Week 7 - M1: Finalize packfiles + storage performance baseline #44

• Owner: @mahis1067 · minigit/objects.py, tests/test_objects.py
• Docs: docs/packfiles.md
• Depends on: Week 5 - M1: Packfile compression support #31, Week 6 - M1: Object-store fixes for push/pull round-trips #38
• Integration: coordinate the shared performance fixture with Week 7 - M4: Network diagnostics + end-to-end performance baseline #47
• Branch from updated main: feat/packfile-finalization
• Week 7 starts the roadmap's Weeks 7-8 optimization + polish phase; finish the linked prerequisites first
• Keep the object API, object hashes, and module ownership unchanged

Steps

1. Finish the packfile and transfer-hardening work; check the implementation against docs/packfiles.md and make the document describe the actual format and packing entry point.
2. Close remaining gaps in packed-only, loose-only, and mixed lookup using fresh store instances.
   • Loose lookup first; corrupt loose objects must not silently fall back to a packed copy
   • Missing -> ObjectNotFoundError; corrupt -> ObjectCorruptError
   • has_object and duplicate writes must preserve those distinctions
3. Finalize validation of pack/index versions, counts, offsets, record boundaries, types, lengths, compression, and hashes. Published packs must preserve the exact bytes of blobs, trees, and commits.
4. Verify temporary-file cleanup and pack-before-index publication. Interrupted publication must leave earlier objects readable; incomplete/unpublished packs must not become visible to readers.
5. With M4, measure packing time, object-read time, and stored bytes on the shared moderately large fixture.
   • Record fixture size, object counts, environment, and commands
   • Compare loose and packed storage separately; report retained loose bytes separately from pack/index bytes
   • Include fresh-store reads so a warm in-memory cache does not hide the initial cost
6. Profile repeated lookups and fix a measured storage bottleneck if one is found. Any index cache must discover newly published packs and preserve corruption checks.
7. Add a small regression for each fix; reuse existing pack and transfer tests. Leave pruning, delta compression, and format redesign outside this issue.
8. Document reproduction commands and remaining performance findings for Week 8 in the PR; scripts/quality-check.sh -> linked PR -> teammate review.

Tests

• Fresh-instance reads of blobs, nested trees, and two-parent commits from multiple packs
• Packed-only and mixed stores; duplicate requests and overlapping objects across packs
• Empty/binary/Unicode payloads retain the original type, bytes, and hash
• Invalid index/record, truncated data, bad zlib, wrong hash, failed publication
• Existing store instance can read a newly published pack; old objects remain readable
• Shared performance fixture plus a small correctness case in the normal test suite; no wall-clock CI threshold

Done when

• Packfile implementation and format/usage documentation agree
• Packed and mixed reads preserve validation and publication guarantees
• Reproducible storage baseline recorded; any optimization has before/after evidence
• Regression and integration checks pass
• Quality check green; PR open; review requested