Fix quadrature rule hash

[connorjward]

Also remove some uses of deprecated @abstractproperty and gives cells and point sets repr()s.

Necessary for firedrakeproject/firedrake#3989

[connorjward]

@ksagiyam please can you take a look at this too?

[ksagiyam]

This PR itself looks almost fine to me, but floats being used in some __repr__() could cause issues in computing cache key in firedrake.tsfc_interface.

[ksagiyam]

I do not think hash() necessarily does perfect hashing, so this definition could in theory go wrong if there was a hash collision. I think we should define __eq__() in child classes comparing sting-by-string, float-by-float, etc.

[connorjward]

I think hash() here is safe as we define __hash__() above. What I am doing is equivalent to what happens in finat.ufl.

If we made child classes overload __eq__() then we have some annoying details to handle (source):

If a class that overrides __eq__() needs to retain the implementation of __hash__() from a parent class, the interpreter must be told this explicitly by setting __hash__ = <ParentClass>.__hash__.

I don't know exactly what the standard thing to do here is but this at least works.

[ksagiyam]

As far as I understand, hash() is not something that guarantees equality. If we want to store fruits (say "orange", "apple", "banana"), the simplest hash keys would be len(name), and the hash table looks like:

0:
1:
2:
3:
4:
5: apple
6: orange -> banana

but orange is not banana. This is called hash collision, and this happens unless hashing is perfect. You might not have seen failures just because the __hash__() method you use is "almost" perfect.
orange and banana can be correctly distinguished by implementing __eq__() correctly.

This makes me think about our having been using md5 to generate cache keys, but maybe I misunderstand something about md5.

[ksagiyam]

I have to focus on my own work the rest of the day, so will come back next week.

[connorjward]

I changed it to

def __eq__(self, other):
        return type(other) is type(self) and repr(other) == repr(self)

anyway which might be better and avoids any confusion.

[connorjward]

but floats being used in some repr() could cause issues in computing cache key in firedrake.tsfc_interface.

An excellent catch. I will think on that.

[connorjward]

This PR itself looks almost fine to me, but floats being used in some repr() could cause issues in computing cache key in firedrake.tsfc_interface.

I have now had a stab at fixing this. UFL provides a format_float function that it uses for this purpose in constantvalue.py.