HsahSetA hash-set analogue that does not own its data.
It can be used to "mark" items without the need to transfer ownership to the map
rust
/// Process arguments while ignoring duplicates
fn process_args(args: impl IntoIterator<Item=String>) {
let mut same= HashRefSet::new();
for argument in args.into_iter()
{
if !same.insert(argument.as_str()) {
// Already processed this input, ignore
continue;
}
//do work...
}
}
# Serialisation support with serde crate
HashRefSet and HashType both implement Serialize and Deserialize from the serde crate if the serde feature is enabled. By default it is not.
# Hashing We use the SHA512 hashing algorithm for the implementation at present. I may implement the ability to choose different types, but as of now I think it is sufficient.
# Drawbacks
Since the item is not inserted itself, we cannot use Eq to double check there was not a hash collision.
While the hashing algorithm used (Sha512) is extremely unlikely to produce collisions, especially for small data types, keep in mind that it is not infallible.
HashRefSet is significantly slower than HashSet, so HashSet should be preferred in most cases.
Even when Clone is required to insert into HashSet, it can be ~10x faster for trivial data structures.
HashRefSet should be used if Clone is either not an option, or Clone is a significantly heavy operation on the type you're inserting.
| Benchmark | Tests | Result |
|--------------------|-----------------------------------------------|-----------------|
| owningstrings | Inserts String into HashSet by cloning | ~4,538 ns/iter |
| nonowningstrings | Inserts str into HashRefSet by reference | ~48,271 ns/iter |
| owningints | Inserts u32 into HashSet by copy | ~937 ns/iter |
| nonowningints | Inserts &u32 into HashRefSet by reference | ~31,089 ns/iter |
HashSetClone to insert a copy of the item into a HashSet is not possible (non-Clone type) or is a significantly heavy operation. (see benchmarks)HashSetMIT