tests(vcs/file_types): all-scope test claims non-binary/non-symlink filtering but fixture has neither

[dekobon]

Summary

all_scope_reproduces_the_whole_tree_ranking asserts that --file-types all ranks "every tracked, non-binary, non-symlink file", but the shared fixture contains neither a binary blob nor a symlink, so the production binary/symlink skip is never exercised — a regression in it would pass green.

Location

• tests/vcs_file_types.rs:31-40 (fixture()), tests/vcs_file_types.rs:75-85 (all_scope_reproduces_the_whole_tree_ranking)

Evidence

The fixture writes only plain UTF-8 text files:

const TRACKED_FILES: &[&str] = &[
    "src/lib.rs", "app/main.py", "README.md",
    "Cargo.toml", "Cargo.lock", "Makefile",
];

fn fixture() -> Repo {
    let repo = Repo::init();
    for rel in TRACKED_FILES {
        repo.write(rel, "one\ntwo\nthree\n");
    }
    ...
}

The all test then asserts:

"`all` ranks every tracked, non-binary, non-symlink file"

But the binary/symlink skip lives in production at src/vcs/git/repo.rs:152, independent of FileTypeScope:

if entry_mode.is_blob() && !entry_mode.is_link() {
    ...
}

Since the fixture has zero binary blobs and zero symlinks, that guard is never triggered. The module doc (lines 2-3) and the assertion message both advertise a binary/symlink contract the suite does not verify.

Expected Behavior

A test whose message claims "non-binary, non-symlink" filtering should include at least one symlink and one binary blob in the tracked set and assert they are excluded under all (the broadest scope, where only the binary/symlink guard — not the extension filter — can drop them).

Actual Behavior

No binary or symlink entry exists in the fixture. Deleting the !entry_mode.is_link() guard (ranking symlinks) or breaking the is_blob() binary handling would leave every test in tests/vcs_file_types.rs green.

Impact

The --file-types extension scoping (issue #576, the suite's primary subject) is well covered and genuinely discriminating. The adjacent enumeration guard at repo.rs:152 — which the all-scope test's own message claims to cover — is untested here, so a silent regression in symlink/binary handling under any scope would not be caught by this suite. Risk is mild (the extension filter is the file's focus) but the assertion message overstates what is verified.

[dekobon]

General Plan for Fix

Root Cause

The all_scope_reproduces_the_whole_tree_ranking test (and the file's module doc) advertise a "non-binary, non-symlink" contract, but the shared fixture() writes only plain text files. The production guard entry_mode.is_blob() && !entry_mode.is_link() (src/vcs/git/repo.rs:152) therefore never fires, so a regression in symlink or binary handling is invisible to this suite.

Implementation Steps

1. Extend the fixture (or add a dedicated fixture variant for the all test) to commit:
   • one symlink (e.g. link.rs -> src/lib.rs), created via std::os::unix::fs::symlink then git add; gate the symlink portion behind #[cfg(unix)] since Windows symlink creation needs elevated privileges.
   • one binary blob (e.g. blob.bin containing NUL bytes), to exercise the is_blob() line-count path.
2. In the all test, assert these two entries are absent from the ranked set (they must be dropped by the repo.rs:152 guard, not by the extension filter — all admits every extension, so only that guard can exclude them). Keep TRACKED_FILES as the asserted-present set.
3. Do not add the symlink/binary to TRACKED_FILES (which other tests treat as the full analyzable/whole-tree baseline); add them as separate fixture entries so the metrics/custom expectations are unchanged.

Tests to Add/Update

• all_scope_reproduces_the_whole_tree_ranking: add symlink + binary fixture entries, assert they are excluded while all TRACKED_FILES remain ranked.
• Optionally a focused symlink_and_binary_excluded_under_every_scope test to pin the guard independently of the extension filter.

Verification

• cargo test -p big-code-analysis --features vcs-git vcs_file_types passes
• Test-via-revert: deleting !entry_mode.is_link() from repo.rs:152 makes the new symlink assertion fail; restoring it passes
• cargo clippy -p big-code-analysis --all-features clean
• Manual verification: the new fixture entries do not perturb the metrics/custom scope expectations