py-multiformats

Python bindings for the official Rust multiformats implementations, wrapped with PyO3:

• rust-multibase — self-describing base encodings
• rust-multihash — self-describing hashes
• rust-multiaddr — self-describing network addresses
• rust-cid — self-describing content identifiers

The package also embeds the canonical multicodec and multibase registries, compiled in at build time from the vendored tables in data/ and refreshed weekly from upstream.

Contents

• Why this package?
• Install
• Usage
• Development
  • Code generation
• License

Why this package?

The existing Python options are four single-format packages (py-multibase, py-multihash, py-multiaddr, py-cid) that have mostly gone quiet, and the pure-Python multiformats package. This one puts all four formats behind a single typed API and lets the Rust reference implementations do the actual work (the same code that runs inside rust-libp2p). Spec fixes arrive by bumping a dependency, parsing and hashing run at native speed, and the codec tables are generated from the canonical registries instead of being copied in once and left to rot. Ships as prebuilt abi3 wheels for CPython ≥ 3.10, no runtime dependencies.

Install

pip install py-multiformats

Usage

from multiformats import multibase, multicodec, multihash
from multiformats.cid import CID
from multiformats.multiaddr import Multiaddr
from multiformats.multihash import Multihash

# multicodec — the codec registry. Every entry is a module constant.
multicodec.DAG_PB                                    # 112 (0x70)
multicodec.code("dag-pb")                            # 112, name -> code
multicodec.name(multicodec.DAG_PB)                   # "dag-pb"
multicodec.tag(multicodec.DAG_PB)                    # "ipld" (accepts constant, code, or name)
multicodec.entries()                                 # [(name, tag, code, status), ...]

# multibase — constants hold the canonical encoding names
encoded = multibase.encode(multibase.BASE58BTC, b"hello")   # "zCn8eVZg"
base, data = multibase.decode(encoded)               # ("base58btc", b"hello")
base == multibase.BASE58BTC                          # True
multibase.bases()                                    # all supported encodings

# multihash
mh = multihash.digest(multicodec.SHA2_256, b"hello world")
mh = multihash.sha2_256(b"hello world")              # same, via convenience function
mh.code == multicodec.SHA2_256                       # True
mh.name                                              # "sha2-256"
mh.size                                              # 32
mh.digest                                            # raw digest bytes
Multihash.from_bytes(mh.to_bytes()) == mh            # True
multihash.codes()                                    # name -> code table

# CID
cid = CID.decode("QmdfTbBqBPQ7VNxZEYEj14VmRuZBkqFbiwReogJgS1zR1n")
cid.version                                          # 0
cid.codec == multicodec.DAG_PB                       # True
cid.codec_name                                       # "dag-pb"
cid.hash.name                                        # "sha2-256"
CID(1, multicodec.RAW, cid.hash)                     # codec by constant, code, or name
str(cid.to_v1())                                     # "bafybeihdwdce..."
cid.to_v1().encode(multibase.BASE64URL)              # any multibase encoding
CID.from_bytes(cid.to_bytes()) == cid                # True

# multiaddr
addr = Multiaddr("/ip4/127.0.0.1/tcp/4001")
list(addr)                                           # [("ip4", "127.0.0.1"), ("tcp", "4001")]
addr.protocols()                                     # ["ip4", "tcp"]
addr = addr.encapsulate("/p2p/QmcgpsyWgH8Y8ajJz1Cu72KnS5uo2Aa2LpzU7kinSupNKC")
addr.decapsulate("/tcp/4001")                        # Multiaddr("/ip4/127.0.0.1")
Multiaddr.from_bytes(addr.to_bytes()) == addr        # True

Anything that fails to parse, decode, or encode raises multiformats.MultiformatsError, a subclass of ValueError.

Development

You need a Rust toolchain and uv:

uv sync                                       # create the venv, install the dev tools, build the extension
uv run pytest                                 # run the test suite
uv run mypy tests/                            # type-check against the stubs
uv run maturin develop                        # rebuild the extension after Rust changes
cargo clippy --all-targets -- -D warnings     # lint the Rust side
cargo test                                    # run the Rust unit tests

The dev tools (maturin, pytest, mypy) are declared as a dependency group in pyproject.toml, so uv sync installs everything.

Code generation

The multicodec and multibase registries are not hand-written. The canonical tables are vendored in data/:

• data/multicodec-table.csv — verbatim copy of multiformats/multicodec table.csv
• data/multibase-table.csv — verbatim copy of multiformats/multibase multibase.csv

build.rs turns both CSVs into Rust at build time: the registry rows as a static ENTRIES table, phf perfect hash maps for the lookups, and a consts module with one constant per entry. The same entries become the Python constants (multicodec.DAG_PB, multibase.BASE58BTC, ...) at import time. build.rs also writes the multicodec.pyi and multibase.pyi stubs — those are committed and CI fails if they drift, so IDE autocomplete always matches the vendored tables.

To pull the latest tables from upstream:

./scripts/update-tables.sh

A scheduled workflow (update-tables.yml) runs the same script weekly and opens a pull request when a registry changed.

Releasing

Releases are tag-driven; nothing is built or uploaded from a laptop.

1. Bump version in pyproject.toml (PEP 440, e.g. 0.1.0a1 for an alpha, 0.1.0 for a final release) and mirror it in Cargo.toml (0.1.0-alpha.1 in semver). Commit and push.

2. Tag the commit and push the tag:

   git tag v0.1.0a1
   git push origin v0.1.0a1

3. The release workflow triggers on v* tags: it builds abi3 wheels for Linux (manylinux + musllinux, x86_64/aarch64), macOS (x86_64/arm64), and Windows (x64), builds the sdist, and publishes everything to PyPI.

Publishing uses trusted publishing (OIDC) — there are no PyPI tokens anywhere. One-time setup, already done for this repository: a (pending) publisher on PyPI pointing at probe-lab/py-multiformats, workflow release.yml, environment pypi, and a matching pypi environment in the GitHub repository settings.

License

Apache-2.0