[oscarbenjamin]

Fixes gh-438

[deftasparagusanaconda]

is this a bug?

from pickle import loads, dumps
from flint import arb

original = arb(2).sqrt()
roundtrip = loads(dumps(original))

print(original)
print(roundtrip)
print(original == roundtrip) # False
print(original.mid() == roundtrip.mid()) # True
print(original.rad() == roundtrip.rad()) # True

expected behaviour: original == roundtrip returns True

---

python: 3.15.2t
python-flint: up to date with 3333d2f

[oscarbenjamin]

That's expected behaviour for arb:

In [1]: from flint import arb

In [2]: a1 = arb(2).sqrt()

In [3]: a2 = arb(2).sqrt()

In [4]: a1
Out[4]: [1.41421356237309 +/- 5.15e-15]

In [5]: a2
Out[5]: [1.41421356237309 +/- 5.15e-15]

In [6]: a1 == a2
Out[6]: False

Conceptually an arb represents an exact number that is only known approximately. Here a1 and a2 are each known to be in the same interval but we don't know if they represent the same number in that interval. Perhaps == should give None rather than False to make this clearer but it is not supposed to be True.

[deftasparagusanaconda]

i think instead of == returning None (which would violate python interfaces), the class shouldn't have an __eq__ method at all

[oscarbenjamin]

It isn't possible to to not have an __eq__ method because if a class does not define __eq__ then it inherits it from object which has its own quirks:

In [14]: class A: pass

In [15]: a1 = A()

In [16]: a2 = A()

In [17]: a1 == a1
Out[17]: True

In [18]: a1 == a2
Out[18]: False