* CI: Intel icc/icpc via oneAPI

Add testing for Intel icc/icpc via the oneAPI images.
Intel oneAPI is in a late beta stage, currently shipping
oneAPI beta09 with ICC 20.2.

* CI: Skip Interpreter Tests for Intel

Cannot find how to add this, neiter the package `libc6-dev` nor
`intel-oneapi-mkl-devel` help when installed to solve this:
```
-- Looking for C++ include pthread.h
-- Looking for C++ include pthread.h - not found
CMake Error at /__t/cmake/3.18.4/x64/cmake-3.18.4-Linux-x86_64/share/cmake-3.18/Modules/FindPackageHandleStandardArgs.cmake:165 (message):
  Could NOT find Threads (missing: Threads_FOUND)
Call Stack (most recent call first):
  /__t/cmake/3.18.4/x64/cmake-3.18.4-Linux-x86_64/share/cmake-3.18/Modules/FindPackageHandleStandardArgs.cmake:458 (_FPHSA_FAILURE_MESSAGE)
  /__t/cmake/3.18.4/x64/cmake-3.18.4-Linux-x86_64/share/cmake-3.18/Modules/FindThreads.cmake:234 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
  tests/test_embed/CMakeLists.txt:17 (find_package)
```

* CI: libc6-dev from GCC for ICC

* CI: Run bare metal for oneAPI

* CI: Ubuntu 18.04 for oneAPI

* CI: Intel +Catch -Eigen

* CI: CMake from Apt (ICC tests)

* CI: Replace Intel Py with GCC Py

* CI: Intel w/o GCC's Eigen

* CI: ICC with verbose make

* [Debug] Find core dump

* tests: use arg{} instead of arg() for Intel

* tests: adding a few more missing {}

* fix: sync with @tobiasleibner's branch

* fix: try ubuntu 20-04

* fix: drop exit 1

* style: clang tidy fix

* style: fix missing NOLINT

* ICC: Update Compiler Name

Changed upstream with the last oneAPI release.

* ICC CI: Downgrade pytest

pytest 6 does not capture the `discard_as_unraisable` stderr and
just writes a warning with its content instead.

* Use new test pinning requirements.txt

* tests: add notes about intel, cleanup

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

* Adding a valgrind build on debug Python 3.9

Co-authored-by: Boris Staletic <boris.staletic@gmail.com>

* Add Valgrind suppression files

- Introduce suppression file, populate it with a first suppression taken from CPython, and fix one leak in the tests
- Suppress leak in NumPy
- More clean tests!
- Tests with names a-e passing (except for test_buffer)
- Suppress multiprocessing errors
- Merge multiprocessing suppressions into other suppression files
- Numpy seems to be spelled with a big P
- Append single entry from valgrind-misc.supp to valgrind-python.supp, and make clear valgrind-python.supp is only CPython

Co-authored-by: Boris Staletic <boris.staletic@gmail.com>

* Enable test_virtual_functions with a workaround

* Add a memcheck cmake target

- Add a memcheck cmake target
- Reformat cmake
- Appease the formatting overlords - they are angry
- Format CMake valgrind target decently

* Update CI config to new action versions

* fix: separate memcheck from pytest

* ci: cleanup

* Merge Valgrind and other deadsnakes builds

Co-authored-by: Boris Staletic <boris.staletic@gmail.com>
Co-authored-by: Henry Schreiner <henryschreineriii@gmail.com>

* Fix STATIC and SHARED flags not being detected

* fix: use classic naming for all lib types

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

Plug leaking function_records in cpp_function initialization in case of exceptions (found by Valgrind in #2746) (#2756)

* Plug leaking function_record objects when exceptions are thrown

* Plug leak of strdup'ed strings in function_record

* Some extra comments about the function_record ownership dance

* Clean up the function_record better, in case of exceptions

* Demonstrate some extra function_record leaks

* Change  DeleteStrings template argument to free_strings runtime argument in destruct(function_record *)

* Zero-state unique_function_record deleter object

* Clarify rvalue reference to unique_ptr parameter in initialize_generic

* Use push_back with const char * instead of emplace_back

Sometimes programmers want to control this, and while it can be changed after the fact, it's commonly set via a CMAKE_ variable; if that variable is set, we should respect that (like the CMAKE_INTERPROCEDURAL_OPTIMIZATION setting).

docs: the order of alternatives for variant types matters, and follows the same rules as overload resolution (#2784)

Found with codespell

* Fix a clang warning from [-Wshadow-field-in-constructor-modified]

warning: modifying constructor parameter 'flag' that shadows a field of 'set_flag' [-Wshadow-field-in-constructor-modified]

* Change name based on review

* Update attr.h: fix a warning found by static code analyzer

Update attr.h: fix a warning found by Visual Studio static code analyzer

Severity:Warning
Code:C6323
Description: Use of arithmetic operator on Boolean type(s).
Location:	C:\src\onnxruntime\debug\pybind11\src\pybind11\include\pybind11\attr.h:547

* Update include/pybind11/attr.h

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

* Update attr.h

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

* Disable builds for 3.10.0a4, and enable a nightly 3.10-dev build

* Fix job name

* Remove deadsnakes job for now

* Add deadsnakes jobs

* There's no deadsnakes 2.7

* Add 3.10 to versions to be discovered by legacy FindPython, and fix debug input to deadsnakes/action

* Try out branch with fix

* Update to deadsnakes/action@v2.1.1

Ignore deprecation warnings about old-style __init__/__setstate__ constructors in the tests (originally done in #2746) (#2759)

* Ignore old-style __init__ warnings

* Simplify ignoreOldStyleInitWarnings with py::exec

* Only wrap single class_::defs to ignore DeprecationWarnings about old-style __init__

