meridian

Meridian is an offline, cryptographically verifiable delta synchronization engine for directory trees. It turns any folder into a content-addressed repository, creates immutable snapshots, and exports compact .mrd bundles containing only the chunks that changed.

It is useful when normal sync is impossible or undesirable: air-gapped machines, removable media, slow links, reproducible deployments, deterministic backups, or audited file transfer.

What makes it different

• True delta bundles: unchanged chunks are not copied.
• Offline application: move a single .mrd file by USB, email, object storage, or any transport.
• Merkle-style integrity: every chunk, snapshot, and bundle is SHA-256 verified.
• Atomic updates: a destination is staged and swapped only after successful reconstruction.
• Conflict protection: a delta applies only to the exact base snapshot it was built from.
• Full bundles: optionally produce a self-contained snapshot that can replace an unrelated directory.
• Content deduplication: repeated chunks are stored once in the local object store.
• No runtime dependencies: Python 3.11+ standard library only.
• Library and CLI: automate it from Python or use the meridian command.

Installation

A prebuilt wheel is included, so installation does not need a compiler or build backend:

python -m venv .venv
# Linux/macOS
source .venv/bin/activate
python -m pip install dist/meridian_sync-1.0.0-py3-none-any.whl

# Windows PowerShell
.venv\Scripts\Activate.ps1
python -m pip install dist\meridian_sync-1.0.0-py3-none-any.whl

To install directly from source instead, run python -m pip install ..

Five-minute workflow

Create a source repository and its first snapshot:

mkdir project
cd project
meridian init
# add or copy your files
meridian snapshot -m v1

After changing files, create the next snapshot:

meridian snapshot -m v2
meridian diff v1 v2

Build a compact delta bundle:

meridian bundle v1 v2 -o ../v1-to-v2.mrd

Prepare a replica at v1 once (or use an existing matching directory):

meridian restore v1 ../replica

Apply the bundle atomically:

meridian apply ../v1-to-v2.mrd ../replica

Validate a bundle before transferring or applying it:

meridian inspect ../v1-to-v2.mrd

Full bundle for unrelated or empty destinations

A normal delta omits chunks already present in the base snapshot. For a self-contained bundle:

meridian bundle v1 v2 -o ../v2-full.mrd --full
meridian apply ../v2-full.mrd ../some-directory --allow-unrelated

This replaces the destination atomically with the exact target tree.

Ignore rules

meridian init creates .meridianignore. Rules are intentionally familiar:

*.tmp
build/
.cache/
!important.tmp

The .meridian repository directory is always excluded. .meridianignore itself is included so that snapshot semantics remain visible and reproducible.

CLI reference

meridian init [PATH] [--chunk-size AVERAGE_BYTES]
meridian snapshot [PATH] [-m LABEL]
meridian snapshots [PATH] [--json]
meridian diff OLD NEW [--repo PATH] [--json]
meridian bundle BASE TARGET -o FILE [--repo PATH] [--full]
meridian inspect BUNDLE [--json]
meridian apply BUNDLE DESTINATION [--allow-unrelated]
meridian verify SNAPSHOT [--repo PATH]
meridian restore SNAPSHOT DESTINATION [--repo PATH] [--overwrite]
meridian gc [PATH]

Snapshot references may be a full ID, an unambiguous ID prefix, or an exact label.

Python API

from pathlib import Path
from meridian_sync import Repository, create_bundle, apply_bundle

repo = Repository.init(Path("data"))
v1 = repo.create_snapshot(label="before")

# mutate files under data/
v2 = repo.create_snapshot(label="after")

create_bundle(repo, v1.id, v2.id, Path("update.mrd"))
repo.restore(v1.id, Path("replica"))
apply_bundle(Path("update.mrd"), Path("replica"))

Data model

A snapshot stores sorted entries for files, directories, and symlinks. File contents are split with deterministic content-defined chunking, so insertions and deletions preserve surrounding chunks instead of shifting every later block. Chunk IDs are SHA-256 hashes of raw bytes. The snapshot root is a canonical hash over paths, types, permissions, sizes, symlink targets, and chunk IDs.

A delta bundle contains:

• bundle.json — format metadata and root hashes;
• base.json — the complete base snapshot manifest;
• target.json — the complete target snapshot manifest;
• objects/<sha256> — only target chunks absent from the base snapshot.

During application, Meridian verifies the current destination against base.json, recovers unchanged chunks directly from the destination, reads changed chunks from the bundle, reconstructs the target in a staging directory, and atomically swaps it into place.

Operational notes

• Permissions are preserved where the operating system allows it.
• Symlinks are stored as symlinks and are never followed while scanning.
• Sockets, devices, and FIFOs are rejected rather than silently mishandled.
• Files that change during scanning abort the snapshot.
• Bundle archives reject unexpected members and malformed object names.
• Applying a normal delta to the wrong base fails without modifying the destination.

Testing

python -m unittest discover -s tests -v

License

MIT

Исходный код · MIT