[bhattsamir]

Problem

queue_shuffle(v, shuffle_cherries=True) returns an invalid label mapping most of the time. When child_order is swapped to [c2, c1], the leaf-label step still writes c2 to new_pair.1 and c1 to new_pair.0:

if child_order[1] < n_leaves {
    label_mapping[new_pair.1] = c2;   // c2 is child_order[0] after a swap
}

So the wrong child is written whenever the children were swapped. Sometimes that child is an internal node index. The mapping then either isn't a permutation of the leaves or doesn't match v_qs.

Repro (phylo2vec 1.3.1, random v with n = 5–20): 925 of 1,000 calls returned an invalid mapping. shuffle_cherries=False is unaffected, which is the only case the existing tests cover.

Fix

Read the leaf labels from child_order, the same way internal nodes are queued and coded a few lines above.

Tests

New test_queue_shuffle_shuffle_cherries (n = 3, 10, 50, 100 random vectors each) checks that, with shuffle_cherries = true:

• v_qs is ordered
• the mapping is a permutation of the leaves
• cophenetic_distances(v_qs)[i][j] == cophenetic_distances(v)[map[i]][map[j]]

The test fails on main and passes with this change. cargo test -p phylo2vec passes (340 unit tests and 43 doc tests). cargo fmt --check and clippy are clean.

🤖 Generated with Claude Code

[Neclow]

Thanks for the PR! Pushed three commits on top:

• refactor: removed the label_mapping uniqueness assert in queue_shuffle. It couldn't catch this bug (the bad mappings held unique internal-node indices, so it passed), and with the fix it can't fire, so it was just building a HashSet on every call.
• test(py): shuffle_cherries=True test in test_reorder.py.
• test(r): queue_shuffle test in test_utils.R (there was none), covering both values of shuffle_cherries.

[Neclow]

We don't usually use loops for the tests here in Rust, we let Python/R have multiple tests.

[Neclow]

Done in ff5c66b (one sample per case, n = 5, 10, 50, 100).