gh-124835: tomllib.loads: Raise TypeError not AttributeError. Improve message

[hukkin]

This PR makes tomllib.loads(b"") throw

TypeError: Expected str object, not 'bool'

instead of

TypeError: a bytes-like object is required, not 'str'

Given an input type that doesn't have the replace attribute, it makes tomllib.loads raise the same new TypeError, instead of an AttributeError. Should this be a separate PR though?

The corresponding Tomli PR is hukkin/tomli#229
The issue report is hukkin/tomli#223
Pinging @encukou as requested here hukkin/tomli#223 (comment)

• Issue: Improve invalid input type handling in tomllib.loads #124835

[picnixz]

Two stances:

• I'm ok with the change but I would let a regular TypeError be propagated instead (e.g., if someone is using a subclass os str for whatever reason).
• I'm not ok with the change because of the "garbage in, garbage out" principle, though it's not really an in/out here. I think loads should only expect a string.

For the second point, we could do something like isinstance(s, str) but this would restrict the user to always use a string instance and not a UserString object for instance.

I don't know which one is better. But the fact that the message does not help is IMO an inconvenience so I like that you want to fix it. Now, how to fix it, is still something I'm pondering.

[picnixz]

I suggest opening a CPython issue so that we can possibly discuss it. In addition, I think a NEWS entry should be added so that people know about the fact that an AttributeError won't be raised (but for now I'll just add the skip news label)

[JelleZijlstra]

Yes, this should have a NEWS entry (and therefore, an issue) as it changes user-visible behavior.

[hukkin]

Thanks! I've opened an issue (#124835) and added a NEWS entry.

[hauntsaninja]

Great, thank you!