Guidance on exception handling for validators?

[poneill]

Hi, I had a question about best practices for handling exceptions with attr validators.

Suppose I wanted to represent a float in the unit interval with runtime validation. The natural way to do that, afaict, would be:

import attr

@attr.s
class UnitFloat:
    """Represent a float in the unit interval."""

    x: float = attr.field(
        validator=[
            attr.validators.instance_of(float),
            attr.validators.ge(0.0),
            attr.validators.le(1.0),
        ],
    )

If I call with:

UnitFloat(1.1) I get: ValueError: 'x' must be <= 1.0: 1.1, which makes sense.

If I call with UnitFloat(1), however, I get: TypeError: ("'x' must be <class 'float'> (got 1 that is a <class 'int'>)..."

This is in accordance with the python docs (https://docs.python.org/3/library/exceptions.html) but requires the user to guard against two kinds of exceptions, depending on how exactly the constructor choked on bad input. This means that code of the form:

try:
    x = UnitFloat(unsanitized_input)
except ValueError:
    print("validation error safely caught!")

is unsafe. Is there an established convention for this pattern, like maybe:

AttrValidationError = (TypeError, ValueError)?

(There is also the option of adding a converter to each field, which would trivially reduce all bad input to ValueError, but this is probably not always desired behavior.)

Thanks!

(attrs version: '23.1.0')

[hynek]

This is in accordance with the python docs (https://docs.python.org/3/library/exceptions.html) but requires the user to guard against two kinds of exceptions, depending on how exactly the constructor choked on bad input.

I think this is very much on purpose, such that the user can differentiate between what went wrong.

---

Generally speaking, attrs doesn't try to be a validation / structuring package (we've got cattrs for that!) but a toolkit to build classes that fit into the Python eco system.

IOW a class that is built using attrs should behave as if someone has written it by hand, adhering to Python standards.

All this to say: there's no attrs-specific guidance to be had, because we're trying to be good Python citizens. But all your suggested approaches are entirely valid, just like with any other Python class.

[poneill]

Thanks, that makes plenty of sense!