fix: stdlib dataclass converted into pydantic dataclass still equal its stdlib dataclass equivalent

[PrettyWood]

Change Summary

Related issue number

closes #2162

Checklist

• Unit tests for the changes exist
• Tests pass on CI and coverage remains at 100%
• Documentation reflects the changes where applicable
• changes/<pull request or issue id>-<github username>.md file added describing change
  (see changes/README.md for details)

[codecov]

Codecov Report

Merging #2182 (b177fb4) into master (de0657e) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master     #2182   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           21        21           
  Lines         4121      4125    +4     
  Branches       829       830    +1     
=========================================
+ Hits          4121      4125    +4     

Impacted Files	Coverage Δ
pydantic/dataclasses.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update de0657e...b177fb4. Read the comment docs.

[samuelcolvin]

Humm, my only reservation is that standard python dataclasses to clever stuff so that different dataclasses with the same fields (and values) are not equal:

from dataclasses import dataclass

@dataclass(frozen=True)
class Foo:
    a: int
    b: float

@dataclass(frozen=True)
class Bar:
    a: int
    b: float

f = Foo(a=10, b=1.0)
b = Bar(a=10, b=1.0)
debug(f == b)  #> False

f2 = Foo(a=10, b=1.0)
debug(f == f2)  #> True

Isn't there a risk that different dataclasses incorrectly being equal is worse than them incorrectly being unequal?

I know "correct" and "incorrect" are subjective here and some might argue (very reasonably) that f and b are the same above, but I think since dataclasses are a python thing we should do our best to match python behaviour.

If we can't reliably match python's standard behaviour, I'd be happy to just raise an error some people don't assume stuff will work which won't.

[samuelcolvin]

I think it shouldn't be that hard to copy python's own logic more or less verabim:

https://github.com/python/cpython/blob/1e5d33e9b9b8631b36f061103a30208b206fd03a/Lib/dataclasses.py#L949-L955

[jtrh]

I think it shouldn't be that hard to copy python's own logic more or less verabim:

https://github.com/python/cpython/blob/1e5d33e9b9b8631b36f061103a30208b206fd03a/Lib/dataclasses.py#L949-L955

I agree that copying Python's own logic is a good idea, but how would you handle the class equality check? Read below for details.

Near the end of that code, Python calls _cmp_fn() to create the comparison function, which starts by checking for class equality (if other.__class__ is self.__class__:) before even taking a look at the fields themselves.

@samuelcolvin Do you think that replacing the class equality check with something like if other.__class__ is self.__class__ or other.__class__ is self.__class__.__bases__[0]: would be reliable enough?

[samuelcolvin]

@samuelcolvin Do you think that replacing the class equality check with something like if other.__class__ is self.__class__ or other.__class__ is self.__class__.__bases__[0]: would be reliable enough?

Yes, I think makes sense.

[PrettyWood]

FYI I thought again about all the issues related to stdlib dataclass to validated dataclass
I started a draft on implementation in #2557.
Feedback more than welcome! 🙏

[PrettyWood]

Will be tackled in #2557