All adaptors bounded by Hash would potentially benefit from a with_hasher variant that allowed supplying a faster Hasher (e.g., ahash or fxhash):

• duplicates
  • duplicates_with_hasher
• duplicates_by
  • duplicates_by_with_hasher
• unique
  • unique_with_hasher
• unique_by
  • unique_by_with_hasher
• all_unique
  • all_unique_with_hasher
• into_group_map
  • into_group_map_with_hasher
• into_group_map_by
  • into_group_map_by_with_hasher
• into_grouping_map
  • into_grouping_map_with_hasher
• into_grouping_map_by
  • into_grouping_map_by_with_hasher
• counts
  • counts_with_hasher
• counts_by
  • counts_by_with_hasher

Implementing each of these is a three step process:

1. Add trailingS = RandomState generic parameter to the adaptor type, where S: BuildHasher
2. Replace relevant calls to HashMap::new and HashMap::default with HashMap::with_hasher
3. Modify the adaptor type's internal constructor to consume an extra hash_builder: S parameter.
4. Add a with_hasher adaptor to Itertools with a hash_builder: S parameter that's passed to the internal constructor.
   • The documentation for this should defer to the non-with_hasher variant.
   • The documentation should include the same warning as HashMap::with_hasher
5. Redefine the non-with_hasher Itertools method to defer to the with_hasher variant.