Summary

Sub-slice 4d of #438. Closes out Slice 4 by adding the prev_graph diff — the input Slice 5's linker will subtract from graph to identify orphaned directories that need removing.

What lands

lockfile_to_hoisted_dep_graph now takes an optional current_lockfile: Option<&Lockfile> and runs a second walk over it (when present and non-empty) with force: true, skipped: BTreeSet::new() to populate result.prev_graph. Ports upstream's wrapper at installing/deps-restorer/src/lockfileToHoistedDepGraph.ts:70-79.

let prev_graph = match current_lockfile {
    Some(current) if current.packages.is_some() => {
        let prev_opts = LockfileToHoistedDepGraphOptions {
            force: true,
            skipped: BTreeSet::new(),
            ..opts.clone()
        };
        Some(build_dep_graph(current, &prev_opts)?.graph)
    }
    _ => None,
};

Why force + empty skipped

Both overrides matter — neither is paranoia:

• force: true — an orphan that landed under the previous install but would now fail installability (host platform changed, dependency was downgraded out of engine range, etc.) still has a directory on disk. Without force, the prev-walk would silently filter it out and the linker would leave the stale dir in place. Pinned by prev_graph_includes_orphan_even_when_now_incompatible.
• skipped: empty — the previous install's filter set says which packages it didn't install, which is unrelated to which directories currently exist. Inheriting the wanted-walk's skipped would underreport prev_graph for unchanged-but-now-incompatible packages.

API change

lockfile_to_hoisted_dep_graph gains a middle current_lockfile: Option<&Lockfile> argument. The previous body splits into a private build_dep_graph helper that the public wrapper calls once or twice depending on whether a current lockfile is present. Existing tests updated (the only callers — None for the no-prev case).

prev_graph shape

Upstream uses prevGraph = {} for both null-current cases; pacquet uses None. The linker treats them the same (no orphans to remove on a fresh install).

Tests

Four new prev_graph tests; the 15 walker tests from 4a-4c carry over with the new signature.

• prev_graph_none_when_current_lockfile_absent — no current lockfile → None, wanted graph still produced.
• prev_graph_none_when_current_lockfile_has_no_packages — current lockfile with packages: None → None.
• prev_graph_contains_orphan_from_current_only_lockfile — package in current but not wanted appears in prev_graph, not in graph.
• prev_graph_includes_orphan_even_when_now_incompatible — darwin-only orphan in current, host is linux, wanted is empty: prev_graph still contains it (proves force: true works), and the wanted-walk's skipped stays empty (proves the prev-walk's skipped set is independent).

Test plan

• All 19 walker tests pass (15 carry-over + 4 new).
• just ready (874 → 878 tests, all green).
• cargo doc --document-private-items, taplo format --check, just dylint clean.
• Sabotage check: changing the prev-walk's options from force: true, skipped: empty to ..opts.clone() re-fails prev_graph_includes_orphan_even_when_now_incompatible (orphan disappears from prev_graph because it would now fail installability).

What's next

Slice 4 is done. Next on the critical path: Slice 5 — linkHoistedModules. Consumes LockfileToDepGraphResult to produce the on-disk tree: walks prev_graph minus graph to rimraf orphans, then imports each graph node from the CAS via Slice 1's import_indexed_dir primitive, then runs linkBins per node_modules/ directory.

Written by an agent (Claude Code, claude-opus-4-7).

Summary by CodeRabbit

• Improvements

  • Added support for computing a previous dependency graph when a current lockfile is provided, improving detection of changes between lockfile states and handling of current-only (orphan) packages.

• Tests

  • Expanded tests for lockfile-comparison behavior, ensuring prev-graph semantics and orphaned package detection (including installability edge cases).