vortex-data / vortex / 16634428462

88%

push

github

[chore] improve binary_numeric conformance (#4069)

⏺ Summary of Work Completed

Based on the git diff, here's a comprehensive summary of the changes
made:

  1. Enhanced Binary Numeric Conformance Testing

- Modified the test harness in
/vortex-array/src/compute/conformance/binary_numeric.rs to handle
overflow/underflow on a per-element basis rather
  than skipping entire operations
- Added detailed documentation explaining the test strategy and
supported operations
  - Implemented per-element overflow tracking that:
    - Runs operations even when some elements would overflow
    - Validates non-overflowing elements produce correct results
- Verifies that elements expected to overflow actually do fail the
checked operation
- Added edge case testing for special values (zero, max, min, NaN,
infinity for floats)

  2. Added Binary Numeric Tests Across Multiple Encodings

Added comprehensive binary numeric conformance tests to 11 different
array encodings:

  - ALP (Adaptive Lossless Floating-Point): Tests for f32/f64 arrays
- ALP-RD (ALP with Reference Data): Tests for f32/f64 arrays with
reference data encoding
- Dict (Dictionary): Tests for dictionary-encoded arrays including
nullable and sliced variants
- BitPacking: Tests for bit-packed integer arrays (u8, u16, u32, u64,
i32)
  - Delta: Tests for delta-encoded arrays with constant values
- FoR (Frame of Reference): Tests for frame-of-reference encoded arrays
  - RunEnd: Tests for run-end encoded arrays with repeated values
  - Sparse: Tests for sparse arrays with various fill values
  - ZigZag: Tests for zigzag-encoded signed integers
  - Primitive: Tests for basic primitive arrays
- Various consistency tests were also added alongside the binary numeric
tests

  3. Test Case Improvements

- Simplified test cases for Delta encoding to use arrays with constant
values (e.g., [1, 1, 1, 1, 1]) which better test the encoding's behavior
  - Added diverse test scenarios including:
    - Basic small arrays
... (continued)

161 of 188 new or added lines in 10 files covered. (85.64%)

46018 of 55411 relevant lines covered (83.05%)

452923.98 hits per line