Custom scalars using numpy arrays fail is_nullish() test

[thomascobb]

I have a custom scalar type for a numpy array:

def serialize_ndarray(value):
    return dict(
        numberType=value.dtype.name.upper(),
        base64=base64.b64encode(value).decode()
    )

ndarray_type = GraphQLScalarType("NDArray", serialize=serialize_ndarray)

is_nullish() does an inequality test on values to work out if they are NaN:

def is_nullish(value: Any) -> bool:
    """Return true if a value is null, undefined, or NaN."""
    return value is None or value is INVALID or value != value

Unfortunately, numpy arrays break this because they override __eq__ to return a pointwise equality array rather than a boolean.

The comment for is_nullish() suggests that value != value is checking for NaN. If this is the case we could replace it with:

def is_nan(value):
    """Return true if a value is NaN"""
    try:
        return math.isnan(value)
    except TypeError:
        return False

def is_nullish(value: Any) -> bool:
    """Return true if a value is null, undefined, or NaN."""
    return value is None or value is INVALID or is_nan(value)

Would this be acceptable? I can provide a PR if it is.

[Cito]

Right, this value != value is a bit obscure - it has been taken over 1:1 from JavaScript. Not sure why they test it that way, probably because isNaN() was only added to the language in ES 6.

I'm not against changing this and using math.isnan instead. However, I'd rather avoid creating an exception and do something like ... or (isinstance(value, float) and isnan(value)) instead.

We should also add a test that is_nullish(value) is False for some custom object with value != value , that would simulate the numpy array. Feel free to send a PR, otherwise I'll care about this.

[thomascobb]

Ok, I think I'm happy with the PR, let me know if any changes are required for it

[Cito]

Nice, thank you! 👍