Containers

Containers in the standard libraries

• Mutable:
  • core::dlist (doubly-linked list - uses @ but avoiding that would require unsafe)
  • core::vec (dynamic array)
  • core::str (string implemented on top of the same dynamic array as core::vec)
  • core::hashmap (hash table - set and map)
  • core::trie (radix trie - set and map, for uint keys only at the moment)
  • std::treemap (balanced binary search tree - set and map)
  • std::smallintmap (dense array-based map)
  • std::deque (ring buffer)
  • std::priority_queue (binary heap)
  • std::bitv
• Persistent:
  • std::list (singly-linked list)
  • std::fun_treemap (unbalanced binary search tree - not very useful)

Obsolete (to be removed)

• std::dvec (~[] in a mut field) - #4985
• std::oldmap (chaining-based hash table using lots of @ and mut fields) - #4986

Wanted

• B-tree (Map and Set implementations) - #4992
• small vector (3-word struct storing small arrays on the stack) - #4991
• LRU cache (doubly-linked list with a hash table pointing at the nodes) - #4988
• persistent balanced binary search tree (map and set) - #4987
• persistent heap
• persistent hash-based map/set
• compile-time lookup tables via syntax extensions - #4864

Issues

• smallintmap should implement a set type - #4984
• Add range searches to TreeMap/TreeSet - #4604
• LinearMap fields should be private - #4764
• add shrink_to_fit method for vec and hashmap - #4960
• split the Map and Set traits into Map, MutableMap, PersistentMap and Set, MutableSet, PersistentSet - #4989
• Move dlist to std - #3549
• benchmark whether 1.5x is a better growth factor for vectors - #4961
• add reserve/reserve_at_least to std::deque - #4994

Stack/Queue-like containers

Suggested API convention

• fn back(&self) -> Option<&self/T>
• fn front(&self) -> Option<&self/T>
• fn push_back(&mut self, value: T)
• fn push_front(&mut self, value: T)
• fn pop_back(&mut self) -> Option<T>
• fn pop_front(&mut self) -> Option<T>

This is the same naming convention as C++, and seems to be the most consistent way to do it.

Python uses append, pop, appendleft and popleft which isn't as uniform and doesn't lead to an obvious naming convention for equivalents to back and front (they just use [0] and [-1]).

Data structures

• vectors can efficiently implement front, back push_back and pop_back
• doubly-linked lists and circular buffers can efficiently implement all of the operations
• a singly-linked list can efficiently implement front, push_front and pop_front, with the addition of back and push_back if it's either circular or if the container wrapper stores two node pointers
• a priority queue technically doesn't implement any of these operations, it shouldn't share traits

Traits

This is the tricky part.... there isn't really an obvious way to divide these up into nice traits. A stack could either use push_front and pop_front or push_back and pop_back. The same thing applies to a queue, which could either use push_back and pop_front or push_front and pop_back`.

A Deque trait would include all of the operations, so at least that part is easy.