Needs a release note. You also missed spots the C function still shows up:

numpy/__init__.pxd
526:    void PyArray_SetStringFunction (object, int)

doc/source/reference/c-api/types-and-structures.rst
223:    methods can also be altered using :c:func:`PyArray_SetStringFunction`.

numpy/__init__.cython-30.pxd
611:    void PyArray_SetStringFunction (object, int)

Maybe also add a note or comment to numpy/_core/code_generators/numpy_api.py above the entry for the function that it is stubbed out and should be removed in the future.

repr can't be modified by a user with the Python or C API.

That's not true is it? You can use printoptions to override how the repr appears too:

In [1]: with np.printoptions(formatter={"all": lambda x: "hello"}):
   ...:     print(repr(np.array([1, 2, 3])))
   ...:
array([hello, hello, hello])

That's not true is it? You can use printoptions to override how the repr appears too:

In [1]: with np.printoptions(formatter={"all": lambda x: "hello"}):
   ...:     print(repr(np.array([1, 2, 3])))
   ...:
array([hello, hello, hello])

I think there's no need to dig into it so let's say that we can override repr/str sufficiently with printoptions.

With set_string_function, that we're removing, we can override whole repr, not only elements repr (but I guess it's a redundant feature):

In [104]: a = np.array([[1,2,3]])

In [105]: a
Out[105]: array([[1, 2, 3]])

In [106]: np.set_string_function(lambda x: "hello")

In [107]: a
Out[107]: hello

Yeah, I think printoptions replaces all the actually useful use cases. If anyone really wants to override how we lay out printing arrays they can do that themselves.

Done!

I do see one downstream project trying to work around us removing the ability to completely override the print formatting for the entire array:

napari/napari#6776

They ended up deciding to set edgeitems so the array size is limited in the printed output, but it's not quite the same as what they had before.

That said, maybe printoptions could grow an option to override printing the entire array?

Since we began deprecating this in 2.0, I think we need to wait for 2.2 to remove it, no?

The C-level function was never deprecated but the python-level function was removed from the public API in NumPy 2.0, skipping the deprecation.

It seems odd to me to have the Python function gone from the public API but to insist on a long deprecation period for the C-level function that is used much less often.

It seems odd to me to have the Python function gone from the public API but to insist on a long deprecation period for the C-level function that is used much less often.

FWIW, the deprecation warning should have been implemented to include the C-side, but the normal deprecation policy is 2 (or maybe even 3 in this case) releases (1 year).
I don't care too much about bending that here, it isn't a typical end-user function. But what is the reason?

It looks like it was deprecated first, then quickly removed before the deprecation was in a release (the comment says it was deprecated in numpy 2.0). I think it was included in the general API cleanup, @mtsokol might remember the details about why it was included for removal so quickly after being deprecated. I don't remember thinking too carefully about this one during review.

It looks like it was deprecated first, then quickly removed before the deprecation was in a release

I am confused: This PR would be the one to remove it, no? Giving us a very brief deprecation period on the Python function.

I am happy to ignore the C version, and had we done it in 2.0, I would have been happy to just do it probably.
But now it does seem a bit like there is no reason to not just wait for 2.2 at least, so I am -0.0 for finalizing the deprecation so quickly?

It looks like it was deprecated first, then quickly removed before the deprecation was in a release (the comment says it was deprecated in numpy 2.0). I think it was included in the general API cleanup, @mtsokol might remember the details about why it was included for removal so quickly after being deprecated. I don't remember thinking too carefully about this one during review.

I think we just made a decision during one of the meetings to put this function in the "remove" group for 2.0 and removed it: #24306 (See Details).

So set_string_function was removed from the main namespace without a deprecation period (the error message says what should be used instead), and we didn't touch C's PyArray_SetStringFunction at all. This PR removes set_string_function implementation and stubs out PyArray_SetStringFunction.

That said, maybe printoptions could grow an option to override printing the entire array?

Sure, then there would be a new option override_repr or something in np.set_printoptions that accepts a callable and completely replaces how arrays are printed, replicating set_string_function functionality.

Aha, so the point is that it was effectively already removed in 2.0 because it only exists as a private function.

That said, maybe printoptions could grow an option to override printing the entire array?

@ngoldbaum I added override_repr option to np.printoptions that allows defining custom repr(...), which replicates:

np.set_string_function(lambda x: "FOO", repr=True)

One thing we forgot yesterday: This needs a release note for the C-API change at least.

EDIT: Sorry, I see it's there, would be nice to put the C-API part into c_api but either way is fine.

Sorry for taking a little while to merge this one. Let's bring this in and thanks for finishing up the work on this deprecation/removal @mtsokol!

I can see the use of adding it, but hadn't realized it was added in this PR. If we add override_repr we should have a relase note for that, and that is also a place to give anyone a chance to bike-shed names.

Sure! Let me tweak the release note separately.

Thanks for adding this functionality back with override_repr!

Just wanted to note that at least one other project that relies on this feature: lovely-numpy, and I personally also have a hackier version, so it's great that it's being left in .

Very useful to be able to quickly see array information in one glance :)