* Factor out detail/value_and_holder.h (from detail/type_caster_base.h)
This is in support of PR #5213:
* trampoline_self_life_support.h depends on value_and_holder.h
* type_caster_base.h depends on trampoline_self_life_support.h
* Fix a minor and inconsequential inconsistency in `copyable_holder_caster`: the correct `load_value()` return type is `void` (as defined in `type_caster_generic`)
For easy future reference, this is the long-standing inconsistency:
* dbf848aff7/include/pybind11/detail/type_caster_base.h (L634)
* dbf848aff7/include/pybind11/cast.h (L797)
Noticed in passing while working on PR #5213.
* Add `DANGER ZONE` comment in detail/init.h, similar to a comment added on the smart_holder branch (all the way back in 2021).
* fix: Make gil_safe_call_once thread-safe in free-threaded CPython
The "is_initialized_" flags is not protected by the GIL in free-threaded
Python, so it needs to be an atomic field.
Fixes#5245
* style: pre-commit fixes
* Apply changes from code review
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Use PyMutex instead of std::mutex in free-threaded build.
PyMutex is now part of the public C API as of 3.13.0b3 and generally has
slightly less overhead than std::mutex.
* style: pre-commit fixes
* Fix instance_map_shard padding
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Use manual padding of instance_map_shard.
The alignas(64) specifier requires aligned allocation, which is not
available on macOS when targeting versions before 10.14.
* Add 'see #5200'
* Update include/pybind11/detail/internals.h
* Update include/pybind11/detail/internals.h
---------
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
* Adds support for TypeGuard and TypeIs
* style: pre-commit fixes
---------
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Support free-threaded CPython (PEP 703)
Some additional locking is added in the free-threaded build when
`Py_GIL_DISABLED` is defined:
- Most accesses to internals are protected by a single mutex
- The registered_instances uses a striped lock to improve concurrency
Pybind11 modules can indicate support for running with the GIL disabled
by calling `set_gil_not_used()`.
* refactor: use PYBIND11_MODULE (#11)
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* Address code review
* Suppress MSVC warning
* Changes from review
* style: pre-commit fixes
* `py::mod_gil_not_used()` suggestion.
* Update include/pybind11/pybind11.h
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>
* Transfer diff from pybind11k fork as-is. New tests are still missing.
* Add `PYBIND11_WARNING_DISABLE_MSVC(4127)` into `PYBIND11_OVERRIDE_IMPL` macro.
* Add test_trampoline_with_pyobject_ptr_return()
* Resolve clang-tidy error: use auto when initializing with a template cast to avoid duplicating the type name [modernize-use-auto,-warnings-as-errors]
* Disabled checking refcount when building with PyPy.
* Clang 3.6, 3.7, 3.9 compatibility.
```
/__w/pybind11/pybind11/tests/test_type_caster_pyobject_ptr.cpp:23:13: error: definition of implicit copy constructor for 'WithPyObjectPtrReturn' is deprecated because it has a user-declared destructor [-Werror,-Wdeprecated]
virtual ~WithPyObjectPtrReturn() = default;
^
```
* Minor clean-up of production code changes.
* Add missing `override` (to resolve clang-tidy error).
* Move PYBIND11_WARNING_POP for a better clang-format outcome.
* Add a pybing function to clear a list.
* Add required error handling.
* Add `/* py-non-const */` as suggested by @Skylion007
---------
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>
* ci: try Python 3.13t
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* fix: support Python 3.13t
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* fix: patch PyPy
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* tests: one more int cast
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* tests: cleanup
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* refactor: use named constant in tests for immortal refcounts
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* docs: move comment about free threaded Python
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* ci: add Python 3.13
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* tests: run the gc for 3.13+
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* Revert "tests: run the gc for 3.13+"
This reverts commit fe8a3ce3d1.
