blake3 0.2.1

Python bindings for the Rust blake3 crate

blake3-py

Python bindings for the official Rust implementation of BLAKE3, based on PyO3. These bindings expose all the features of BLAKE3, including extendable output, keying, and multithreading. The basic API matches that of Python's standard hashlib module.

Examples

from blake3 import blake3

# Hash some input all at once. The input can be bytes, a bytearray, or a memoryview.
hash1 = blake3(b"foobarbaz").digest()

# Hash the same input incrementally.
hasher = blake3()
hasher.update(b"foo")
hasher.update(b"bar")
hasher.update(b"baz")
hash2 = hasher.digest()
assert hash1 == hash2

# Hexadecimal output.
print("The hash of 'hello world' is", blake3(b"hello world").hexdigest())

# Use the keyed hashing mode, which takes a 32-byte key.
zero_key = b"\0" * 32
message = b"a message to authenticate"
mac = blake3(message, key=zero_key).digest()

# Use the key derivation mode, which takes a context string. Context strings
# should be hardcoded, globally unique, and application-specific.
context = "blake3-py 2020-03-04 11:13:10 example context"
key_material = b"some super secret key material"
derived_key = blake3(key_material, derive_key_context=context).digest()

# Extendable output. The default digest size is 32 bytes.
extended = blake3(b"foo").digest(length=100)
assert extended[:32] == blake3(b"foo").digest()
assert extended[75:100] == blake3(b"foo").digest(length=25, seek=75)

# Hash a large input with multithreading. Note that this can be slower
# for short inputs, and you should benchmark it for your use case on
# your platform. As a rule of thumb, don't use multithreading for inputs
# shorter than 1 MB.
large_input = bytearray(1_000_000)
hash3 = blake3(large_input, multithreading=True).digest()

# Copy a hasher that has already accepted some input.
hasher1 = blake3(b"foo")
hasher2 = hasher1.copy()
hasher1.update(b"bar")
hasher2.update(b"baz")
assert hasher1.digest() == blake3(b"foobar").digest()
assert hasher2.digest() == blake3(b"foobaz").digest()

Installation

pip install blake3

As usual with Pip, you might need to use sudo or the --user flag with the command above, depending on how you installed Python on your system.

There are binary wheels available on PyPI for most environments. But if you're building the source distribution, or if a binary wheel isn't available for your environment, you'll need to install the Rust toolchain.

Thread Safety and the GIL

Like the hashlib functions in the Python standard library, we release the GIL while hashing, to avoid blocking other threads for a potentially long time. However, this allows race conditions: it's possible for other threads to access a hasher or an input buffer while hashing is going on. This is worse than an ordinary Python race condition. It's undefined behavior in the C/C++/Rust sense. But this seems to be the standard way to do hashing in Python. In any case, it should be rare for real world programs to share a hasher between threads. For more details about this issue, see the comments on usafe code in lib.rs.