This is a façade over std::vector that makes it act like std::unordered_map. This may seem like something very stupid to do, but it has some advantages over std::unordered_map:
- It takes significantly less memory
- It does not require the key type to have a hash function
- It preserves insertion order
- It can be faster (if indexed by
std::string, it's faster if it contains less than about 200 elements)
Vector can find elements in linear time, hashtable in constant time. That's a pretty elementary part of computer science. However, better complexity does not always mean it's faster.
Surprisingly, on GCC 9, it can be twice as fast as std::unordered_map with sizes around 30. If the size is large enough, the unfavourable complexity kicks in and the speed will start lagging after std::unordered_map. The breakpoint is around size 200. It is not as good on older compilers like MSVC 2015, where the breakpoint is only around 20.
This is the good part:
Now, why is it so? My guess is that it has something to do with memory locality, less dynamic allocation, better vectorisation possibilities and maybe speculative execution.