LinkedLruHashMap: Bad state when remove() removes the last entry

[hugovangalen]

My own code crashed in the method _removeLru with a null pointer access when calling _tail.next.

After debugging I discovered that the cause seems to be in the remove() method.

There exists a check whether the removed entry is equal to _head, or it is equal to _tail.

If it happens to be the last item that gets removed, that is not handled correctly. Right now it branches into the entry == _head, resets _head, but leaves _tail in its old value, which is wrong.

If my understanding is correct, in that case both _tail and _head need to be reset to null. Adding a test for entry == _head && entry == _tail to reset the _tail and _head to null appears to fix the problem:

  @override
  V remove(Object key) {
    final entry = _entries.remove(key);
    if (entry != null) {
      if (entry == _head && entry == _tail) {
        _head = null;
        _tail = null;
      } else if (entry == _head) {
        _head = _head.next;
      } else if (entry == _tail) {
        _tail.previous.next = null;
        _tail = _tail.previous;
      } else {
        entry.previous.next = entry.next;
      }

      return entry.value;
    }
    return null;
  }

[cbracken]

Fixed by #386 and released in v0.26.1. Thanks for diagnosing and reporting this!

[hugovangalen]

@cbracken No problem. I'd like to stress that there are still issues with dangling references. See ticket #412.