Skip to content


Folders and files

Last commit message
Last commit date

Latest commit



21 Commits

Repository files navigation


This library is intended to be an exact as possible python translation of the TweetNaCl minimal elliptic curve cryptography library. Interested parties are invited to compare the contents of pure_pynacl/ and tweetnacl/TweetNaCl.c. tweetnacl/TweetNaCl.c has been carefully and deliberately modified from the original in the first several commits of this repo, mostly to remove the unnecessary preprocessor statements.


The original goal of pure_pynacl was to provide a limited but sufficient crypto library for arcane or obscure platforms, for which porting, adopting, or maintaining dependencies would be challenging. Using pypy could make execution slightly faster, however, pypy is not bytecode-equivalent to CPython or any other python runtime. The bytecode incompatibility of pypy with CPython and also the inability of translating down to C for RPython circumscribes its potential as a passive optimizer for expensive parts of large programs that cannot feasibly platform on pypy. (Evidently doing this is anathema to the project, which seems to have more rigorous than convenience goals.) If python (CPython) has been ported to an exotic or unusual platform, or even if it could be, porting pypy will likely be at least as challenging as porting a standard crypto library like {open|libre}ssl. Any performance gains from using pure_pynacl combined with the convenience of a complete python source are likely not going to exceed the benefit of porting a standard crypto library to the platform.

There are critical junctures in history where momentous conventions are set by often unwitting actors. The creation of C and the subsequent popularity and open nature of the C/Unix ecosystem is one of them. Supposing some other generic language family, lisp, smalltalk, ocaml, eiffel, erlang, forth, or any purported language most excellent for systems programming were better than C, it would still be irrelevant. Macroeconomic inertia dictates that C be the universal language of platforms and the volume and depth of its mindshare could not be diverted by expense of fortune or force of decree. Any platform that wants more than irrelevance will market itself with a C-compatible operating system or driver and associated libraries. Thus, any python project wanting to port to such an obscure platform (for which CPython has already been ported), would likely be wiser to port a C-based crypto lib than run pure_pynacl even as a fallback.

For non-obscure platforms, python has very good CFFI with CTypes and there are several python crypto libraries that use this internally if not SWIG.

Finally, NaCl, the superset of TweetNaCl, does not even provide the complete set of crypto features expected from a modern crypto library. Among other caveats throughout the docs: "On the contrary: the [public key] crypto_box function guarantees repudiability [emph. original]." Scraping standard crypto library output with import subprocess would still likely be better than import pure_pynacl.

What remains then is the crypto learning and intellectual exercise I enjoyed creating this library, the C-compatible types, and multidimensional random data factories used for testing. You are welcome to use this project however you desire as governed by the license. GPLv[latest] is my default, but Apache 2.0 was requested and I am fine keeping it that way.


tox run -e ALL


python -m build --build-wheel


From PyPI

pip install pure_pynacl

From source

git clone

Using pure_pynacl

Consult the NaCl documentation for complete details of correct usage.

import pure_pynacl as pynacl
result = pynacl.crypto_sign_ed25519_tweet(sm, smlen, m, n, sk)

Comparison tests

Unit tests that verify the parity of TweetNaCl and pure_pynacl are located in a file called tests/ This file also contains tests that compare the parity of C bitwise operations and bitwise operations on the Int type defined in These tests can be run with the following command.

$ tox run -e unit

Currently, the parity testing is only parallel, meaning that for each function in the libraries, identical input data is supplied and the output data is expected to be identical.

Parallel testing


     |             |
     v             v
    fcn           fcn
     |             |
     v             v
    out     ==    out

Antiparallel testing

Antiparallel testing is much less symmetric and only covers the external API as the implementation of the internal utilities are not necessarily invertible. Input data is sent through each function in one library and then through the corresponding counter function in the other library such that the output data is identical to the input data.

          in                     in
          |                      |
          v                      v
     TweetNacl.fcn       TweetNacl.inv_fcn
          |                      |
          v                      v
 pure_pynacl.inv_fcn      pure_pynacl.fcn
          |                      |
          v                      v
         out == in              out == in

          in                     in
          |                      |
          v                      v
   pure_pynacl.fcn      pure_pynacl.inv_fcn
          |                      |
          v                      v
  TweetNacl.inv_fcn         TweetNacl.fcn
          |                      |
          v                      v
         out == in              out == in

C99 bit shifting operations

Right-shifts of negative signed numbers are implementation-defined. All of the other cases that are not explicitly undefined are specified.

From C99

The integer promotions are performed on each of the operands. The type of the result is that of the promoted left operand. If the value of the right operand is negative or is greater than or equal to the width of the promoted left operand, the behavior is undefined.

The result of E1 << E2 is E1 left-shifted E2 bit positions; vacated bits are filled with zeros. If E1 has an unsigned type, the value of the result is E1 x 2^E2, reduced modulo one more than the maximum value representable in the result type. If E1 has a signed type and nonnegative value, and E1 x 2^E2 is representable in the result type, then that is the resulting value; otherwise, the behavior is undefined.

The result of E1 >> E2 is E1 right-shifted E2 bit positions. If E1 has an unsigned type or if E1 has a signed type and a nonnegative value, the value of the result is the integral part of the quotient of E1 / 2^E2. If E1 has a signed type and a negative value, the resulting value is implementation-defined.

a << b and a >> b

  1. a and b are promoted before operation; the result has the type of promoted(a)

  2. if b < 0 a << b and a >> b are undefined

  3. if b > len(a) a << b and a >> b are undefined

  4. if a < 0 a << b and a >> b are undefined

a << b 4. if a >= 0 and b >= 0 if a not signed a << b == a*2**b%bits(a) elif a is signed if a*2**b in type(a) a << b == a*2**b else a << b is undefined

a >> b 5. if a >= 0 and b >= 0 a >> b == a//2**b

ctypes multidimensional pointers


A pure python implementation of TweetNaCl.







No packages published