Skip to content

Latest commit

 

History

History
95 lines (73 loc) · 4.76 KB

CHANGELOG.md

File metadata and controls

95 lines (73 loc) · 4.76 KB
Raw

static_aabb2d_index changelog

All notable changes to the static_aabb2d_index crate will be documented in this file.

2.0.0 - 2023-09-04

Fixed 🐛

  • Fixed building index properly for integer types. Previously due to integer division truncation and the way the hilbert coordinate values were computed the index was not properly sorted/structured when using integer values (this lead to a less optimal tree for querying, despite everything still querying properly). Hilbert coordinate values are now computed using f64 to properly scale and work in all integer cases. This improves performance of the index when using integers.

Changed 🔧

  • ⚠️ MINOR BREAKING: numeric type used is now required to successfully cast to a f64 (previously expected to cast to/from u16). This is a very minor breaking change, and only arises when using a non-builtin numeric type that for whatever reason cannot cast to a f64 but is able to cast to/from a u16. Major version number was bumped since this is technically a breaking change for a weird edge use case.
  • Numeric type used no longer requires casting to/from a u16.
  • Rearranged math expressions when building hilbert coordinate values for performance improvement (~5%).
  • Implemented the IndexableNum trait for i8, u8, and i16 types (those types are now supported, this came as part of the fix for integer types).

1.1.0 - 2023-08-31

Changed 🔧

  • Bumped rust edition to 2021.
  • Changed internal Vecs to boxed slices (smaller size, and makes clear they don't change in size).
  • Internal code improvements for clarity.
  • Performance improvement when building an index with large number of item (8-10% measured for 1_000_000 items).
  • Avoid some allocations when building an index by determining the exact length of the level_bounds array before constructing it (performance improvement).
  • Added uninitialized memory optimization when the feature flag unsafe_optimizations is on. This further improves performance when building an index by avoiding zeroing the array of AABBs allocated for the all the boxes. Care was taken to avoid undefined behavior due to reading from or creating references to any uninitialized memory.

1.0.0 - 2023-03-02

Added ⭐

  • Added item_indices method to get a slice over the indices for all item boxes added.

Changed 🔧

  • ⚠️ BREAKING: Index now supports being empty (no longer errors when building the index if item count is 0). When the index is empty all queries will yield no results.
  • ⚠️ BREAKING: min_x, min_y, max_x, and max_y functions on index replaced with single bounds function which returns the total bounds as an AABB or None if index item count is 0.
  • ⚠️ BREAKING: fixed inconsistency in visit_query function to return break result the same as the visit_query_with_stack function.
  • ⚠️ BREAKING: fixed inconsistency in visit_neighbors function to return break result the same as the visit_neighbors_with_queue function.
  • ⚠️ BREAKING: renamed map_all_boxes_index function to all_box_indices and changed signature to return a slice rather than indexing into a slice internally.
  • Improved doc comments.

0.7.1 - 2023-02-22

Changed 🔧

  • Removed unsafe optimization involving uninitialized memory, the code did not strictly uphold the invariants required of a Vec at all times which could lead to undefined behavior. To properly perform this optimization will require more pointer manipulation spread across the code or new APIs from the Rust standard library. Index bounds checking is still toggled by the unsafe_optimizations feature.
  • INTERNAL: replaced get_at_index and set_at_index macros with simple inlined functions and simplified some function signatures to use slices rather than Vec.

0.7.0 - 2023-02-18

Added ⭐

  • ⚠️ BREAKING: Added total_cmp method to IndexableNum trait to eliminate using partial_cmp and unwrap which can panic if NaN is present (was used for nearest neighbors query). If implementing IndexableNum for your own type you must implement total_cmp for your type).
  • Added forbid(unsafe_code) attribute to crate if unsafe_optimizations feature is not enabled. This ensures there is no unsafe code used in this crate unless unsafe_optimizations feature is turned on.
  • Added CHANGELOG.md to the project for tracking changes.

Changed 🔧

  • BREAKING: renamed feature allow_unsafe to unsafe_optimizations.
  • Use std::cmp::min and std::cmp::max in implementing IndexableNum for integers (ensures use of any available compiler intrinsic for optimizations).
  • Use f32::min, f32::max, f64::min, and f64::max in implementing IndexableNum (ensures use of any available compiler intrinsic for optimizations).