Question

The hashtable and map is hashtable is implemented as a hash function but map is implemented as a tree.

My question is, in what situation, hashtable can not be used but a map is a must?

Answer 1

One motivation for choosing to use a map over a hashtable is the constraints which each one places on the key type used in the template instantiation. As described in the documentation for hash_map in the SGI implementation of STL, an instantiation hash_map requires provision of a functor which hashes K. The STL includes a functor, std::hash, which does this, but it is only implemented for a limited set of types T.

Instantiation of std::map on the other hand only requires a functor which compares objects of type K to generate a weak ordering. The standard functor std::less will work for any T which defines an operator<.

This means that, for many user-defined types, adding the support necessary to use the type as a key in a std::map is much less than that required to use it in a std::hash_map.

Aside from the question of overhead,

1. Only std::map, not std::hash_map, guarantees that the keys will be ordered, so if this is a requirement, use std::map
2. hash_map is not part of the standard, so while some STL implementations include it, this is not true of all implementations. Using hash_map therefore has potential impact on the portability of your program.