Expand description
§index-db
A B+tree indexing primitive for storage engines: ordered keys mapped to values, with fast point lookups, laid out as a tree of fixed-fan-out nodes rather than on the heap one entry at a time.
The one public type is BPlusTree, an ordered map. Keys are kept sorted
across the tree, so a lookup is a binary search at each level and the height
grows only with the logarithm of the entry count. Beyond point operations it
supports ordered iteration and range scans, forward and in reverse. The node
layout — sorted keys packed into fixed-capacity arrays, internal nodes routing
to children — is the same structure a storage engine persists as an on-disk
index; this release keeps the tree in memory.
§Example
use index_db::BPlusTree;
let mut index = BPlusTree::new();
index.insert(42_u32, "answer");
index.insert(7, "lucky");
assert_eq!(index.get(&42), Some(&"answer"));
assert_eq!(index.get(&13), None);
// Ordered range scan.
let keys: Vec<_> = index.range(0..50).map(|(&k, _)| k).collect();
assert_eq!(keys, vec![7, 42]);
assert_eq!(index.remove(&7), Some("lucky"));
assert_eq!(index.len(), 1);§Scope
v1.0.0 is the stable in-memory ordered map. The public API is frozen until
2.0: search, insert, delete (with merge and redistribute), forward and reverse
range scans, and bulk construction from sorted input. The tree is Sync, so
any number of threads may read it at once. Node access runs through an
internal storage seam so that a page-backed, concurrent (write-side) backend
over page-db can be added in a 1.x release without changing the tree
algorithm or breaking this API. See dev/ROADMAP.md.