Please support deterministic output (PROV-N)

[olebole]

With Python 3 (3.6, 3.7), the output with PROV-N (at least) is not deterministic:

import prov.model as prov

d = prov.ProvDocument()
d.set_default_namespace('https://example.com')
d.entity('id', [(prov.PROV_TYPE, 'foo'), (prov.PROV_TYPE, 'bar')])
print(d.get_provn())

Gives sometimes

document
  default <https://example.com>
  
  entity(id, [prov:type="foo", prov:type="bar"])
endDocument

and sometimes

document
  default <https://example.com>
  
  entity(id, [prov:type="bar", prov:type="foo"])
endDocument

This makes it difficult to create doctests here.

[trungdong]

Hi @olebole,

Attributes are stored internally in a set of values. I think this is why the values do not follow a deterministic order when listed. I can see why it is an issue for your testing, but I wonder if you could have the same test done in a different way. The two documents that you provided above are equivalent.

[olebole]

Could that be wrapped by sorted()? Having a deterministic output would greatly improve doctests. Currently, I write

'''Example:

>>> import prov.model as prov
>>> d = prov.ProvDocument()
>>> d.set_default_namespace('https://example.com')
>>> d.entity('id', [(prov.PROV_TYPE, 'foo'), (prov.PROV_TYPE, 'bar')])
>>> print(d.get_provn())
document
  default <https://example.com>
  
  entity(id, [prov:type="foo", prov:type="bar"])
endDocument
'''

which is nicely readable as tutorial for the user, and would be a good doctest. Putting the expected output into a string, parsing it, and comparing with the input doc would probably work (right?), but is then not usable as documentation anymore.
If two documents are equivalent (and created in the same manner), shouldn't they have the same serialization?