repro(5526): distinct filtering bug in hybrid search

[arthurgousset]

Description

Reproduces hybrid search bug. The previous reproduction failed because the vector dimensions were not extreme.enough.

// Instead of 4D vectors, try higher dimensions with more extreme differences
"_vectors": {"default": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]}  // 8D max similarity
"_vectors": {"default": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0]}  // 8D min similarity

Important

Do not merge, this is a PR to reproduce the bug not to be merged.

Other changes

NA

Tested

Successfully reproduced the bug

cargo test search::hybrid::reproduce_hybrid_distinct_bug -- --nocapture

running 1 test
SEARCH RESULTS:
  Hit 0: id=2, product_id="DUPLICATE_PRODUCT_ID"
  Hit 1: id=1, product_id="DUPLICATE_PRODUCT_ID"

thread 'search::hybrid::reproduce_hybrid_distinct_bug' panicked at crates/meilisearch/tests/search/hybrid.rs:699:9:
BUG REPRODUCED! Found 2 total results but only 1 unique product_ids. This means distinct filtering failed in hybrid search! Results: ["DUPLICATE_PRODUCT_ID", "DUPLICATE_PRODUCT_ID"]
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
test search::hybrid::reproduce_hybrid_distinct_bug ... FAILED

failures:

failures:
    search::hybrid::reproduce_hybrid_distinct_bug

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 790 filtered out; finished in 1.62s

error: test failed, to rerun pass `-p meilisearch --test integration`

Related issues

• closes [ID]

Backwards compatibility

NA not to be merged, only reproduces bug.

Documentation

NA