Hmm, how does this help prevent atom collisions?

The code roughly does this:

$old_atoms = load_existing_atoms_from_db();
foreach ($atoms_being_published as $atom) {
  if ($atom is not one of $old_atoms) {
    $atom->save();
  }
}

Prior to this patch, we miss the existing atom when loading, so $old_atoms always comes back empty (the query builds like ... WHERE bookPHID = "phabdev" ..., which no rows match, because that isn't a PHID).

Since we fail to load the existing atom, the check to see if the atom we're publishing is being updated or created always decides to create the atom.

Then, when we save it, the atom's hash collides with the existing hash in the database, because there really is an existing atom, we just failed to load it earlier.