Fix matrix coercion with numpy 2.1

[antonio-rojas]

numpy 2.1 requires a copy argument for numpy.array. Fixes test failures with numpy 2.1:

**********************************************************************
File "/usr/lib/python3.12/site-packages/sage/matrix/matrix1.pyx", line 723, in sage.matrix.matrix1.Matrix.numpy
Failed example:
    b = numpy.array(a); b
Exception raised:
    Traceback (most recent call last):
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 716, in _run
        self.compile_and_execute(example, compiler, test.globs)
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 1146, in compile_and_execute
        exec(compiled, globs)
      File "<doctest sage.matrix.matrix1.Matrix.numpy[9]>", line 1, in <module>
        b = numpy.array(a); b
            ^^^^^^^^^^^^^^
      File "sage/matrix/matrix1.pyx", line 673, in sage.matrix.matrix1.Matrix.numpy (build/cythonized/sage/matrix/matrix1.c:14076)
        def numpy(self, dtype=None):
    TypeError: numpy() got an unexpected keyword argument 'copy'
**********************************************************************
File "/usr/lib/python3.12/site-packages/sage/matrix/matrix1.pyx", line 727, in sage.matrix.matrix1.Matrix.numpy
Failed example:
    b.dtype
Exception raised:
    Traceback (most recent call last):
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 716, in _run
        self.compile_and_execute(example, compiler, test.globs)
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 1146, in compile_and_execute
        exec(compiled, globs)
      File "<doctest sage.matrix.matrix1.Matrix.numpy[10]>", line 1, in <module>
        b.dtype
        ^
    NameError: name 'b' is not defined
**********************************************************************
File "/usr/lib/python3.12/site-packages/sage/matrix/matrix1.pyx", line 730, in sage.matrix.matrix1.Matrix.numpy
Failed example:
    b.shape
Exception raised:
    Traceback (most recent call last):
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 716, in _run
        self.compile_and_execute(example, compiler, test.globs)
      File "/usr/lib/python3.12/site-packages/sage/doctest/forker.py", line 1146, in compile_and_execute
        exec(compiled, globs)
      File "<doctest sage.matrix.matrix1.Matrix.numpy[11]>", line 1, in <module>
        b.shape
        ^
    NameError: name 'b' is not defined
**********************************************************************

[kiwifb]

Is it compatible with numpy 2.0?

[antonio-rojas]

Is it compatible with numpy 2.0?

works locally, I see some CI issues though

[kiwifb]

Should it be copy="copy"?

[kiwifb]

OK, I have read the manual and now I understand your patch. copy expect a boolean and the default is True. I think that should be the default for the numpy function rather than None (which I am guessing could be taken as False with some casting).

[antonio-rojas]

OK, I see the issue is incompatibility with numpy 1, which is still shipped in sage-the-distro. Let's try default to True. This parameter should only matter if you're calling numpy on a numpy object itself, which shouldn't be too common.

[antonio-rojas]

Works locally now with 2.1, 2.0 and 1.26

[github-actions]

Documentation preview for this PR (built with commit 8e2b65e; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[kiwifb]

LGTM

[tornaria]

https://github.com/void-linux/void-packages/blob/master/srcpkgs/sagemath/patches/numpy_2.1a.patch

https://github.com/void-linux/void-packages/blob/master/srcpkgs/sagemath/patches/numpy_2.1b.patch

[user202729]

Just come across this. While this fixes the tests, it isn't really correct:

• It breaks Liskov substitution principle — Matrix_numpy_dense inherits from Matrix_dense, yet Matrix_numpy_dense doesn't allow passing copy=True to .numpy() method.
• It doesn't really work correctly (see https://numpy.org/doc/stable/numpy_2_0_migration_guide.html#adapting-to-changes-in-the-copy-keyword and https://github.com/scipy/scipy/pull/20172/files for the proper fix)
• It allows the user to pass copy=False which (by the interface contract) must return a view of the internal array, which is dangerous (depends specifically on the fact that the type is implemented with a numpy array). And this cannot be undone without a long deprecation period.
• The copy=copy is in fact useless — self.list() is guaranteed to always return a copy, so even if copy=False is passed, it will make a copy anyway in numpy 1.26 and raise an error in numpy 2.0.

A fix is provided in #39152 . Maybe someone can take a look?