* ci: drop macos ARM for now, need pin updates
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* fix: use Python 3.13 API if on 3.13
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* Call concat with proper namespace in cast.h
* Apply suggestions from code review
* tests: add test for ADL on concat
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* fix: fully qualify all usages of concat
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
* Transfer bug fixes from #4888 wholesale. Full test coverage for all fixes is still missing.
* Add cmake option(PYBIND11_DISABLE_HANDLE_TYPE_NAME_DEFAULT_IMPLEMENTATION) and use in some tests.
* API: Make `numpy.h` compatible with both NumPy 1.x and 2.x
* TST: Update numpy dtype flags test to not covert flags to char
* API: Add `numpy2.h` instead and make `numpy.h` safe
This means that users of `numpy.h` cannot be broken, but need to
update to `numpy2.h` if they want to compile for NumPy 2.
Using Macros simply and didn't bother to try to remove unnecessary
code paths.
* API: Rather than `numpy2.h` use a define for the user.
* Thread `PYBIND11_NUMPY2_SUPPORT` through things and try to adept test matrix
* Small fixups (shouldn't matter)?
* Fixup. Does upgrading scipy help? (it shouldn't?)
(Some other small fixup)
* Use NumPy 2 nightlies for ubuntu-latest job also
* BUG: Fix numpy.bool check
* TST: Fix complexwarning
* BUG: Fix the fact that only the 50 slot is filled with the copy alias
(There were 3 functions all doing the same, only this slot survived 2.x)
* TST: One more test tweak
* TST: Use "long" name for long, since it changed on windows
* TST: Apparently we didn't always have ulong, so just use `L`
* TST: Enforce dtype='l' for test as default isn't long anymore on windows
* Rename macro and invert logic to PYBIND11_NUMPY_1_ONLY
* PYBIND11_INTERNAL_NUMPY_1_ONLY_DETECTED
* Test and code comment expansion
* CI: Use pre-releases of numpy/scipy from pip via explicit version
* CI: NumPy 2 only available on almalinux (as it is Python >=3.9)
* MAINT: Match name more exactly and adopt error phrasing
* MAINT: Pushed early, move helper to be private member
* fix error message compilation when using NumPy 1.x-only backcompat
* silence name shadowing warning
* chore: minor optimization
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>
Co-authored-by: Henry Schreiner <henryschreineriii@gmail.com>
* ci: trying things for Windows Clang failure
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* WIP: try using older clang
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* tests: skip broken test
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* tests: try to skip test in tests
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* fix(tests): Prefer __version__ over MSVC
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* chore: avoid warning on Clang
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* Update tests/test_exceptions.py
* Update tests/test_exceptions.py
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
* stl.h: Use C++11 range-loops with const reference
This saves copy-ctor and dtor for non-trivial types by value
Found by clazy (range-loop-reference)
* test_smart_ptr.cpp cleanup
Introduce `pointer_set<T>`
https://github.com/boostorg/unordered/issues/139
> Based on the standard, the first move should leave a in a "valid but unspecified state";
---------
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>
* fix: Use PyObject_VisitManagedDict() of Python 3.13
Use PyObject_VisitManagedDict() and PyObject_ClearManagedDict() in
pybind11_traverse() and pybind11_clear() on Python 3.13 and newer.
* Add Python 3.13 CI
* tests: don't get numpy/scipy on 3.13 yet
* ci: move 3.13 to upstream
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
---------
Signed-off-by: Henry Schreiner <henryschreineriii@gmail.com>
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
* Equivalent of 5718e4d080
* Resolve clang-tidy errors.
* Moving test_PPCCInit() first changes the behavior!
* Resolve new Clang dev C++11 errors:
```
The CXX compiler identification is Clang 17.0.0
```
```
pytypes.h:1615:23: error: identifier '_s' preceded by whitespace in a literal operator declaration is deprecated [-Werror,-Wdeprecated-literal-operator]
```
```
cast.h:1380:26: error: identifier '_a' preceded by whitespace in a literal operator declaration is deprecated [-Werror,-Wdeprecated-literal-operator]
```
* Resolve gcc 4.8.5 error:
```
pytypes.h:1615:12: error: missing space between '""' and suffix identifier
```
* Specifically exclude `__clang__`
* Snapshot of debugging code (does NOT pass pre-commit checks).
