Add dense_integer_mapt type

This gives a map-like interface but vector-like lookup cost, ideal for
mapping keys that have a bijection with densely packed integers, such as
the location numbers of a well-formed goto_modelt or goto_programt.

Author: smowton

src/util/dense_integer_map.h

@@ -0,0 +1,296 @@
+/*******************************************************************\
+
+Module: Dense integer map
+
+Author: Diffblue Ltd
+
+\*******************************************************************/
+
+/// \file
+/// Dense integer map
+
+#ifndef CPROVER_UTIL_DENSE_INTEGER_MAP_H
+#define CPROVER_UTIL_DENSE_INTEGER_MAP_H
+
+#include <limits>
+#include <unordered_set>
+#include <vector>
+
+#include <util/invariant.h>
+#include <util/optional.h>
+
+/// Identity functor. When we use C++20 this can be replaced with std::identity.
+class identity_functort
+{
+public:
+  template <typename T>
+  constexpr T &&operator()(T &&t) const
+  {
+    return std::forward<T>(t);
+  }
+};
+
+/// A map type that is backed by a vector, which relies on the ability to (a)
+/// see the keys that might be used in advance of their usage, and (b) map those
+/// keys onto a dense range of integers. You should specialise
+/// key_to_dense_integer for your key type before using. If it maps your keys
+/// onto too sparse a range, considerable memory will be wasted, as well as time
+/// spent skipping over the unused indices while iterating.
+///
+/// Value type V must be default constructible.
+///
+/// The type is optimised for fast lookups at the expense of flexibility.
+/// It makes one compromise on the iterface of std::map / unordered_map: the
+/// iterator refers to a pair<key, optionalt<value>>, where the value optional
+/// will always be defined. This is because the backing store uses optionalt
+/// this way and we don't want to impose the price of copying the key and value
+/// each time we move the iterator just so we have a <const K, V> & to give out.
+template <class K, class V, class KeyToDenseInteger = identity_functort>
+class dense_integer_mapt
+{
+public:
+  typedef std::vector<K> possible_keyst;
+
+private:
+  int64_t offset;
+  typedef std::vector<std::pair<K, V>> backing_storet;
+  backing_storet map;
+  std::vector<bool> value_set;
+  std::size_t n_values_set;
+  possible_keyst possible_keys_vector;
+
+  std::size_t key_to_index(const K &key) const
+  {
+    auto key_integer = KeyToDenseInteger{}(key);
+    INVARIANT(((int64_t)key_integer) >= offset, "index should be in range");
+    auto ret = (std::size_t)key_integer - offset;
+    INVARIANT(ret < map.size(), "index should be in range");
+    return ret;
+  }
+
+  void mark_index_set(std::size_t index)
+  {
+    if(!value_set[index])
+    {
+      ++n_values_set;
+      value_set[index] = true;
+    }
+  }
+
+  bool index_is_set(std::size_t index) const
+  {
+    return value_set[index];
+  }
+
+public:
+  template <class UnderlyingIterator, class UnderlyingValue>
+  class iterator_templatet
+    : public std::iterator<std::forward_iterator_tag, UnderlyingValue>
+  {
+    typedef std::iterator<std::forward_iterator_tag, UnderlyingValue>
+      base_typet;
+    typedef iterator_templatet<UnderlyingIterator, UnderlyingValue> self_typet;
+    typedef dense_integer_mapt<K, V, KeyToDenseInteger> map_typet;
+
+  public:
+    iterator_templatet(UnderlyingIterator it, const map_typet &underlying_map)
+      : underlying_iterator(it), underlying_map(underlying_map)
+    {
+      it = skip_unset_values(it);
+    }
+
+    // Convert iterator to const_iterator
+    // (redundant when defined in the const_iterator itself)
+    operator iterator_templatet<
+      typename backing_storet::const_iterator,
+      const typename backing_storet::value_type>() const
+    {
+      return {underlying_iterator, underlying_map};
+    }
+
+    self_typet operator++()
+    {
+      self_typet i = *this;
+      underlying_iterator = advance(underlying_iterator);
+      return i;
+    }
+    self_typet operator++(int junk)
+    {
+      underlying_iterator = advance(underlying_iterator);
+      return *this;
+    }
+    typename base_typet::reference operator*() const
+    {
+      return *underlying_iterator;
+    }
+    typename base_typet::pointer operator->() const
