Fix robin_hash::erase() method

[zhicme]

Bug Description

There is a bug in the range-erase implementation:

if (first == last) {
  return mutable_iterator(first);
}

auto first_mutable = mutable_iterator(first);
auto last_mutable = mutable_iterator(last);
for (auto it = first_mutable.m_bucket; it != last_mutable.m_bucket; ++it) {
  if (!it->empty()) {
    it->clear();
    m_nb_elements--;
  }
}

The problem is that erase(first, last) treats the bucket array as linear rather than circular when clearing the erased range.

As a result, when the erased range logically wraps around the end of the bucket array, the implementation does not correctly account for the circular probe sequence before performing the backward-shift cleanup. This can leave the table in an inconsistent state and may cause incorrect behavior afterward.

In particular, this may lead to a situation where the map ends up containing duplicate keys, which violates the map invariant.

#include <tsl/robin_map.h>
#include <iostream>

struct BadHash {
  std::size_t operator()(int) const noexcept {
    return 6;
  }
};

int main() {
  tsl::robin_map<int, int, BadHash> map(8);
  map.max_load_factor(0.95f);

  map.insert({1, 10});
  map.insert({2, 20});
  map.insert({3, 30});
  map.insert({4, 40});

  for (const auto& kv : map) {
    std::cout << kv.first << ":" << kv.second << "\n";
  }

  map.erase(map.find(1), map.end());
  auto res = map.insert({3, 300});

  std::cout << "inserted_again = " << res.second << "\n";
  std::cout << "size = " << map.size() << "\n";

  for (const auto& kv : map) {
    std::cout << kv.first << ":" << kv.second << "\n";
  }
}
// outputs:
// 3:30
// 4:40
// 1:10
// 2:20
// inserted_again = 1
// size = 3
// 3:30
// 4:40
// 3:300

To fix this issue, I reimplemented erase, added two overloaded version of next_bucket(), and introduced a new helper function cyclic_distance() to correctly handle circular bucket traversal.
I am not sure if there is a better way to fix this issue, but I hope this fix is helpful for you. 😄

[zhicme]

I’m very sorry, I forgot to switch branches before making the changes. Would you like me to submit a new PR?

[Tessil]

Very sorry for the long delay and thanks for the catch!

The change looks good, could you just apply the clang-format one the change and eventually add a small regression test? Thanks!