Introduction

Git stores objects in two physical formats. Loose objects are individual files; packed objects live inside a pack file with a delta-compressed body. Both forms describe the same logical objects, addressed by the same hashes.

Loose layout

Each loose object is at .git/objects/xx/yyy..., where xx is the first two hex chars of the SHA and yyy... is the remaining 38. The file is the zlib-deflated bytes of <type> <size>\0<content>:

find .git/objects -type f -name '??' -prune -o -type f -print | head
git cat-file -p <sha>

Pack layout

.git/objects/pack/pack-<hash>.pack
.git/objects/pack/pack-<hash>.idx

The .idx file maps SHA to byte offset within the .pack; without it, the pack would require a linear scan.

How objects move between forms

New objects are written loose. They become packed when:

• git gc runs (manual or automatic).
• git repack is invoked.
• A fetch or push transfers them as a pack.
• git fast-import writes them directly into a pack.

git gc
git repack -a -d
git count-objects -v

count-objects -v reports loose count, total disk usage, and pack stats.

Trade-offs

• Loose: simple, easy to inspect, fast to write a single object. Slow when there are millions.
• Packed: compact, delta-compressed, fast to bulk-read. Requires repacking to incorporate new objects.

Auto-gc

Git triggers an auto-gc when too many loose objects accumulate. Tunables:

git config gc.auto 6700           # threshold for loose objects
git config gc.autoPackLimit 50    # threshold for number of packs
git config gc.auto 0              # disable auto-gc (not recommended)

Inspecting a pack

git verify-pack -v .git/objects/pack/pack-<hash>.idx | head
git verify-pack -s .git/objects/pack/pack-<hash>.idx

You will see object types, packed sizes, deltas, and chain depths.

Recovering from corruption

git fsck --full
cp -r broken-repo backup
git unpack-objects < .git/objects/pack/pack-<hash>.pack

unpack-objects explodes a pack back to loose form, useful when a single bad object inside a pack is taking down many operations.

Looking up an object

When Git resolves a SHA, it searches in this order: loose object file, then each pack's idx, then alternates (objects/info/alternates). The lookup is essentially constant-time per pack thanks to the idx's fanout table. The performance impact of having many small packs vs one big pack is real, which is why gc consolidates:

git count-objects -v
cat .git/objects/info/alternates 2>/dev/null
git verify-pack -s .git/objects/pack/pack-<hash>.idx

Alternates let multiple repos share a pool of objects on disk; git clone --shared creates such a setup and saves space at the cost of some safety.

Common mistakes

Manually deleting pack- files to reclaim space; that destroys reachable history. Use git gc. Disabling auto-gc on long-lived repos and ending up with millions of loose objects, slowing every operation. Running git gc --aggressive regularly when normal gc would do; aggressive packing is expensive. Finally, copying a repo with cp -r while a pack is being written, ending up with a corrupt index. Use git clone --local for safe copies.