build(deps): bump the cargo group across 1 directory with 3 updates

[dependabot]

Bumps the cargo group with 3 updates in the / directory: pyo3, rand and wide.

Updates pyo3 from 0.29.2 to 0.29.3

Release notes

Sourced from pyo3's releases.

PyO3 0.29.3

This release is a series of fixes worth landing to the PyO3 0.29 series; the PyO3 0.30 release is planned to follow within a few weeks, however during the development cycle a number of rough edges were discovered which were deemed worth a backport.

The patch broadly includes:

• a further fix the to the opt-out of raw-dylib linking for Windows introduced in PyO3 0.29.1 (the switch to raw-dylib linking was introduced in PyO3 0.29.0),
• fixes to support GraalPy on windows,
• a handful of small performance optimizations,
• a number of non-breaking fixes to the experimental-inspect type stub generation (expect more improvements in 0.30), and
• various other bugfixes such as to FFI definitions, to resolve crashes, and to fix clippy lints caused by PyO3 macros.

For a full list of the exact fixes please consult the CHANGELOG.

Thank you to the following contributors for the improvements:

@​adamgerhant-ai @​alex @​bschoenmaeckers @​charliermarsh @​chirizxc @​davidhewitt @​dependabot[bot] @​dylanpulver @​exg @​FranciscoTGouveia @​JMak-Security @​jonasdedden @​jorenham @​konstin @​latent-9 @​mejrs @​ngoldbaum @​Person-93 @​PeterFaiman @​sneakers-the-rat

Changelog

Sourced from pyo3's changelog.

[0.29.3] - 2026-09-30

Packaging

• Support GraalPy on Windows. #6406

Added

• Add FFI definitions PyModule_GetState_DuringGC, PyModule_GetToken_DuringGC, PyObject_GetTypeData_DuringGC, PyType_GetModuleState_DuringGC, PyType_GetBaseByToken_DuringGC, PyType_GetModule_DuringGC, and PyType_GetModuleByToken_DuringGC for Python 3.15 and up. #6419

Changed

• Optimize u128/i128 conversions on Python 3.15+. #6160
• Preallocate output sets when extracting Python set and frozenset values into Rust hash sets. #6225
• Optimize PyBytes::as_bytes() when not using abi3 or abi3t. #6377

Removed

• Remove FFI definitions PyExc_RecursionErrorInst and Py_UseClassExceptionsFlag (not present in supported Python versions). #6421

Fixed

• Revert removal of FFI definition _PyFrameEvalFunction, _PyInterpreterState_GetEvalFrameFunc and _PyInterpreterState_SetEvalFrameFunc, which are not currently treated as CPython private API despite their name. #6195
• Fix FFI definition PyUnstable_Eval_RequestCodeExtraIndex FFI binding to link to the private CPython symbol on Python versions before 3.12. #6195
• Fix FFI definitions PyByteArray_GET_SIZE, PyList_GET_SIZE, and PySet_GET_SIZE to use an atomic load for free-threaded Python. #6230
• experimental-inspect: deduplicate repeated members of a type union in the generated stubs, and put a space on both sides of the | rather than only before it. #6273
• experimental-inspect: __next__ and __anext__ returning Option<T> (or PyResult<Option<T>>) are now introspected as returning T, since None stops the iteration instead of being yielded. #6274
• Fix clippy::clone_on_copy warnings triggered on nightly Rust by #[pyclass(from_py_object)] on classes which implement Copy. #6309
• experimental-inspect: __pow__, __rpow__ and __get__ now introspect their trailing argument as defaulting to None, matching the CPython slot wrappers which substitute None when it is omitted. #6363
• experimental-inspect: fix the stubs of a package nested inside another package being written into a directory named after its parent instead of its own name. #6365
• Fix many unresolved symbols when linking for PyPy due to incorrect link names in pyo3-ffi. #6389
• Fix empty tuple type hints being rendered as invalid tuple[] instead of tuple[()] in generated stubs and PyStaticExpr display. #6393
• experimental-inspect: escape stub docstrings to prevent invalid Python syntax. #6394
• experimental-inspect: fix the generated type annotation for PyBuffer on python <3.12 by using typing_extensions.Buffer instead of collections.abc.Buffer. #6395
• Fix string constants containing control characters rendering as invalid Python in Display for PyStaticExpr. #6397
• Fix crash when a detached thread is terminated while trying to reattach during interpreter finalization. #6404
• Fix link failures on 32-bit Windows when raw-dylib linking is disabled. #6410
• Fix DLL load failures on Windows with PyPy when raw-dylib linking is disabled. #6410
• Fix FFI definition PyVectorcall_Call failing to link on Python 3.11 and older. #6410
• Fix many unresolved data symbols when linking for PyPy due to incorrect link names in pyo3-ffi. #6421
• Fix linker errors on PyPy for outdated FFI definitions where PyPy moved from a function to a macro. #6422
• Fix crash when GC traverses a #[pyclass] type object during class attribute initialization. #6453

