Using and operator vs fuzzy_and while querying assumptions

[skirpichev]

The problem is that builtin and operator is not symmetric wrt non-boolean args, esp. None:

In [1]: from sympy.core.logic import fuzzy_and

In [2]: None and False

In [3]: False and None
Out[3]: False

In [4]: fuzzy_and([None, False])
Out[4]: False

In [5]: fuzzy_and([False, None])
Out[5]: False

That's important if we consider assumptions. We have a lot of statements like self.is_integer and other.is_integer in the code (see, for example, some cases in the core). It seems, this is a wrong pattern and we should use fuzzy_and instead.

Thanks to @smichr who brings my attention to this in #8490.

[asmeurer]

Ditto with or and not. This is another case where the new assumptions are cleaner.

[skirpichev]

with or and not

Yes,

>>> None or False
False
>>> False or None
>>> 

and this is a problem too:

>>> not None
True

In the end, for old assumptions - we can't use python's boolean operations at all.

[pelegm]

Where can I find information about the "new" assumptions?

[skirpichev]

sympy/assumptions

[smichr]

self.is_integer and other.is_integer

It's ok to use and for the positive assertion. It's just the returning of such an expression to a routine that uses the negative result for an inference that causes problem. The same goes for if foo or bar: that's ok but if not (foo or bar) will give False when a None might be more correct.

[skirpichev]

Yes, but we still have errors. I hope I was careful enough for And, but we have Or and Not cases too...

[skirpichev]

No activity. I believe, this is fixed in the diofant, so it's not a problem for me. Closed.

[smichr]

Although this may not be a problem for diofant it should be checked in SymPy to see if these issue still exist.