Bug report

Bug description:

I have many dataclasses created on demand by dataclasses.make_dataclass. 3.14 changes behaviour forinstances of the dataclass.

In 3.13 and before, theinstances have a__annotations__ attribute with the expected contents.

In 3.14, that attribute no longer exists, but callinginstance.__annotate_func__ does return what was in__annotations__ before.

The3.14 howto does say "you may access theannotations data member manually"

importdataclassesimportinspectdc=dataclasses.make_dataclass("foo", (("one",int), ("two",str), ("three",complex)))value=dc(1,'hello',3+4j)print(f"{dir(value)=}")try:print(inspect.get_annotations(value))exceptExceptionasexc:print("inspect.get_annotations exception: ",type(exc),exc)ifhasattr(dc,"__annotate_func__"):print(f"{value.__annotate_func__()=}")

For Python 3.13:

dir(value)=['__annotations__', '__class__', '__dataclass_fields__', '__dataclass_params__ ....

For Python 3.14:

dir(value)=['__annotate_func__', '__annotations_cache__',  '__class__', '__dataclass_fields__' ....

In both casesinspect.get_annotations() does give TypeError.

The release notes cover annotations, but nothing about changes to dataclasses instances.

Perhaps the annotations attributes shouldn't be on instances, but they have been. I also see__annotate_func__ on instances of regular objects from Python classes.

CPython versions tested on:

3.14, 3.13

Operating systems tested on:

Linux

Linked PRs

• gh-134370 Added clarification on instances annotations #134387
• [3.14] gh-134370: Added clarification on instance annotations (GH-134387) #134509