* Splitting out include/pybind11/detail/pragma_warning_block.h

* Always including pragma_warning_block.h before common.h (resolved 8 CI failures).

* Crash when printing Unicode to redirected cout
Add failing tests

* Fix Unicode crashes redirected cout

* pythonbuf::utf8_remainder check end iterator

* Remove trailing whitespace and formatting iostream

* Avoid buffer overflow if ostream redirect races
This doesn't solve the actual race, but at least it now has a much lower
probability of reading past the end of the buffer even when data races
do occur.

Updating ci_sh_def.yml, ci_sh_def.yml.patch after `git merge master`. Pushing directly for this mechanical update.

Re-enabling Python 3.10-dev jobs, made possible by PR #2967. Pushing directly because this just affects the CI config.

Bumps [pre-commit/action](https://github.com/pre-commit/action) from v2.0.2 to v2.0.3.
- [Release notes](https://github.com/pre-commit/action/releases)
- [Commits](https://github.com/pre-commit/action/compare/v2.0.2...9b88afc9cd57fd75b655d5c71bd38146d07135fe

)

Signed-off-by: dependabot[bot] <support@github.com>

Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Adding ci_sh_def.yml, based on PR #2930.

* Adding exludes for .patch files.

* Bug fix: adding back `!is_alias<Class>(ptr)` that were accidentally omitted.

* Introducing PYBIND11_SH_AVL, PYBIND11_SH_DEF macros. Applying PYBIND11_SH_DEF to test_factory_constructors.py to complete test coverage.

* Using PYBIND11_SH_DEF in test_methods_and_attributes.cpp, for more complete test coverage.

* Using PYBIND11_SH_DEF in test_multiple_inheritance.cpp, for more complete test coverage.

* Cleaning up test_classh_mock.cpp.

* Better explanations for PYBIND11_SH_AVL, PYBIND11_SH_DEF.

* Disabling 3.10-dev builds.

When the user defines _GLIBCXX_USE_CXX11_ABI=0 to force the pre-c++11 ABI, numpy.h assumes that is_trivially_copyable is available.
It is not necessarily the case. This patch uses clang's feature detection instead.
The workaround is for certain libstdc++ versions, so the test should target these particular versions.

* Connecting PYBIND11_INTERNALS_VERSION to PYBIND11_USE_SMART_HOLDER_AS_DEFAULT.

* Adding section: Classic / Conservative / Progressive cross-module compatibility

Revert "Adding documentation with associated test: Using py::classh but with fallback to classic pybind11"

This reverts commit 49f8f60e.

The direct commit was accidental. I meant to create a PR.

* correcting Werror for Intel

* adding ward for Intel

* adding wards for intel

* another ward for Intel

* missed one intel ward

* exact match for intel compiler

* removing inline limits

* disable warnings about inline limits

* formatter suggestion

* more indent

* hopefully make formatter happy

* addressed review

* fix &&

* Update tests/CMakeLists.txt

Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>

Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>

* Add a failure test for overloaded functions

* Allow function pointer extraction from overloaded functions

* chore: add myself to CODEOWNERS

This will ensure I get notified about pull requests where these files change.

* Update .github/CODEOWNERS

* Using new smart_holder::reclaim_disowned in smart_holder_type_caster for unique_ptr.

* Systematically renaming was_disowned to is_disowned (because disowning is now reversible: reclaim_disowned).

* Systematically renaming virtual_overrider_self_life_support to trampoline_self_life_support (to reuse existing terminology instead of introducing new one).

* Systematically renaming test_class_sh_with_alias to test_class_sh_trampoline_basic.

* Adding a Trampolines and std::unique_ptr section to README_smart_holder.rst.

* MSVC compatibility.

* Porting subset of absltest code from reproducer provided by @elkhrt. Baseline for debugging ASAN heap-use-after-free.

* Moving Py_DECREF to resolve ASAN heap-use-after-free failure.

* Fixing trivial formatting issue.

* Workaround for clang 3.6 and 3.7.

It was only useful for easily harvest this from the GitHub Actions CI results, mostly out of curiosity:

52 C++ function argument 1 is evaluated first.
90 C++ function argument 2 is evaluated first.

These results came for the final CI run for PR #2936; that PR has nothing else to do with the results.

Pushing directly. This tiny change is not worth a PR.

[skip actions]

Fixing very minor oversights discovered while proof-reading the initial version for the first time on GitHub. Pushing directly. Not worth a PR.

* Adding initial README_smart_holder.rst.

* Adding README_smart_holder.rst to MANIFEST.in and test_files.py.