pacquet/crates/workspace-projects-filter/src/glob.rs:21

• In is_match, only the pattern is normalized to forward slashes. On Windows, PathBuf::to_string_lossy() typically yields \ separators for project root dirs, so glob dir filtering will fail to match even when the logical paths line up. Normalize the candidate string too (and keep the normalization buffer alive while splitting).

This issue also appears on line 13 of the same file.

pub fn is_match(candidate: &str, pattern: &str) -> bool {
    let pattern = pattern.replace('\\', "/");
    let pattern = pattern.strip_suffix('/').unwrap_or(&pattern);
    let candidate = candidate.strip_suffix('/').unwrap_or(candidate);

    let pattern_segments: Vec<&str> = pattern.split('/').collect();
    let candidate_segments: Vec<&str> = candidate.split('/').collect();
    match_segments(&pattern_segments, &candidate_segments)
}

pacquet/crates/workspace-projects-filter/src/parse_project_selector.rs:36

• Doc comment has a duplicated word: "directory directory-selectors".

This issue also appears on line 35 of the same file.

/// Parse one raw `--filter` selector string against `prefix` (the
/// directory directory-selectors resolve relative to).

pacquet/crates/workspace-projects-graph/src/create_projects_graph.rs:204

• resolve_directory only does an exact lexical-normalized lookup. Upstream createProjectsGraph has an additional slow-path scan to handle case mismatches on case-insensitive filesystems (so link:../B can still match a workspace project rooted at ../b). Without a similar fallback, inter-project edges may be missed on Windows/macOS when dependency path casing differs from the actual directory casing.

This issue also appears on line 198 of the same file.

/// Resolve a local-path dependency to a sibling by directory: join the
/// path onto the importer's root, normalize, and look it up in the
/// by-directory index.
fn resolve_directory(importer: usize, path: &str, lookups: &Lookups) -> Option<PathBuf> {
    let resolved = lexical_normalize(&lookups.node_keys[importer].join(path));
    lookups.by_dir.get(&resolved).map(|&index| lookups.node_keys[index].clone())
}

pacquet/crates/workspace-projects-filter/src/filter.rs:143

• filter_graph clones adjacency lists on every traversal (node.dependencies.clone() and graph.get(id).cloned()), which can become a noticeable overhead for large workspaces because each visited node allocates and copies its edge list. Consider changing pick_subgraph/adj to borrow adjacency slices (e.g., Fn(&Path) -> Option<&[PathBuf]>) so traversal can walk without cloning.

This issue also appears on line 137 of the same file.

    let forward = |id: &Path| projects_graph.get(id).map(|node| node.dependencies.clone());
    let reversed_graph = selectors
        .iter()
        .any(|selector| selector.include_dependents)
        .then(|| reverse_graph(projects_graph));
    let reverse = |id: &Path| reversed_graph.as_ref().and_then(|graph| graph.get(id).cloned());

pacquet/crates/workspace-projects-filter/src/glob.rs:21

• In is_match, only the pattern is normalized to forward slashes. On Windows, PathBuf::to_string_lossy() typically yields \ separators for project root dirs, so glob dir filtering will fail to match even when the logical paths line up. Normalize the candidate string too (and keep the normalization buffer alive while splitting).

This issue also appears on line 13 of the same file.

pub fn is_match(candidate: &str, pattern: &str) -> bool {
    let pattern = pattern.replace('\\', "/");
    let pattern = pattern.strip_suffix('/').unwrap_or(&pattern);
    let candidate = candidate.strip_suffix('/').unwrap_or(candidate);

    let pattern_segments: Vec<&str> = pattern.split('/').collect();
    let candidate_segments: Vec<&str> = candidate.split('/').collect();
    match_segments(&pattern_segments, &candidate_segments)
}

pacquet/crates/workspace-projects-filter/src/parse_project_selector.rs:36

• Doc comment has a duplicated word: "directory directory-selectors".

This issue also appears on line 35 of the same file.

/// Parse one raw `--filter` selector string against `prefix` (the
/// directory directory-selectors resolve relative to).

pacquet/crates/workspace-projects-graph/src/create_projects_graph.rs:204

• resolve_directory only does an exact lexical-normalized lookup. Upstream createProjectsGraph has an additional slow-path scan to handle case mismatches on case-insensitive filesystems (so link:../B can still match a workspace project rooted at ../b). Without a similar fallback, inter-project edges may be missed on Windows/macOS when dependency path casing differs from the actual directory casing.

This issue also appears on line 198 of the same file.

/// Resolve a local-path dependency to a sibling by directory: join the
/// path onto the importer's root, normalize, and look it up in the
/// by-directory index.
fn resolve_directory(importer: usize, path: &str, lookups: &Lookups) -> Option<PathBuf> {
    let resolved = lexical_normalize(&lookups.node_keys[importer].join(path));
    lookups.by_dir.get(&resolved).map(|&index| lookups.node_keys[index].clone())
}

pacquet/crates/workspace-projects-filter/src/filter.rs:143

• filter_graph clones adjacency lists on every traversal (node.dependencies.clone() and graph.get(id).cloned()), which can become a noticeable overhead for large workspaces because each visited node allocates and copies its edge list. Consider changing pick_subgraph/adj to borrow adjacency slices (e.g., Fn(&Path) -> Option<&[PathBuf]>) so traversal can walk without cloning.

