C API: Remove deprecated `PySlice_GetIndicesEx()`, or remove the deprecation

[vstinner]

@serhiy-storchaka deprecated the PySlice_GetIndicesEx() function in Python 3.6.1.

We are now at Python 3.16, 10 releases later. So should we remove PySlice_GetIndicesEx() functon? Or should we keep it just remove the deprecation and keep it forever?

In PyPI top 15,000 projects (at 2026-09-17), I found 79 projects by searching for PySlice_GetIndicesEx:

• PyICU-binary (2.7.4)
• PyQt5 (5.15.11)
• Theano (1.0.5)
• Theano-PyMC (1.1.2)
• arize_phoenix_sqlean (0.1.2)
• biopython (1.88)
• bitarray-hardbyte (2.3.8)
• btrees (6.5)
• catboost (1.2.10)
• cffi (2.1.1)
• cvxopt (1.3.3)
• cython (3.3.0)
• ddtrace (4.15.0)
• dlib (20.0.1)
• ethos_u_vela (5.2.0)
• glpk (0.4.8)
• gmpy2 (2.3.1)
• igraph (1.0.0)
• instanttensor (0.2.0)
• kuzu (0.11.3)
• ladybug (0.20.4)
• lxml (6.1.3)
• mercurial (7.2.4)
• minorminer (0.2.22)
• mmcif (1.1.1)
• nanobind (3.0.1)
• ncnn (1.0.20260526)
• ntgcalls (2.2.5)
• numba (0.67.0)
• numpy (2.5.3)
• onnxoptimizer (0.4.2)
• open_spiel (2.0.2)
• opencc (1.4.2)
• opentimelineio (0.18.1)
• orjson (3.12.0)
• pandas (3.0.5)
• passagemath_combinat (10.8.11)
• passagemath_modules (10.8.11)
• passagemath_standard_no_symbolics (10.8.11)
• pillow (12.3.0)
• polyscope (2.6.1)
• praat_parselmouth (0.4.7)
• protovalidate (2.0.0)
• pybind11 (3.1.0)
• pybind11_global (3.1.0)
• pycuda (2026.1)
• pydemumble (0.1.3)
• pyfastx (2.3.1)
• pygame (2.6.1)
• pygame_ce (2.5.8)
• pyglm (2.8.3)
• pyicu (2.16.2)
• pymatching (2.4.0)
• pyodbc (5.3.0)
• pyopencl (2026.1.4)
• pyqt5_sip (12.19.0)
• pyrsistent (0.20.0)
• pysimdjson (7.0.2)
• pysqlite3 (0.6.0)
• pytensor (3.3.2)
• pywhispercpp (1.5.1)
• pywinpty (3.0.5)
• real_ladybug (0.15.3)
• recordclass (0.24.1)
• regex (2026.9.10)
• rhino3dm (8.35.0)
• sip (6.16.1)
• sqlcipher3 (0.6.2)
• sqlcipher3_wheels (0.5.7)
• sqlean_py (3.50.4.5)
• stpyv8 (13.1.201.22)
• symusic (0.6.0)
• ur_rtde (1.6.5)
• viennarna (2.7.2)
• winrt_windows_devices_enumeration (3.2.1)
• winrt_windows_foundation (3.2.1)
• winsdk-1.0.0b10
• wxpython (4.3.1)
• xatlas (0.0.11)

That's a long list of projects. So I'm not sure that it's a good idea to break them if we can continue to maintain PySlice_GetIndicesEx(). But I would prefer to keep the deprecation.

Can we do something to reduce the list of affected projects?

Linked PRs

• gh-158077: Remove deprecated PySlice_GetIndicesEx() function #158089

[serhiy-storchaka]

The PySlice_GetIndicesEx() function was redefined as a PySlice_GetIndicesEx() macro, which is safe to use (because it is a macro, and length is calculated after calling PySlice_Unpack() with other arguments). If projects do not define Py_LIMITED_API or define it >= 0x03060100, they are safe and do not need to change their code.

[vstinner]

Oh, I didn't recall the complex situation of this function. Hum.

I don't think that it has to be that complicated. I propose to remove the deprecated and unsafe function implementation, leave a static inline function implementation, and remove the deprecation. The C API would not be affected.

I created PR gh-158089 to do that.

[vstinner]

PySlice_GetIndicesEx() function was reimplemented as a macro by commit b2a5be0 from issue gh-72054.

Py_DEPRECATED(3.7) was only added later, by commit 3c8724f.

[gvanrossum]

Can we do something to reduce the list of affected projects?

Try opening issues or PRs to the most commonly-used projects on the list. (IIRC Hugo maintains a list of PyPI projects by popularity.)