Dependency specification missing for python3.14

[giochanturia]

There is no case defined for numpy using python3.14 in pyproject.toml. As a result, installing packages that depend on opencv-python and a newer version of numpy does not work as expected and causes conflicts between numpy versions.

opencv-python/pyproject.toml

Line 5 in 81b4c07

"numpy==2.1.3; python_version=='3.13'",

[lucalista]

I also reported an issue related to a problematic numpy version (#1182) that was fixed with a manual upgrade of numpy version.

[MDSALMANSHAMS]

This one, #1155 and #1160 are all the same root cause, and it looks resolved since 4.13.0.90.

The opencv wheel was never the problem on 3.14 — the wheels are cp37-abi3, so a 3.14 interpreter picks them up fine (both logs in #1155/#1160 show opencv_python-4.12.0.88-cp37-abi3-win_amd64.whl downloading successfully). What failed was the numpy requirement it pulled in. 4.12.0.88 capped it:

4.12.0.88   numpy<2.0; python_version < "3.9"
            numpy<2.3.0,>=2; python_version >= "3.9"     <- the cap

On 3.14 that cap resolves to numpy 2.2.6, which has no cp314 wheel, so pip fell back to building numpy from source — which is exactly the numpy-2.2.6.tar.gz → Preparing metadata (pyproject.toml) ... error both reporters hit.

The cap is gone from 4.13.0.90 onward (checked the PyPI JSON metadata for each release):

4.13.0.90   numpy>=2; python_version >= "3.9"
4.13.0.92   numpy>=2; python_version >= "3.9"
4.14.0.94   numpy>=2; python_version >= "3.9"
5.0.0.93    numpy>=2; python_version >= "3.9"

So pip on 3.14 now resolves to a numpy that actually ships cp314 wheels, and no source build is attempted.

For the specific ask in this issue — a python_version case for 3.14 in the build requirements — pyproject.toml on 4.x now has one:

"numpy==2.1.3; python_version=='3.13'",
"numpy==2.3.2; python_version=='3.14'",

@asmorkalov this looks safe to close, along with #1155 and #1160, unless someone can still reproduce it on 4.14.0.94 or 5.0.0.93.