|0.4.2||Nov 17, 2018|
|0.4.1||Feb 14, 2018|
|0.3.5||Jan 14, 2018|
|0.3.3||Dec 29, 2017|
|0.2.7||Nov 2, 2016|
#46 in #hashmap
116,458 downloads per month
Used in 55 crates (14 directly)
Awesome hash table implementation in just Rust (stable, no unsafe code).
Note: the crate ordermap has been renamed with no change in functionality to indexmap; please use it under its new name.
Please read the API documentation here
This was inspired by Python 3.6's new dict implementation (which remembers the insertion order and is fast to iterate, and is compact in memory).
Some of those features were translated to Rust, and some were not. The result was ordermap, a hash table that has following properties:
- Order is independent of hash function and hash values of keys.
- Fast to iterate.
- Indexed in compact space.
- Preserves insertion order as long as you don't call .remove().
- Uses robin hood hashing just like Rust's libstd HashMap.
- It's the usual backwards shift deletion, but only on the index vector, so it's cheaper because it's moving less memory around.
Does not implement (Yet)
- .reserve() exists but does not have a complete implementation
OrderMap derives a couple of performance facts directly from how it is constructed, which is roughly:
Two vectors, the first, sparse, with hashes and key-value indices, and the second, dense, the key-value pairs.
- Iteration is very fast since it is on the dense key-values.
- Removal is fast since it moves memory areas only in the first vector, and uses a single swap in the second vector.
- Lookup is fast-ish because the hashes and indices are densely stored. Lookup also is slow-ish since hashes and key-value pairs are stored in separate places. (Visible when cpu caches size is limiting.)
- In practice, OrderMap has been tested out as the hashmap in rustc in PR45282 and the performance was roughly on par across the whole workload.
- If you want the properties of OrderMap, or its strongest performance points fits your workload, it might be the best hash table implementation.
- Insertion order is preserved (.swap_remove() perturbs the order, like the method name says).
- Implements .pop() -> Option<(K, V)> in O(1) time.
- OrderMap::new() is empty and uses no allocation until you insert something.
- Lookup key-value pairs by index and vice versa.
- No unsafe.
- Supports IndexMut.
Where to go from here?
- Ideas and PRs for how to implement insertion-order preserving remove (for example tombstones) are welcome. The plan is to split the crate into two hash table implementations a) the current compact index space version and b) the full insertion order version.
Ideas that we already did
It can be an indexable ordered map in the current fashion (This was implemented in 0.2.0, for potential use as a graph datastructure).
Idea for more cache efficient lookup (This was implemented in 0.1.2).
Current indices: Vec<Pos>. Pos is interpreted as (u32, u32) more or less when .raw_capacity() fits in 32 bits. Pos then stores both the lower half of the hash and the entry index. This means that the hash values in Bucket don't need to be accessed while scanning for an entry.
- 0.4.2 (ordermap)
- Inserted more deprecation information in the documentation. Note: the crate ordermap has been renamed with no change in functionality to indexmap; please use it under its new name.
- Renamed crate to indexmap; the ordermap crate is now deprecated and the types OrderMap/Set now have a deprecation notice.
- This is the last release series for this ordermap under that name, because the crate is going to be renamed to indexmap (with types IndexMap, IndexSet) and no change in functionality!
- The map and its associated structs moved into the map submodule of the
crate, so that the map and set are symmetric
- The iterators, Entry and other structs are now under ordermap::map::
- Internally refactored OrderMap<K, V, S> so that all the main algorithms (insertion, lookup, removal etc) that don't use the S parameter (the hasher) are compiled without depending on S, which reduces generics bloat.
- Entry<K, V> no longer has a type parameter S, which is just like the standard HashMap's entry.
- Minimum Rust version requirement increased to Rust 1.18
- Documentation improvements
- The .retain() methods for OrderMap and OrderSet now traverse the elements in order, and the retained elements keep their order
- Added new methods .sort_by(), .sort_keys() to OrderMap and .sort_by(), .sort() to OrderSet. These methods allow you to sort the maps in place efficiently.
- Document insertion behaviour better by @lucab
- Updated dependences (no feature changes) by @ignatenkobrain
- Add OrderSet by @cuviper!
- OrderMap::drain is now (too) a double ended iterator.
- In all ordermap iterators, forward the collect method to the underlying iterator as well.
- Add crates.io categories.
- The methods get_pair, get_pair_index were both replaced by get_full (and the same for the mutable case).
- Method swap_remove_pair replaced by swap_remove_full.
- Add trait MutableKeys for opt-in mutable key access. Mutable key access is only possible through the methods of this extension trait.
- Add new trait Equivalent for key equivalence. This extends the Borrow trait mechanism for OrderMap::get in a backwards compatible way, just some minor type inference related issues may become apparent. See #10 for more information.
- Implement Extend<(&K, &V)> by @xfix.
- Fix deserialization to support custom hashers by @Techcable.
- Add methods .index() on the entry types by @garro95.
- Add methods .with_hasher(), .hasher().
- Support ExactSizeIterator for the iterators. By @Binero.
- Use Box<[Pos]> internally, saving a word in the OrderMap struct.
- Serde support, with crate feature "serde-1". By @xfix.
- Add iterator .drain(..) by @stevej.
- Add method .is_empty() by @overvenus.
- Implement PartialEq, Eq by @overvenus.
- Add method .sorted_by().
- Add iterators .values() and .values_mut().
- Fix compatibility with 32-bit platforms.
- Add .retain().
- Add OccupiedEntry::remove_entry and other minor entry methods, so that it now has all the features of HashMap's entries.
- Improved .pop() slightly.
- Improved performance of .insert() (#3) by @pczarn.
- Generalize Entry for now, so that it works on hashmaps with non-default hasher. However, there's a lingering compat issue since libstd HashMap does not parameterize its entries by the hasher (S typarm).
- Special case some iterator methods like .nth().
- Disable the verbose Debug impl by default.
- Fix doc links and clarify docs.
- Add more HashMap methods & compat with its API.
- Experimental support for .entry() (the simplest parts of the API).
- Add .reserve() (placeholder impl).
- Add .remove() as synonym for .swap_remove().
- Changed .insert() to swap value if the entry already exists, and return Option.
- Experimental support as an indexed hash map! Added methods .get_index(), .get_index_mut(), .swap_remove_index(), .get_pair_index(), .get_pair_index_mut().
- Implement the 32/32 split idea for Pos which improves cache utilization and lookup performance.
- Initial release.