[minor] Add a test project

[matt-edmondson]

Fixes #96. Surfaced #97 along the way.

Why

The solution had one project and no tests. This tool deletes files — an off-by-one in grouping, or a mistake in the keep/delete decision, destroys user data, and nothing stood between a refactor and that outcome.

21 tests over Deduplicator, FileHasher and FileScanner.

The test that matters most

DryRunEquivalenceTests asserts that the set of files DryRun says it would delete is exactly the set Deduplicate does delete.

DryRun is the safety net users rely on before letting this tool remove anything, so a divergence between the two would be the worst defect this codebase could have: the preview would be a lie, and the thing it lied about is irreversible.

The two verbs share GroupByHash, FindDuplicates and SelectFileToKeep, so today they agree by construction. That is worth pinning rather than assuming — it holds only while both keep calling the same three methods and while SelectFileToKeep stays deterministic, and neither was enforced by anything.

It also asserts every group retains exactly one survivor, since deleting a whole group would lose the content entirely.

Real filesystem, not a fake

The tests drive throwaway directories under the temp path rather than an abstraction, because that is what the production code uses: FileScanner calls Directory.EnumerateFiles and Deduplicator calls File.Delete directly. Faking those would exercise a seam that does not exist and would not catch a mistake in the delete path — the part that destroys data. TempTree names each directory after the calling test, so a leaked directory names the test that leaked it.

Also covered

Grouping across nested directories; unique files never offered for deletion; empty files as duplicates of each other; the shortest-filename keep rule; that only the file name and not the path length decides it; deterministic tie-breaking regardless of input order; reclaimed byte counting; a second pass over a deduplicated tree being a no-op; scanning recursion plus the missing- and empty-directory cases; hash agreement across names; the empty-file SHA-256; and that parallel hashing matches single-file hashing.

Verification

21/21 pass. Mutation-checked, by substitution rather than deletion:

Mutation	Tests failed
SelectFileToKeep returns .Last() instead of .First()	3
DeleteDuplicates stops sparing the keeper (deletes every copy)	5

The tree was restored from a backup and re-verified after each run — git diff clean, no Last(), no && false residue.

One thing to know before merging

InternalsVisibleTo names the test assembly FileDeduplicator.Test, without the ktsu. prefix every other repository uses. That is not a typo: this repository has no AUTHORS.md, so ktsu.Sdk resolves an empty AuthorsNamespace and names the assembly FileDeduplicator rather than ktsu.FileDeduplicator. It is the only one of six repositories I checked that is missing the file.

I left it alone here — correcting it renames the published artifact, which does not belong in a test-only change. #97 has the detail, including the warning that fixing it must update this literal in the same commit or the test project silently loses access to internals. There is a comment at the attribute saying so.

Caveat

I could not run the repo's own analyzers locally: this sandbox's SDK is 10.0.111 (Roslyn 5.0.0.0) and ktsu.Sdk.Analyzers 2.28.0 requires 5.9.0.0, so CSC refuses it with CS9057. The .NET analyzers and EnforceCodeStyleInBuild did run and are clean. CI is the first place KTSU0001–0007 will actually see this — the new test project is the likeliest place for a KTSU0001 standard-package complaint.

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud