hash of URIRef is not the same across python runs

[drewp]

python -c 'import rdflib; print hash(rdflib.URIRef("hi"))'
13312079974552043
python -c 'import rdflib; print hash(rdflib.URIRef("hi"))'
13312079974020587

It would be nice if this was stable. The fix is in term.py, where we should perhaps use hash(type(self).name) instead of hash(type(self)) .

[joernhees]

i'm not entirely sure this is a must, but it's definitely a should...

i'll implement this with the full qualified name though: type(self).__module__ + '.' + type(self).__name__

also the current implementation of Literal.__hash__ uses hash(None) if no language is specified, which at least on my system isn't stable.

[joernhees]

@drewp: what's your actual use case for this? discussing this with @gromgull it turns out that python 3.3+ use randomized hashing in order to counter DoS attacks (e.g., http://lemire.me/blog/archives/2014/04/23/do-you-realize-that-you-are-using-random-hashing/ ). Hence fixing this from that perspective doesn't seem to make much sense at all.

I'm still somewhat for merging #501 as the current implementation doesn't do what it probably was intended to do: currently it hashes type(i) for rdflib.term.Identifiers, so essentially their memory location, instead of just their fully qualified class names. The change in #501 fixes that. For Python <3.3 it makes hash stable as the hash of a unicode string and for Python 3.3+ it will be as randomized as hash of a str.

Any other thoughts on this?

[drewp]

Use case: I was generating C code for an arduino, and to get a C variable name for a URIRef, I used 'pwm%s' % (hash(dev) % 99999) and hoped for the best. (And didn't think about negative numbers, apparently.) My system was rebuilding and reuploading on every run, since the generated C code didn't have the same checksum each time. Nothing was broken; it just ran slow. hash(str(dev)) is my workaround in python2.

Probably I should use hashlib.md5(dev).hexdigest() if I want a stable hash, or 'pwm_'+dev.encode('base64').strip('=\n') if I want to try something that's properly unique for each URI. Or re.sub(r'[^0-9a-zA-Z]', '_', dev) for something that's probably unique and also readable.

[rchateauneu]

Another use case is creating two rdflibURIRef("strign") from two strings with identical content. These two URIRef are not equal due to different hash values, as far as I can tell.