[ABA-14] test(vortex-fastlanes): repro for ZeroBytes scheme reporting nbytes()==0

[abnobdoss]

Summary

• Adds a single #[ignore]-d regression test reproducing the ABA-14 bug: BitPackedArray built from all-zero input reports nbytes() == 0
• No fix included — a semantics decision is needed from upstream before the fix approach can be chosen
• Test name: issue_aba14_bitpack_zerobytes_reports_nonzero_nbytes in encodings/fastlanes/src/bitpacking/array/bitpack_compress.rs

Linear

https://linear.app/abanoubdoss/issue/ABA-14

Root cause (confirmed on develop)

When bitpack_to_best_bit_width processes an all-zero array it selects bit_width=0 (zero packed bits is cheapest). bitpack_primitive short-circuits at line 147–149 with Buffer::<T>::empty(). The VTable exposes that buffer as the sole physical buffer, so Array::nbytes() (which sums buffer lengths) returns 0 for an array that logically holds 1024 elements.

Downstream consequence

vortex-compressor/src/estimate.rs maps after_nbytes == 0 from sample compression to EstimateScore::ZeroBytes. EstimateScore::is_valid (line 142–149) returns false for ZeroBytes, so the scheme that achieves perfect compression of all-zero data is silently excluded from the compressor's selection.

Open question

Should nbytes() for a perfectly-compressible scheme like ZeroBytes (bit_width==0) report:

• Virtual uncompressed size (e.g. len * ptype.byte_width()) — preserves comparability with uncompressed arrays in the estimator, but misrepresents actual storage
• Actual byte size (currently: 0) — truthful about storage, but breaks the estimator's assumption that 0 bytes means "bad sample" rather than "perfect compression"

Currently 0 causes vortex-compressor/src/estimate.rs:122–149 to classify the result as ZeroBytes and exclude it from selection. The TODO comment at line 113–116 in estimate.rs already acknowledges this ambiguity.

Validation

• Test reproduces the bug: cargo test -p vortex-fastlanes -- issue_aba14 (with #[ignore] removed) panics with nbytes=0 with bit_width=0
• All existing tests continue to pass with #[ignore] restored

🤖 Generated with Claude Code

Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com