tests/std_filter_tests/ has not been compiled since October 2025

[leynos]

Summary

tests/std_filter_tests/ contains 66 test functions across 13 files that
have not been compiled since 2025-10-26. They are tracked in git, they look
maintained, and they run nowhere.

Evidence

The directory is absent from the test-binary list:

$ cargo nextest list --list-type binaries-only | grep -i filter
$ # (no output)

There is no [[test]] stanza for it in Cargo.toml, no tests/std_filter_tests.rs
entry file, and no other file under tests/ declares those modules.

Root cause

Commit b2d46d6 ("Tightens lint policy, documents API, and hardens runtime
paths", #222, 2025-10-26) renamed the entry file into the directory it points at:

tests/{std_filter_tests.rs => std_filter_tests/mod.rs} | 2 ++

Two things broke at once, and neither produced a diagnostic:

1. The test binary disappeared. Cargo builds one integration-test binary per
   tests/*.rs. Moving the only top-level entry file into a subdirectory left
   nothing for Cargo to build, so the suite stopped existing rather than
   starting to fail.

2. Every #[path] prefix became wrong. mod.rs still declares its
   submodules relative to the old location:

   #[path = "std_filter_tests/collection_filters.rs"]
   mod collection_filters;

   From inside tests/std_filter_tests/, that resolves to
   tests/std_filter_tests/std_filter_tests/collection_filters.rs, which does
   not exist. Were the binary restored without fixing these, it would fail to
   compile.

This is the failure mode where a rename is silently correct to the compiler
because the thing it broke is no longer built at all.

Coverage lost

File	Area
collection_filters.rs	collection filters
command_filters/grep_filter_tests.rs	grep filter
command_filters/shell_filter_tests.rs	shell filter
command_filters/windows_filter_tests.rs	Windows shell bypass, streaming, metacharacters
hash_filters.rs	hash filters
io_filters.rs	I/O filters
network_functions.rs	network functions
path_filters.rs	expanduser, realpath, relative_to, with_suffix
which_filter_tests.rs, which_filter_common.rs	which resolver

Some of this is covered elsewhere — expanduser has a BDD scenario, and the
which resolver has stdlib_which_tests — but the overlap has not been
assessed. Assessing it is part of the work: restoring 66 tests that duplicate
existing coverage is not automatically worth it, and knowing which are unique is
the point.

Why it matters now

This surfaced while addressing a review finding on #505. Codex reported an
unannotated std::env::var_os("PATH") in
tests/std_filter_tests/command_filters/windows_filter_tests.rs, reasoning that
"a Windows make lint compiles the module with -D warnings and rejects the
existing lookup". That reasoning is sound but the premise is not: no make lint compiles that file, on any platform, because Cargo never builds it. The
annotation added in #505 is correct in substance and harmless, but it does not
fix a live failure.

Related: #518 proposes Windows CI. That would not have caught this either — the
gap is Cargo target wiring, not platform coverage.

Suggested work

1. Restore the target: move mod.rs back to tests/std_filter_tests.rs, or add
   an explicit [[test]] stanza with path.
2. Fix the #[path] prefixes, which are wrong under either arrangement.
3. Expect compile failures and test failures. Nine months of drift against code
   that has moved considerably — EnvLock, EnvVarGuard, the which resolver
   refactor in Refactor which resolver: OS-specific workspace fallback #241, and the localization work in Externalize Fluent localization; Windows test context refactor + audit #260 all landed since.
4. Assess overlap with existing coverage before repairing each file. Delete what
   is genuinely redundant rather than resurrecting it out of completeness.
5. Add a guard so this cannot recur silently. A test asserting the expected set
   of test binaries, or a check that every tests/*/mod.rs has a corresponding
   target, would have caught it the day it broke.

Acceptance criteria

• cargo nextest list --list-type binaries-only includes the suite.
• Every retained file compiles and passes.
• Files whose coverage is genuinely duplicated elsewhere are deleted, with
  the duplication named in the commit message.
• A guard exists against an orphaned test tree recurring unnoticed.
• The #[expect(clippy::disallowed_methods)] added to
  windows_command_setup in Enforce the environment mandate with clippy disallowed-methods (#504) #505 is re-checked once the file is actually
  compiled.

[coderabbitai]

🔗 Related PRs

leynos/whitaker#235 - Implement rstest helper call-site collection (8.2.2) [merged]
#410 - Adopt the Whitaker Dylint suite in the lint gate and CI [merged]
leynos/rstest-bdd#583 - Restore the coverage ratchet (#582) [merged]
#478 - Narrow no_std_fs_operations exemptions with Whitaker module-path suppressions (#462) [open]
leynos/mdtablefix#428 - Adopt camino/cap-std paths across integration tests (#418) [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!