* Experiment using -DCMAKE_BUILD_TYPE=Debug for Centos 8.

* Moving comment because the current placement seems to mess up the cmake command.

* Using `echo > VAR_XXX` and `$(cat VAR_XXX)` trick to emulate using variables (https://github.com/actions/starter-workflows/issues/68#issuecomment-524937002).

Removing `std::get_deleter` `const_cast` to retest because the code has changed significantly. (#3045)

* Add const T to docstring generation.

* Change order.

* See if existing test triggers for a const type.

* Add tests.

* Fix test.

* Remove experiment.

* Reformat.

* More tests, checks run.

* Adding `test_fmt_desc_` prefix to new test functions.

* Using pytest.mark.parametrize to 1. condense test; 2. exercise all functions even if one fails; 3. be less platform-specific (e.g. C++ float is not necessarily float32).

Co-authored-by: Ralf W. Grosse-Kunstleve <rwgk@google.com>

* Drop constraints on casting of std::shared_ptr

std::shared_ptrs can be shared across python and C++ by design.

* Correctly report casting error

It is important to return an empty handle.
Simply returning None, would skip the error handling in
simple_collector / unpacking_collector, although a python exception is set.
A function call would then be processed with a (wrong) None argument!

* Return None for nullptr

* Revert "Drop constraints on casting of std::shared_ptr"

This reverts commit 7cf53ae8b4f99c0382845acf8bc63b64924dc891.

Adjusting valgrind suppression for pypocketfft to resolve systematic failures that started to appear on 2020-05-27. (#3022)

* Set visibility of exceptions to default.

Co-authored-by: XZiar <czktc2007@gmail.com>

* add test

* update docs

* Skip failed test.

test_call_policies: Explicitly check free-functions and static methods

* 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.