[Python-Dev] The new SHA-3 is the old SHA-3

Christian Heimes christian at python.org
Mon Nov 4 20:58:30 CET 2013

Good news everybody!

A while ago John Kelsey has presented NIST's plans to change SHA-3 [1]
in order to make it faster but also weaker. Last Friday he posted a mail
on NIST's internal hash-forum mailing list. NIST is planing to drop
these plans and to move forward with the SHA-3 draft in its original
state. I can keep the sha3 module in Python 3.4 and don't have to remove
it. Maybe I'm going to add the new SHAKE functions to hashlib, too.

But first let me explain the background. NIST is going to standardize
four replacements for SHA2: SHA3-224, SHA3-256, SHA3-384 and SHA3-512.
(The number after the hyphen is the output size in bits.) These are
going to be standardized variants of the Keccak-f[1600] sponge
construction. [2]

Similar to a real sponge Keccak has a capacity, can absorb data at some
rate and can be squeezed to output data. The rate parameter affects
performance of the Keccak construction (higher rate == faster). A higher
capacity results in better collision resistance. Keccak-f[1600] rate +
capacity = 1600 bits with capacity/2 collision and pre-image resistance.
Contrary to a real sponge the squeeze function can return arbitrarily
amounts of data.

The old and new parameters for the SHA-3 are

  SHA3-224: capacity=448, output=224
  SHA3-256: capacity=512, output=256
  SHA3-384: capacity=768, output=384
  SHA3-512: capacity=1024, output=512

NIST was about to standardize SHA3-224/SHA3-256 with capacity=256 and
SHA3-384 /SHA3-512 with capacity=512. That's (most like) off the stove.

However NIST is going to standardize two additional functions: SHAKE128
and SHAKE256. These functions support arbitrary length output with a
collision resistance of 128/256 bits with a capacity of 256/512. The
SHAKE functions can be easily implemented on top of the current code and
PEP 247 API with mandatory length arguments for digst() and hexdigest().


[2] http://keccak.noekeon.org/

More information about the Python-Dev mailing list