1],
           ((char *)&(dbl))[2],
           ((char *)&(dbl))[3],
           ((char *)&(dbl))[4],
           ((char *)&(dbl))[5],
           ((char *)&(dbl))[6],
           ((char *)&(dbl))[7],
           dbl);
}

int main()
{
    hexdump_double(6.1026988574311E-320);
}

Which produces some exciting output:

double: 40 30 00 00 00 00 00 00 (6.1027e-320)

As you can see, that little floating-point number isn't any random pattern of bits. However, it also doesn't look to be related to "16."

Where Zend_Amf documentation states that ActionScript Number type is returned as a PHP float, what is meant is the class Number documented by Adobe: Adobe.com Flex documentation. It does not mean that any "number" will be passed as a double.

An int with value less than 2^29 will be returned transmitted in AMF as an integer type, and I assume Zend_Amf will return this as an integer.

How are you transmitting the AMF object from your ActionScript? Is it feasible to dump the bytes being sent?