fix(hardlink)!: use (device, inode) pair as deduplication key

[Copilot]

• Add (device, inode) pair as deduplication key
• Add test proving device number is used in deduplication key
• Fix Reflection doc comment: inode numbers are NOT guaranteed unique (only (dev, ino) pairs are)
• Add deterministic tie-breaker sort by (ino, dev) before dropping dev in From<HardlinkList>
• Fix misleading comment: duplicate inode from JSON is rejected (DuplicatedInode), not silently merged

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[KSXGitHub]

@copilot Very cool to point this out. But can you check whether you actually use the device number to in deduplication?

[github-actions]

Performance Regression Reports

commit: fe98f44

There are no regressions.

[Copilot]

@copilot Very cool to point this out. But can you check whether you actually use the device number to in deduplication?

Yes, the device number is used. HardlinkList is keyed on InodeKey { dev: u64, ino: InodeNumber } (a private struct in hardlink_list.rs), so two files with the same inode number but different device numbers hash to different map entries and are never merged.

Added a dedicated test same_ino_on_different_devices_are_treated_separately (commit 38aacab) that inserts four files — two sharing (dev=1, ino=100) and two sharing (dev=2, ino=100) — and asserts the list has exactly 2 entries with 2 paths each, confirming cross-device isolation.

[Copilot]

Pull request overview

This PR fixes hardlink deduplication across multiple filesystems by switching the internal key from inode-only to a (device, inode) pair, preventing false merges when scanning multiple roots on different devices while keeping the public API and JSON schema unchanged.

Changes:

• Introduce an internal InodeKey { dev, ino } and key the HardlinkList map by it.
• Thread dev through hardlink recording and update iterators accordingly.
• Keep reflection/JSON ino-only (with a documented limitation) and update tests to cover same-inode/different-device behavior.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/hardlink/hardlink_list.rs	Adds InodeKey and rekeys the internal DashMap; updates add() to accept dev.
src/hardlink/hardlink_list/iter.rs	Updates iterator key types to InodeKey; Item::ino() returns key.ino.
src/hardlink/hardlink_list/reflection.rs	Strips dev when producing reflection; uses dev=0 placeholder on load; updates docs.
src/hardlink/aware.rs	Extracts dev from metadata and passes it into HardlinkList::add().
src/hardlink/hardlink_list/test.rs	Updates test table/call sites for dev and adds a cross-device same-inode test.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[KSXGitHub]

@copilot The following reviews:

• fix(hardlink)!: use (device, inode) pair as deduplication key #347 (comment).
• fix(hardlink)!: use (device, inode) pair as deduplication key #347 (comment).
• fix(hardlink)!: use (device, inode) pair as deduplication key #347 (comment).

Do you agree with these reviews?

If you do, implement them.

If you don't, provide a reason why you disagree and skip it.

[Copilot]

@copilot The following reviews:

• fix(hardlink)!: use (device, inode) pair as deduplication key #347 (comment).
• <fix(hardlink)!: use (device, inode) pair as deduplication key #347 (comment)...

All three reviews are correct and have been implemented in commit eb92bdc:

1. Doc comment — updated to state that only (device, inode) pairs are unique, not inode numbers alone, and explicitly notes the multi-filesystem limitation.

2. Deterministic sort tie-breaker — From<HardlinkList> now sorts by (ino, dev) before stripping dev, so entries that share an inode number across filesystems get a stable, reproducible order in JSON output.

3. Misleading "incorrectly merged" comment — corrected to accurately describe the actual behaviour: such duplicates cause ConversionError::DuplicatedInode and are rejected, not silently merged.