+    {
+      return &*underlying_iterator;
+    }
+    bool operator==(const self_typet &rhs) const
+    {
+      return underlying_iterator == rhs.underlying_iterator;
+    }
+    bool operator!=(const self_typet &rhs) const
+    {
+      return underlying_iterator != rhs.underlying_iterator;
+    }
+
+  private:
+    UnderlyingIterator advance(UnderlyingIterator it)
+    {
+      return skip_unset_values(std::next(it));
+    }
+
+    UnderlyingIterator skip_unset_values(UnderlyingIterator it)
+    {
+      auto iterator_pos = std::distance(
+        underlying_map.map.begin(),
+        (typename backing_storet::const_iterator)it);
+      while((std::size_t)iterator_pos < underlying_map.map.size() &&
+            !underlying_map.value_set.at(iterator_pos))
+      {
+        ++iterator_pos;
+        ++it;
+      }
+      return it;
+    }
+
+    UnderlyingIterator underlying_iterator;
+    const map_typet &underlying_map;
+  };
+
+  typedef iterator_templatet<
+    typename backing_storet::iterator,
+    typename backing_storet::value_type>
+    iterator;
+  typedef iterator_templatet<
+    typename backing_storet::const_iterator,
+    const typename backing_storet::value_type>
+    const_iterator;
+
+  dense_integer_mapt() : offset(0), n_values_set(0)
+  {
+  }
+
+  template <typename Iter>
+  void setup_for_keys(Iter first, Iter last)
+  {
+    int64_t highest_key = std::numeric_limits<int64_t>::min();
+    int64_t lowest_key = std::numeric_limits<int64_t>::max();
+    std::unordered_set<int64_t> seen_keys;
+    for(Iter it = first; it != last; ++it)
+    {
+      int64_t integer_key = KeyToDenseInteger{}(*it);
+      highest_key = std::max(integer_key, highest_key);
+      lowest_key = std::min(integer_key, lowest_key);
+      INVARIANT(
+        seen_keys.insert(integer_key).second,
+        "keys for use in dense_integer_mapt must be unique");
+    }
+
+    if(highest_key < lowest_key)
+    {
+      offset = 0;
+    }
+    else
+    {
+      auto map_size = (highest_key - lowest_key) + 1;
+      INVARIANT(
+        map_size > 0, // overflowed?
+        "dense_integer_mapt size should fit in std::size_t");
+      INVARIANT(
+        ((std::size_t)map_size) <= std::numeric_limits<std::size_t>::max(),
+        "dense_integer_mapt size should fit in std::size_t");
+      offset = lowest_key;
+      map.resize((highest_key - lowest_key) + 1);
+      for(Iter it = first; it != last; ++it)
+        map.at(key_to_index(*it)).first = *it;
+      value_set.resize((highest_key - lowest_key) + 1);
+    }
+
+    possible_keys_vector.insert(possible_keys_vector.end(), first, last);
+  }
+
+  const V &at(const K &key) const
+  {
+    std::size_t index = key_to_index(key);
+    INVARIANT(index_is_set(index), "map value should be set");
+    return map.at(index).second;
+  }
+  V &at(const K &key)
+  {
+    std::size_t index = key_to_index(key);
+    INVARIANT(index_is_set(index), "map value should be set");
+    return map.at(index).second;
+  }
+
+  V &operator[](const K &key)
+  {
+    std::size_t index = key_to_index(key);
+    mark_index_set(index);
+    return map.at(index).second;
+  }
+
+  std::pair<iterator, bool> insert(const std::pair<const K, V> &pair)
+  {
+    std::size_t index = key_to_index(pair.first);
+    iterator it{std::next(map.begin(), index), *this};
+
+    if(index_is_set(index))
+      return {it, false};
+    else
+    {
+      mark_index_set(index);
+      map.at(index).second = pair.second;
+      return {it, true};
+    }
+  }
+
+  std::size_t count(const K &key) const
+  {
+    return index_is_set(key_to_index(key));
+  }
+
+  std::size_t size() const
+  {
+    return n_values_set;
+  }
+
+  const possible_keyst &possible_keys() const
+  {
+    return possible_keys_vector;
+  }
+
+  iterator begin()
+  {
+    return iterator{map.begin(), *this};
+  }
+
+  iterator end()
+  {
+    return iterator{map.end(), *this};
+  }
+
+  const_iterator begin() const
+  {
+    return const_iterator{map.begin(), *this};
+  }
+
+  const_iterator end() const
+  {
+    return const_iterator{map.end(), *this};
+  }
+
+  const_iterator cbegin() const
+  {
+    return begin();
+  }
+
+  const_iterator cend() const
+  {
+    return end();
+  }
+};
+
+#endif // CPROVER_UTIL_DENSE_INTEGER_MAP_H