Commits

• 451d99f release: 0.29.3
• a19b491 ci: Exclude the Python extensions article from guide checks (#6472)
• d76560a ci: exclude current release tag from link check (#6463)
• 91c0e7d fix crash with GC during attribute initialization (#6453)
• 9d467b1 only override env vars on Python 3.8
• 945e62a bump GraalPy versions tested in CI, test graalpy on windows (#6406)
• 9712753 fix 3.8 ffi checks
• 42a8fcc fix clippy lint (#6334)
• 018dbcc Add DuringGC FFI bindings for 3.15 (#6419)
• 75ada93 fix FFI cases where PyPy replaced function with macro (#6422)
• Additional commits viewable in compare view

Updates rand from 0.10.2 to 0.10.3

Changelog

Sourced from rand's changelog.

[0.10.3] — 2026-09-20

Fixes

• Fix WeightedIndex panic when the sum of float weights is infinite; return Error::Overflow instead (#1808)
• Fix spurious Error::NonFinite from Uniform::new_inclusive on large finite float ranges such as 0.0..=f64::MAX (#1821)
• Fix possible panic due to sampling a deserialized Uniform<char> (#1831)

Changes

• Report exact remaining lengths from WeightedIndex::weights() and reduce overhead when reading weights (#1838)

#1808: rust-random/rand#1808 #1821: rust-random/rand#1821 #1831: rust-random/rand#1831 #1838: rust-random/rand#1838

Commits

• 9e7d328 Prepare rand 0.10.3 (#1840)
• f73ce74 Optimize WeightedIndex weight lookup and iteration (#1838)
• ef9e044 Avoid panic from deserialized Uniform\<char> where range == 0 (#1831)
• c994eb1 docs: fix angle unit in quick start example (#1839)
• 33dea4f Test that WeightedIndex rejects INFINITY with Error::Overflow (#1822)
• 94c9078 Fix Uniform::new_inclusive overflow on large finite float ranges (#1821)
• bb1262f Use Xoshiro256PlusPlus in examples/rayon-monte-carlo.rs (#1805)
• 521fab6 Stop pinning dependencies (#1820)
• 3f7c433 Stop pinning dependencies
• cf4f73e sample_efraimidis_spirakis: error on more than amount non-finite weights (#1814)
• Additional commits viewable in compare view

Updates wide from 1.7.0 to 1.7.1

Changelog

Sourced from wide's changelog.

1.7.1

• #327: Fixed i/u32x4::to_bitmask incorrect result in fallback path

1.7

• Added Select trait for arbitrary types in the select method
• Added support for select with mixed types (e.g., selecting f32x4 values with u32x4 mask)
• Added from_i16x16 and From<i16x16> to i32x16.
• Added from_i8x32 and From<i8x32> to i16x32.
• Added runtime-index shuffle functions. Previous functions swizzle and swizzle_relaxed have been deprecated. See [PR #317](lokathor/wide#317) for more information.
• Optimize u8xL shifts by a constant on AVX by using shift instructions on u16x(L/2), followed by masking out the bits shifted across bytes. See [PR #319](Lokathor/wide#319), which supersedes #313.
• Added from_ne_bytes and to_ne_bytes to all SIMD vector types. See this issue for more information.
• Added u8x64 and i8x64 types. See this PR.
• Fixed the signum documentation: the docs previously misstated the grammar and the set of inputs that map to 1.0/-1.0. The behavior itself was already correct and matches [f32::signum].
• Optimized float signum. This changes the bit-patterns of returned NaNs.
• Optimized copysign.
• Added add_mul_lo and add_mul_hi for u32xN and u64xN: a fused multiply-add over the low W bits of each lane, keeping the low or high half of the product. u64xN uses vpmadd52lo/hi on AVX-512-IFMA when W == 52.
• Added unpack_lo and unpack_hi for u32xN and u64xN.

1.6.1

• Reverted a compile time optimization that triggered a compiler bug. See this issue for more information.
• Fixed code path errors that resulted when avx512f was enabled without avx512dq, which is a valid potential build. CI checks have been added for this combination to prevent recurrence.

1.6.0

• Potential Change: i64xN functions to_bitmask, any, all and none no longer guarantee that the sign-bit is used to decide whether an element is considered "true" or "false". This is being counted as a "bug" in the docs for previously being too specific.
• Added cast functions cast_unsigned, cast_signed, to_bits, from_bits.

... (truncated)

Commits

• a051907 chore: Release wide version 1.7.1
• 95ac1c8 changelog
• 4bdf5d6 fix: Fix non-SIMD code paths and add CI tests for them (#327)
• See full diff in compare view