[Summary: Both algorithms derive from the SHA2 specification. There's no copyright infringement here.]
To make sure I understand, the project uses modified code, where the original code (at https://github.com/thomdixon/pysha2/blob/master/sha2/sha256....) is under the MIT license, but the modified version (at https://github.com/bkkcoins/klondike/blob/master/utils/sha25... ) is under the GPLv2-or-later, and doesn't attribute the original source.
To answer your question, if the copyright holder believes that someone at github is infringing on copyright then the easiest thing to do is to have the copyright holder send in a DCMA notice. See https://help.github.com/articles/dmca-takedown .
That said, I don't agree that there is copyright infringement. The argument is that:
> The only real differences is that the derived code has been made a bit more compact and less readable, the capitalization of the class has changed, a method name and a few field names were modified, and a little bit of the message/data plumbing was changed. Otherwise the core computational aspects, down to the order of operations, destructuring assignment, variable names, and auxiliary routines are all the same—all things that are hard to duplicate coincidentally.
However, if you look at http://en.wikipedia.org/wiki/SHA-2#Pseudocode you'll see that it also has the same core computational aspects, variable names, etc. These all come from a common source - the SHA specification. Take a look at http://csrc.nist.gov/groups/STM/cavp/documents/shs/sha256-38... and compare.
What this means is the different implementers used the same reference, and not that one implementer infringed the other's copyright. While it's true that 'all things that are hard to duplicate coincidentally', there's no coincidence here.
Now, remove those parts which are based directly on the standard. Those parts are quite different. Compare:
def digest(self):
mdi = self._counter & 0x3F
length = struct.pack('!Q', self._counter<<3)
if mdi < 56:
padlen = 55-mdi
else:
padlen = 119-mdi
r = self.copy()
r.update('\x80'+('\x00'*padlen)+length)
return ''.join([struct.pack('!L', i)
for i in r._h[:self._output_size]])
-to-
def finalize(self):
tailbytes = self._length & 0x3f
if tailbytes < 56: padding = 55 - tailbytes
else: padding = 119 - tailbytes
self.update(b"\x80" + (b"\0" * padding) +
struct.pack("!Q", self._length << 3))
The 0x3f is another way to say "mod 64". It's something that a C programmer would think is faster, but profiling in Python just now shows that they have identical performance.
Overall these implementations for how to pad is what you would expect from the algorithm.
The last part of the implementations, however, are quite different. The first makes a copy() then reaches into the copy's internals. This means people can continue to update the hash even after the digest was generated. This matches the API from Python's hashlib.
The latter uses an in-place modification of the hash, and the name 'finalize' reflects that it's a different implementation. It modifies state in-place, and requires a get_bytes() to get the actual bytes. This doesn't match the Python hashlib API at all.
What I see is that the parts derive from the NIST specification are similar, though not identical, and the parts which don't follow naturally from that specification are quite different.
Therefore I disagree with you, and say that there is no infringement.