[] should evaluate to an empty list, not any list

[tuukkamustonen]

>>> from voluptuous import Schema
>>> Schema({})({})
{}
>>> Schema({})({'foo': 1})
Traceback (most recent call last):
...
voluptuous.error.MultipleInvalid: extra keys not allowed @ data['foo']
>>> Schema(int)(5)
5
>>> Schema(2)(5)
Traceback (most recent call last):
...
voluptuous.error.MultipleInvalid: not a valid value
>>> Schema([])([])
[]
>>> Schema([])([1])  # shouldn't this throw error?
[1]

Wouldn't it be logical to evaluate list for any list and [] for an exact value (an empty list)?

[tusharmakkar08]

I also agree with you here. One problem which I could think is the backward compatibility loss in here.
@alecthomas What do you think ?

[alecthomas]

Yes, agreed. I don't think there is a significant issue with backwards compatibility, as even following the description of how [...] works you would expect this to be the case. That is, I think relying on [] to allow arbitrary list values is relying on undefined behaviour.

[tusharmakkar08]

@tuukkamustonen : Please submit the PR for the same. If you need any help, reach out to us.

[tuukkamustonen]

Should we also change {} to evaluate as is (an empty dict) always (even with extra=True)?

If one wants to define a free-form dict:

• With extra=False, he needs to use dict anyway.
• With extra=True, he can use dict instead of {}.

{...} would still be evaluated as nested schema and not as is.

That would break backwards-compatibility but would maybe make logic even more coherent (classes evaluated as types, instances evaluate as exact... except non-empty dict).

@alecthomas @tusharmakkar08 What do you think?

[tuukkamustonen]

Ping @alecthomas @tusharmakkar08

[tusharmakkar08]

Agree on this. It would make voluptuous more consistent.

[tusharmakkar08]

@tuukkamustonen : Are you working on PR for this issue?

[tuukkamustonen]

@tusharmakkar08: Not yet, but I'll take a look at it today...