Report that NamedTuple and dataclass are incompatile instead of crashing.

[tyralla]

Fixes #18527

The fix is pretty simple. I could not find a situation where combining NamedTuple and dataclass makes sense, so emitting an error and just not applying the dataclass transformations seems sensible.

[sobolevn]

While we are at it: what about TypedDict?

[tyralla]

dataclass is also not really compatible with TypedDict, but here, things work a little differently, and Mypy's understanding seems to be consistent with Python's runtime behaviour:

from dataclasses import dataclass
from typing import TypedDict

@dataclass
class A(TypedDict):
    i: int

a = A(i=1)
assert a["i"] == 1
a.i
# Mypy -> error: "A" has not attribute "i" [attr-defined]
# Python -> AttributeError: 'dict' object has no attribute 'i'

[sterliakov]

First I was excited when I saw that issue: "wow, we can make a dataclass with convenient unpacking, nice!". However, it really doesn't work at runtime:

from dataclasses import dataclass
from typing import NamedTuple

@dataclass
class Foo(NamedTuple):
    x: int
    y: str

Foo(0, '')

results in

Traceback (most recent call last):
  File "/tmp/a.py", line 9, in <module>
    Foo(0, '')
  File "<string>", line 3, in __init__
AttributeError: can't set attribute

So this is probably the most reasonable fix. Can we expand the message or add some note about runtime incompatibility? It really isn't immediately obvious why not use that combination.

[tyralla]

So this is probably the most reasonable fix. Can we expand the message or add some note about runtime incompatibility? It really isn't immediately obvious why not use that combination.

Maybe "A NamedTuple cannot be a dataclass (AttributeError during initialization)."?

[sobolevn]

🚀

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[sobolevn]

Thank you!