Writing profile to disk fails because hashlib does not support usedforsecurity keyword

[Martin-Rey]

Hi all,

I am trying to write a pynbody.profile.Profile object, using its build-in method write():

pynbody/pynbody/analysis/profile.py

Line 481 in 6b0ded6

def write(self):

which fails with the following trace:

Traceback (most recent call last):
  File "/Users/martinrey/Documents/stellar_halos/plot_surface_brightness.py", line 70, in <module>
    profile2d.write()
  File "/opt/anaconda3/lib/python3.7/site-packages/pynbody-0.48-py3.7-macosx-10.9-x86_64.egg/pynbody/analysis/profile.py", line 551, in write
    filename = self._generate_hash_filename()
  File "/opt/anaconda3/lib/python3.7/site-packages/pynbody-0.48-py3.7-macosx-10.9-x86_64.egg/pynbody/analysis/profile.py", line 522, in _generate_hash_filename
    (hashlib.md5(self._x, usedforsecurity=False)).hexdigest()
TypeError: openssl_md5() takes no keyword arguments

Thanks to https://stackoverflow.com/questions/54717862/how-do-i-know-if-the-usedforsecurity-flag-is-supported-by-hashlib-md5 and sphinx-doc/sphinx#7611, I have tracked this to the fact that my platform (Mac OS) does not support the usedforsecurity keyword used in the hashing here:

pynbody/pynbody/analysis/profile.py

Lines 468 to 477 in 6b0ded6

	def _generate_hash_filename(self):
	"""Create a filename for the saved profile from a hash using the binning data"""
	import hashlib
	try:
	filename = self.sim.base.filename + '.profile.' + \
	(hashlib.md5(self._x, usedforsecurity=False)).hexdigest()
	except AttributeError:
	filename = self.sim.filename + '.profile.' + \
	(hashlib.md5(self._x, usedforsecurity=False)).hexdigest()

From python, hash lib is indeed missing the keyword

In [1]: import hashlib, inspect                                                                                                        

In [2]: inspect.getfullargspec(hashlib.new)                                                                                            
Out[2]: FullArgSpec(args=['name', 'data'], varargs=None, varkw='kwargs', defaults=(b'',), kwonlyargs=[], kwonlydefaults=None, annotations={})

for the following platform configuration

In [3]: import sys, platform, pynbody, numpy; print("%s\r\n%s\r\npynbody %s\r\nnumpy %s\r\n"%(sys.version, platform.platform(), pynbody
   ...: .__version__, numpy.__version__))                                                                                              
3.7.4 (default, Aug 13 2019, 15:17:50) 
[Clang 4.0.1 (tags/RELEASE_401/final)]
Darwin-19.6.0-x86_64-i386-64bit
pynbody 0.47
numpy 1.17.2

From what I have understood, the keyword usedforsecurity is unsafe because not supported across all platforms. But my understanding of hashing algorithms is fairly limited (to say the least) and I don't understand whether it is necessary in this context.

Hope that helps,
Martin

[apontzen]

Thanks, are you able to do a pull request and perhaps even a test? This is a feature I've never actually used.

[Martin-Rey]

Yes, I can look and figure out a fix to this. As for a test, I will try but this seems to arise from differences in backend implementations on different operating systems, and I am unsure how to capture this in the Travis build. Will have a go anyway.

[Martin-Rey]

Solved and merged in #599