spearman_rank_corr panics on NaN input (partial_cmp unwrap in compute_ranks)

[krishna3554]

Summary

compute_ranks inside spearman_rank_corr sorts with partial_cmp(...).unwrap(), which panics when either input slice contains NaN. Since spearman_rank_corr accepts arbitrary &[f64] financial series — where NaN is a routine representation of missing bars/indicators — any NaN in the input crashes the caller instead of returning None like the sibling pearson_corr handles gracefully.

Location

• File: crates/czsc-core/src/utils/corr.rs
• Function: compute_ranks (nested in spearman_rank_corr), line 136

indexed_data.sort_by(|a, b| a.0.partial_cmp(&b.0).unwrap());

Problem

f64::partial_cmp returns None whenever either operand is NaN (NaN compares unordered to everything, including itself), so .unwrap() panics with "called Option::unwrap() on a None value" on first NaN encountered by the comparator.

Context that makes this likely to be hit in practice:

1. The crate's own conventions elsewhere treat missing data as NaN (typical for pandas→Rust ported pipelines feeding czsc-python).
2. The public contract documented on these functions only mentions empty/mismatched-length handling ("当数据为空或长度不一致时返回 None") — a caller has no reason to pre-filter NaN.
3. pearson_corr, called at the end of spearman_rank_corr, tolerates NaN inputs (it would return Some(NaN) rather than crash) — so the two correlation helpers have inconsistent failure behavior for identical inputs.

Secondary note: even if the panic were avoided, NaN values would fall through the tie-grouping loop as self-unequal singletons and receive sequential ranks, silently producing a meaningless coefficient. Returning None (or an error) is the semantically correct outcome.

Trigger / Reproduction

Static analysis finding — not confirmed by execution; derived from the comparator at master (701e480a):

let x = [1.0, 2.0, f64::NAN];
let y = [1.0, 2.0, 3.0];
spearman_rank_corr(&x, &y); // panics: unwrap on None
// pearson_corr(&x, &y);     // returns Some(...) without panicking

Expected Behavior

Either return None when any element is non-finite (mirroring the documented None-cases), or use a total-order comparator such as total_cmp (stable since Rust 1.62) so NaN sorts deterministically — though returning None better matches the function's existing contract.

Actual Behavior

Panic propagates out of spearman_rank_corr into the calling signal/trader pipeline.

Impact

A single NaN bar in a series passed to Spearman-based signals aborts the whole computation thread/process instead of degrading to "no correlation", a meaningful robustness gap for a market-data library where NaN is the standard sentinel for absent data.

Suggested Direction

Replace the comparator with a.0.total_cmp(&b.0) if NaN should be ranked deterministically, or add an early if x.iter().any(|v| v.is_nan()) || y.iter().any(|v| v.is_nan()) { return None; } guard in spearman_rank_corr. A unit test with NaN input would lock the chosen behavior in.