Fix dataclass_transform decorator for pydantic dataclasses

[cdce8p]

Remove kw_only_default and add additional field_specifier for stdlib dataclass fields.

This change should be backported to 1.10.X since the next mypy release will break typing otherwise.

Ref https://peps.python.org/pep-0681/#dataclass-transform-parameters
Fixes #5117

[samuelcolvin]

Not sure if this will conflict with #5009 which is a WIP implementation of dataclasses for V2 which I'm intending to complete this week?

[cdce8p]

Not sure if this will conflict with #5009 which is a WIP implementation of dataclasses for V2 which I'm intending to complete this week?

Just looked at it, as long as the public interface isn't changed (i.e. dataclasses.field and positional arguments still works) it shouldn't conflict. It might even be slightly helpful if you add a pydantic specific field method. That could / should be added to dataclass_transform then as well.

If you really want though, I could also change the target to 1.10.X.

[samuelcolvin]

otherwise LGTM.

[samuelcolvin]

otherwise LGTM

[cdce8p]

due to import time problems in the past, we avoid doing module level import of stuff unless it's absolutely necessary.

I guess if we remove dataclass from __init__.py, maybe this would be fine, but we'll need to change this or on v1.10.

dataclasses is already imported in other modules, e.g. networks.py. So it won't add any additional import time.

pydantic/pydantic/networks.py

Line 3 in 6ab4207

import dataclasses as _dataclasses

Should the implementation keep the dataclass_transform for runtime introspection?

As the import time is no longer a concern, I reverted the change. With that dataclass_transform will be accessible at runtime again.

[dmontagu]

Closing in favor of #5111

[dmontagu]

Oh I realize this is for main, not 1.10.X, so I'll reopen it.

[cdce8p]

I think it would make sense to move forward with this PR as well now that the changes have already been merged into 1.10.X. (A backport is obviously no longer necessary). /CC: @dmontagu

[dmontagu]

Thanks @cdce8p!