In PHP source code function uniqid() have following C code: (I removed some types to shorten it)
//...
struct timeval tv;
gettimeofday(&tv, NULL);
int sec = (int) tv.tv_sec;
int usec = (int) (tv.tv_usec % 0x100000);
// The max value usec can have is 0xF423F,
// so we use only five hex digits for usecs.
printf("%08x%05x", sec, usec);
//...
If we put criticism aside, they try to produce 64 bit timestamp.
0xF423F is probably CLOCKS_PER_SEC - 1 (CLOCKS_PER_SEC is decimal 1000000),
but where this 0x100000 come from and what could be the reason to use modulus instead of bitwise and?
She or he could write the Unique ID as printf("%08x%08x", sec, usec)
sample output:
55189926000eb16f
5518997900051219
5518997a0005171b
The zeros in position 8 to 10 are consistent, they don't add entropy, so he wants to get rid of those zeros. The new UID will be 3 bytes shorter with the same entropy. He could simply use printf("%08x%05x", sec, usec);
sample output:
55189926eb16f
5518997951219
5518997a5171b
But that's on the assumption that usec is guaranteed to be less than 0x100000 otherwise UID will be up to 16 bytes long. You need % 0x100000 for insurance. It's also the same as & 0xFFFFF. Technically the insurance should be % 1000000 (decimal), but it doesn't really matter, it's still the same entropy.
Or we could just use the 16 byte version because saving 3 lousy bytes don't matter these days.