This issue also appears on line 137 of the same file.

    let forward = |id: &Path| projects_graph.get(id).map(|node| node.dependencies.clone());
    let reversed_graph = selectors
        .iter()
        .any(|selector| selector.include_dependents)
        .then(|| reverse_graph(projects_graph));
    let reverse = |id: &Path| reversed_graph.as_ref().and_then(|graph| graph.get(id).cloned());

pacquet/plans/TEST_PORTING.md (2)

266-266: ⚡ Quick win

Clarify the directional reference to hoist stubs.

The phrase "the two known_failures hoist stubs below stay" is confusing because the referenced hoist stubs are in the "Hoisting" section (lines 123-124), which appears above this new section in the file, not below.

📝 Suggested clarification

-selected projects is still a follow-up (the install fan-out is
-unfiltered, so the two `known_failures` hoist stubs below stay).
+selected projects is still a follow-up (the install fan-out is
+unfiltered, so the two `known_failures` hoist stubs in the "Hoisting"
+section remain blocked).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@pacquet/plans/TEST_PORTING.md` at line 266, The sentence referencing "the two
`known_failures` hoist stubs below stay" is misleading because the Hoisting
section (which contains those `known_failures` stubs) is above this paragraph;
update the sentence in TEST_PORTING.md to explicitly reference the Hoisting
section or change "below" to "above" (e.g., "the two `known_failures` hoist
stubs in the 'Hoisting' section remain") so readers can locate the
`known_failures` hoist stubs unambiguously.

---

300-302: 💤 Low value

Make stubbed test documentation consistent.

The first stubbed test includes the test name in parentheses (git_diff_selection_unimplemented), but the next two only say "Stubbed." For consistency and discoverability, either include the stub name for all three or omit it for all three.

📝 Suggested fix for consistency

If the stub names differ, include them:

-- [x] `TypeScript repo: workspace/projects-filter/test/index.ts:348` `select changed packages`. Stubbed (`git_diff_selection_unimplemented`).
-- [x] `TypeScript repo: workspace/projects-filter/test/index.ts:480` `select changed packages when operating under a git worktree`. Stubbed.
-- [x] `TypeScript repo: workspace/projects-filter/test/index.ts:553` `selection should fail when diffing to a branch that does not exist`. Stubbed.
++ [x] `TypeScript repo: workspace/projects-filter/test/index.ts:348` `select changed packages`. Stubbed (`git_diff_selection_unimplemented`).
++ [x] `TypeScript repo: workspace/projects-filter/test/index.ts:480` `select changed packages when operating under a git worktree`. Stubbed (`git_diff_selection_unimplemented`).
++ [x] `TypeScript repo: workspace/projects-filter/test/index.ts:553` `selection should fail when diffing to a branch that does not exist`. Stubbed (`git_diff_selection_unimplemented`).

Or, if they all use the same stub and the rejection path is already documented on line 296, omit the parenthetical entirely:

-- [x] `TypeScript repo: workspace/projects-filter/test/index.ts:348` `select changed packages`. Stubbed (`git_diff_selection_unimplemented`).
++ [x] `TypeScript repo: workspace/projects-filter/test/index.ts:348` `select changed packages`. Stubbed.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@pacquet/plans/TEST_PORTING.md` around lines 300 - 302, The three checklist
items referencing tests "select changed packages"
(workspace/projects-filter/test/index.ts:348), "select changed packages when
operating under a git worktree" (index.ts:480), and "selection should fail when
diffing to a branch that does not exist" (index.ts:553) are inconsistent: the
first shows the stub name (git_diff_selection_unimplemented) in parentheses
while the other two just say "Stubbed." Make them consistent by either adding
the stub name "(git_diff_selection_unimplemented)" to the two latter entries or
removing the parenthetical from the first entry so all three simply read
"Stubbed."; update the three checklist lines accordingly and ensure the exact
stub identifier git_diff_selection_unimplemented is used if you choose to
include names.

pacquet/crates/cli/src/cli_args.rs (1)

150-151: 💤 Low value

Consider moving instead of cloning.

filter and filter_prod are owned values (moved out at line 112) and are only used once here. Direct assignment would avoid the clone:

-                    cfg.filter.clone_from(&filter);
-                    cfg.filter_prod.clone_from(&filter_prod);
+                    cfg.filter = filter;
+                    cfg.filter_prod = filter_prod;

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@pacquet/crates/cli/src/cli_args.rs` around lines 150 - 151, The two lines use
clone_from on owned values; replace these clones by moving the owned values into
the config fields instead of cloning—i.e., stop calling
cfg.filter.clone_from(&filter) and cfg.filter_prod.clone_from(&filter_prod) and
assign the owned filter and filter_prod into cfg.filter and cfg.filter_prod
respectively so the ownership is moved rather than cloned.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes