#385 loose ends: CHANGELOG claims one path key (D13 split it), and the D8/D13 regression has no unit guard

[FelixKrueger]

Found by the #385 coverage re-audit (plans/08062026_se-output-collision-preflight/REAUDIT.md, 2026-08-08 — the first audit to run against the final tree; the three original audits all predate the D7–D13 review round). Verdict was INCOMPLETE on 3 items; no behavioural gap. Two are real loose ends:

1. The CHANGELOG entry contradicts what shipped. CHANGELOG.md:206-211 ends: "All four now use the same key as the pre-flight." D13 then split the key in two, and #389 refined the --passthrough half again. Today the pre-flight uses the case-folded collision_key; the three input-identity checks use the case-preserving path_identity_key; --passthrough uses both. Nothing in the CHANGELOG discloses the split (grep for path_identity_key / "identity key" returns nothing). The bug the entry describes is genuinely fixed — both keys share lexical_normalise — so the entry is stale rather than wrong about the outcome. Fix: correct the final sentence to describe both keys and why they differ (case-folding is right for output paths, wrong for input identity).

2. The D8→D13 reversal has no unit-level guard. CODE_REVIEW_D13.md finding 3 asked for a four-line test, and it was never added:

let cli = Cli::parse_from(["trim_galore", "--paired",
    "Sample_R1.fastq.gz", "Sample_R2.fastq.gz", "SAMPLE_R1.fastq.gz", "SAMPLE_R2.fastq.gz"]);
assert!(cli.validate().is_ok(), "case-only variants are distinct inputs");

D8 over-applied case-folding to the input-identity checks, which rejected two genuinely distinct files on a case-sensitive filesystem; D13 reversed it. The io.rs key-level tests pin the keys — but what broke was the wiring in cli.rs, and the only thing covering that today is the #216 CI validation job (.github/workflows/ci.yml:625-630). No files on disk, so the test runs identically on macOS and Linux.

The re-audit's third gap (norm_path's doc block not naming path_identity_key) is already tracked in #399, whose resolution removes that reader-facing surface anyway.

[FelixKrueger]

Fixed by #411 (squash-merged to dev as 2e00f32). Both items done, and the second turned out to be a finding rather than a task.

The CHANGELOG claim is corrected: it said "All four now use the same key as the pre-flight", which D13 falsified by splitting the key (and #389 refined the --passthrough half again). It now describes both keys and why they differ — case-folding is right for output paths, wrong for input identity.

The unit guard could not be written as specified. The four-liner from CODE_REVIEW_D13, copied verbatim into this issue, cannot run: validate() checks input existence, so the nonexistent paths never reach the identity check (it fails with "Input file not found"). A real-file version cannot run on macOS either — Sample_R1.fastq.gz and SAMPLE_R1.fastq.gz are one file on APFS, verified. So a test that had been sitting in the queue since #385 was never runnable by anyone.

It is therefore #[cfg(target_os = "linux")], creates four real files, and asserts the filesystem actually kept them distinct before trusting its own assertion — otherwise it would pass vacuously on any case-insensitive volume. Verified before merge by temporarily retargeting it to macOS, where it compiled and failed on exactly that premise check. It has now run for real on CI's ubuntu leg and passed, confirming the D8→D13 behaviour is correct.

Closing manually — merges into dev don't auto-close.