[pdq] Replace POSIX dependent functions with STL equivalent

[ianwal]

Summary

While I was testing vpdq on Windows I had to make a couple changes to get it to build with pdq.

They're solid changes in general, so I thought I would upload them.

Use C++11 STL over OS dependent functions in general:

• Change file pointers to streams
  • File pointers were used with getline() which is POSIX only. Replaced file pointers only wherever necessary with streams and use std::getline(). I tried to change as little as possible here.
• Use C++ random device instead of random() which is POSIX only
  • This could have been done with rand() but I just used the C++11 one because it's a C++ project
• Fix __builtin_popcount where hamming*() should have been used
  • There's a define in pdqhamming.h which chooses either a lookup table or __builtin_popcount based on the USE_BUILTIN_POPCOUNT directive but some functions were directly calling __builtin_popcount regardless.

None of these changes should affect bindings. The Python binding doesn't call any of the changed functions since they're almost all file IO.

I'm not sure why __builtin_popcount is being used at all. std::bitset::count and __builtin_popcount generate the same assembly. I can replace __builtin_popcount with std::bitset::count if you would like.

The getline problem is a really good example of why it's not good to do using namespace std; since std::getline and getline are both functions.

I vaguely remember reading something about Buck for building and if you guys use that for PDQ I'm not sure how these changes would be affected by that.

Test Plan

Run included PDQ tests and vPDQ tests.

[Dcallies]

Thanks for the detailed explanation in the summary. I had a few questions that normally would have forced a roundtrip, but I was able to answer them from reading your summary.

As for the test plan, I don't know if there are dedicated unittests for it, the blast radius for the C++ implementation might be slightly larger if it is the one linked to the python bindings, though you mentioned it's not in the summary.

[ianwal]

Did you have any thoughts on using std::bitset::count?

[Dcallies]

If it produces the right output, then I'm game!