`extra !=` does not work as expected

[konstin]

#6476 has recently added (https://github.com/feast-dev/feast/pull/6476/changes#diff-50c86b7ed8ac2cf95bd48334961bf0530cdc77b5a56f852c5c61b89d735fd711L26) the following lines to pyproject.toml:

feast/pyproject.toml

Lines 26 to 27 in f296d4b

	"pyarrow>=21.0.0; extra != 'flink'",
	"pyarrow>=16.1.0,<21.0.0; extra == 'flink'",

The extra != 'flink'" of this does not work as expected: Neither pip or uv support extra != markers, with uv first evaluating feast on its own without the base package and pip always evaluating it to true. As we're in the process of formally removing extra != due to being broken by design (https://discuss.python.org/t/ban-negative-extras-for-extras-marker/108022), we found this one usage and wanted give you a heads up about this feature being broken.

When you try to [uv] pip install feast[flink], it currently downgrades to installing 0.63.0, to avoid the unsatisfiable 0.64.0.

CC @XuananLe

[konstin]

For context, the tool output:

$ uv pip install --no-build "feast[flink]"
Resolved 68 packages in 94ms
Prepared 33 packages in 799ms
Installed 68 packages in 30ms
[...]
 + feast==0.63.0
[...]
warning: The package `feast==0.63.0` does not have an extra named `flink`

$ uv pip install --no-build "feast[flink]==0.64.0"
  × No solution found when resolving dependencies:
  ╰─▶ Because feast[flink]==0.64.0 depends on pyarrow>=16.1.0,<21.0.0 and feast==0.64.0 depends on pyarrow{python_version < '0'}>=21.0.0, we can conclude that
      feast==0.64.0 and feast[flink]==0.64.0 are incompatible.
      And because you require feast[flink]==0.64.0, we can conclude that your requirements are unsatisfiable.

$ pip install --only-binary :all: "feast[flink]"
[...]
Successfully installed Jinja2-3.1.6 MarkupSafe-3.0.3 PyYAML-6.0.3 SQLAlchemy-2.0.51 annotated-doc-0.0.4 annotated-types-0.7.0 anyio-4.14.1 ast-serialize-0.6.0 attrs-26.1.0 bigtree-1.5.1 certifi-2026.6.17 charset_normalizer-3.4.7 click-8.4.2 cloudpickle-3.1.2 colorama-0.4.6 dask-2026.6.0 dill-0.3.9 fastapi-0.139.0 feast-0.63.0 fsspec-2026.6.0 greenlet-3.5.3 gunicorn-26.0.0 h11-0.16.0 httptools-0.8.0 idna-3.18 jsonschema-4.26.0 jsonschema-specifications-2025.9.1 librt-0.12.0 locket-1.0.0 mmh3-5.2.1 mypy-2.1.0 mypy_extensions-1.1.0 numpy-2.5.0 packaging-26.2 pandas-2.3.3 partd-1.4.2 pathspec-1.1.1 prometheus_client-0.25.0 protobuf-7.35.1 psutil-7.2.2 pyarrow-24.0.0 pydantic-2.13.4 pydantic-core-2.46.4 pygments-2.20.0 pyjwt-2.13.0 python-dateutil-2.9.0.post0 python-dotenv-1.2.2 pytz-2026.2 referencing-0.37.0 requests-2.34.2 rpds-py-2026.6.3 six-1.17.0 starlette-1.3.1 tabulate-0.10.0 tenacity-8.5.0 toml-0.10.2 toolz-1.1.0 tqdm-4.68.3 typeguard-4.5.2 typing-extensions-4.16.0 typing-inspection-0.4.2 tzdata-2026.2 urllib3-2.7.0 uvicorn-0.34.0 uvicorn-worker-0.3.0 uvloop-0.22.1 watchfiles-1.2.0 websockets-16.0
[...]

$ pip install --only-binary :all: "feast[flink]==0.64.0"
[...]
ERROR: Cannot install feast and feast[flink]==0.64.0 because these package versions have conflicting dependencies.
                                                                                                                                                                    
The conflict is caused by:
    feast 0.64.0 depends on pyarrow>=21.0.0; extra != "flink"
    feast[flink] 0.64.0 depends on pyarrow<21.0.0 and >=16.1.0; extra == "flink"
[...]

[shuchu]

Thank you for the info. @konstin . I'll keep my eyes open on this issue.

[shuchu]

@franciscojavierarceo @ntkathole @HaoXuAI @XuananLe Regarding this issue, I'm thinking of putting the feastt[flink] feature into a dedicate branch like "feast/flink" temporary and waiting for the pyflink to support pyarrow >= 21.0. what's you guys' opinion? as proposed, I will do:
1, create a new branch to store the feature of pyflink
2, revert the pyflink (and related) PRs. Update the document and make a notice about this change.

[ntkathole]

@shuchu we can also do broad base + upper-bound in extra:

In [project.dependencies]:

"pyarrow>=16.1.0",

In [project.optional-dependencies.flink]:

"pyarrow<21.0.0",