Numpy version fix

[Houbein]

In the latest 4.12.0.88 release of opencv-python the Numpy version requirement was changed from a minimum version fix to a maximum version fix.
But the maximum version fix is set to an older version of Numpy, which caused Numpy to be downgraded on all our machines.
Please DO NOT DO THIS! I absolute hate maximum version locks when they are not absolutly necessary.
We have been running opencv-python with Numpy 2.3.1 without any issues, there is no need to forcibly downgrade everyone to Numpy 2.2.6.

[TheExDeus]

Yeah, same. I hope this was just an oversight.

[yangw8]

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
opencv-python 4.12.0.88 requires numpy<2.3.0,>=2; python_version >= "3.9", but you have numpy 1.26.0 which is incompatible.
opencv-python-headless 4.12.0.88 requires numpy<2.3.0,>=2; python_version >= "3.9", but you have numpy 1.26.0 which is incompatible.这个怎么解决啊

[schm0]

Requiring Numpy 2.x for Python 3.9 makes it incompatible with older Tensorflow versions....
tensorflow-intel 2.17.1 depends on numpy<2.0.0 and >=1.23.5; python_version <= "3.11"

[Tenzer]

#1121 should fix this.

[Houbein]

So we are now 5 months on. The release candidate for numpy 2.4 has just come out, while this ticket for supporting numpy 2.3 is stil open.
I have tested with numpy 2.4 just like I did with 2.3 and found 0 issues.
Again, please either speed up your release cycle or relax version fixing, preferably both, it's really getting annoying now.

[asemoon]

Hi Everyone,
Has this been addressed? the numpy is stuck in a version where it is required to be updated by other packages now we are stuck.

[Houbein]

There are now several tickets complaining about this.
Again, there is ZERO code change needed to support even numpy 2.4, it just works, the problem is not the code, just your much too strict version fixing.

[guillaume-mueller]

I can install numpy==2.4.1 with opencv-python-headless==4.11.0.86 but I can only install numpy>=2,<2.3.0 with opencv-python-headless==4.12.0.88, what the hell?

[Tenzer]

Looking at this again, I wonder if this actually already is fixed by #1142 which already has been merged, and just needs to be released?

I'm not totally sure when the version restrictions in setup.py and pyproject.toml apply. I guess one might be for when building the wheel and one when installing the wheel?

opencv-python/pyproject.toml

Lines 3 to 5 in bfb85dc

	"numpy<2.0; python_version<'3.9'",
	"numpy==2.0.2; python_version>='3.9' and python_version<'3.13'",
	"numpy==2.1.3; python_version=='3.13'",

opencv-python/setup.py

Lines 26 to 27 in bfb85dc

	'numpy<2.0; python_version<"3.9"',
	'numpy>=2; python_version>="3.9"',

[Tenzer]

Version 4.13.0.90 was released yesterday which now depend on numpy >= 2.

[schm0]

just edit the METDATA file and remove the numpy requirements.
It even seems to work with Numpy versions 1.x.
I did run 4.12 with python 3.11 and numpy 1.26.4 for quite some time.

[Houbein]

Fixed by #1142