* Revert "Snapshot of debugging code (does NOT pass pre-commit checks)."
This reverts commit 1d4f9ff263.
* [ci skip] Order Dependence Demo
* Revert "[ci skip] Order Dependence Demo"
This reverts commit d37b5409d4.
* One way to deal with the order dependency issue. This is not the best way, more like a proof of concept.
* Move test_PC() first again.
* Add `all_type_info_add_base_most_derived_first()`, use in `all_type_info_populate()`
* Revert "One way to deal with the order dependency issue. This is not the best way, more like a proof of concept."
This reverts commit eb09c6c1b9.
* clang-tidy fixes (automatic)
* Add `is_redundant_value_and_holder()` and use to avoid forcing `__init__` overrides when they are not needed.
* Streamline implementation and avoid unsafe `reinterpret_cast<instance *>()` introduced with PR #2152
The `reinterpret_cast<instance *>(self)` is unsafe if `__new__` is mocked,
which was actually found in the wild: the mock returned `None` for `self`.
This was inconsequential because `inst` is currently cast straight back to
`PyObject *` to compute `all_type_info()`, which is empty if `self` is not
a pybind11 `instance`, and then `inst` is never dereferenced. However, the
unsafe detour through `instance *` is easily avoided and the updated
implementation is less prone to accidents while debugging or refactoring.
* Fix actual undefined behavior exposed by previous changes.
It turns out the previous commit message is incorrect, the `inst` pointer is actually dereferenced, in the `value_and_holder` ctor here:
f3e0602802/include/pybind11/detail/type_caster_base.h (L262-L263)
```
259 // Main constructor for a found value/holder:
260 value_and_holder(instance *i, const detail::type_info *type, size_t vpos, size_t index)
261 : inst{i}, index{index}, type{type},
262 vh{inst->simple_layout ? inst->simple_value_holder
263 : &inst->nonsimple.values_and_holders[vpos]} {}
```
* Add test_mock_new()
* Experiment: specify indirect bases
* Revert "Experiment: specify indirect bases"
This reverts commit 4f90d85f9f.
* Add `all_type_info_check_for_divergence()` and some tests.
* Call `all_type_info_check_for_divergence()` also from `type_caster_generic::load_impl<>`
* Resolve clang-tidy error:
```
include/pybind11/detail/type_caster_base.h:795:21: error: the 'empty' method should be used to check for emptiness instead of 'size' [readability-container-size-empty,-warnings-as-errors]
if (matching_bases.size() != 0) {
^~~~~~~~~~~~~~~~~~~~~~~~~~
!matching_bases.empty()
```
* Revert "Resolve clang-tidy error:"
This reverts commit df27188dc6.
* Revert "Call `all_type_info_check_for_divergence()` also from `type_caster_generic::load_impl<>`"
This reverts commit 5f5fd6a68e.
* Revert "Add `all_type_info_check_for_divergence()` and some tests."
This reverts commit 0a9599f775.
According to https://docs.python.org/3.13/whatsnew/3.13.html:
Add PyThreadState_GetUnchecked() function: similar to
PyThreadState_Get(), but don't kill the process with a fatal error if
it is NULL. The caller is responsible to check if the result is
NULL. Previously, the function was private and known as
_PyThreadState_UncheckedGet().
* Workaround NVCC parse failure in `cast_op`
There is a bug in some CUDA versions (observed in CUDA 12.1 and 11.7 w/ GCC 12.2),
that makes `cast_op` fail to compile:
`cast.h:45:120: error: expected template-name before ‘<’ token`
Defining the nested type as an alias and using it allows this to work
without any change in semantics.
Fixes#4606
* style: pre-commit fixes
* Add comments to result_t referencing PR
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>