* Fix leak in the test_copy_move::test_move_fallback

* Fix leaking PyMethodDef in test_class::test_implicit_conversion_life_support

* Plumb leak in test_buffer, occuring when a mutable buffer is requested for a read-only object, and enable test_buffer.py

* Fix weird return_value_policy::reference in test_stl_binders, and enable those tests

* Cleanup nodelete holder objects in test_smart_ptr, and enable those tests

Fix invalid access when reinterpret_casting a non-pybind11 PyObject* to instance* (found by Valgrind in #2746) (#2755)

* enum: Add Enum.value property

* simplify

* address review

* Do not set docstring for function when it's empty

* No need to check pointer for `free`

* Use ternary operator to conditionally set `ml_doc`

* Update pytest to 6.2.1 in tests/requirements.txt

* Pin pytest to last supported version for 3.5

* Suppress PytestUnraisableExceptionWarning and use sys.__unraisablehook__ instead of sys.unraisablehook

* Fix filterwarnings mark on old pytest and old Python versions

* Cleanup ignore_pytest_unraisable_warning decorator

* docs: mention PYTHONPATH in installing.rst

When pybind11 is included as a submodule, the user needs to update their
Python module search path.  Otherwise, the first c++ compilation command
in docs/basics.rst will fail.

* docs: add a note about compiling the example

This note shows how to modify the compilation command for the example
when the pybind11 source has been included as a Git submodule.

* docs: add a note about compiling the example

Added an internal link to the docs

* docs: updated a note about compiling the example

Also updated the command substitution syntax for consistency

... by leaving the unused parameters anonymous.

(This can be checked e.g. by compiling python_example with -Wall
-Wextra.)

* Avoid thread termination in scoped_released

Do not call `PyEval_RestoreThread()` from `~gil_scoped_release()` if python runtime is finalizing, as it will result in thread termination in Python runtime newer than 3.6, as documented in https://docs.python.org/3/c-api/init.html#c.PyEval_RestoreThread
Similarly do not call `PyThreadState_DeleteCurrent` from `~gil_scoped_acquire()` if runtime is finalizing.

Discovered while debugging PyTorch crash using Python-3.9 described in  https://github.com/pytorch/pytorch/issues/47776



* Simplify _Py_IsFinalizing() availability check

* Fix typo

* Add version agnostic `detail::finalization_guard()`

* Move `finalization_guard` to detail/common.h

And rename it to `is_finalizing`

* Move `is_finalizing()` back to pybind11.h

* Simplify `is_finalizing()` check

One should follow documentation rather than make any assumptions

* feat: disarm

* docs: fix comment

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

* minor cleanup: fixing or silencing flake8 errors

* ci: lock CMake to non-Universal version

* Update .github/workflows/ci.yml

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

fix: also throw in the move-constructor added by the PYBIND11_OBJECT macro, after the argument has been moved-out (if necessary) (#2701)

which was just before set to True in instance->allocate_layout()

* fix: regression with installed pybind11 overriding discovered one

Closes #2709

* docs: wording incorrect

* ci: drop pypy2 linux, add Python 10 dev

* ci: fix mistake

* ci: commented-out PGI 20.11, drop 20.7

Adjusting `type_caster<std::reference_wrapper<T>>` to support const/non-const propagation in `cast_op`. (#2705)


* Allow type_caster of std::reference_wrapper<T> to be the same as a native reference.

Before, both std::reference_wrapper<T> and std::reference_wrapper<const T> would
invoke cast_op<type>. This doesn't allow the type_caster<> specialization for T
to distinguish reference_wrapper types from value types.

After, the type_caster<> specialization invokes cast_op<type&>, which allows
reference_wrapper to behave in the same way as a native reference type.

* Add tests/examples for std::reference_wrapper<const T>

* Add tests which use mutable/immutable variants

This test is a chimera; it blends the pybind11 casters with a custom
pytype implementation that supports immutable and mutable calls.

In order to detect the immutable/mutable state, the cast_op needs
to propagate it, even through e.g. std::reference<const T>

Note: This is still a work in progress; some things are crashing,
which likely means that I have a refcounting bug or something else
missing.

* Add/finish tests that distinguish const& from &

Fixes the bugs in my custom python type implementation,
demonstrate test that requires const& and reference_wrapper<const T>
being treated differently from Non-const.

* Add passing a const to non-const method.

* Demonstrate non-const conversion of reference_wrapper in tests.

Apply formatting presubmit check.

* Fix build errors from presubmit checks.

* Try and fix a few more CI errors

* More CI fixes.

* More CI fixups.

* Try and get PyPy to work.

* Additional minor fixups. Getting close to CI green.

* More ci fixes?

* fix clang-tidy warnings from presubmit

* fix more clang-tidy warnings

* minor comment and consistency cleanups

* PyDECREF -> Py_DECREF

* copy/move constructors

* Resolve codereview comments

* more review comment fixes

* review comments: remove spurious &

* Make the test fail even when the static_assert is commented out.

This expands the test_freezable_type_caster a bit by:
1/ adding accessors .is_immutable and .addr to compare identity
from python.
2/ Changing the default cast_op of the type_caster<> specialization
to return a non-const value. In normal codepaths this is a reasonable
default.
3/ adding roundtrip variants to exercise the by reference, by pointer
and by reference_wrapper in all call paths.  In conjunction with 2/, this
demonstrates the failure case of the existing std::reference_wrpper conversion,
which now loses const in a similar way that happens when using the default cast_op_type<>.

* apply presubmit formatting

* Revert inclusion of test_freezable_type_caster

There's some concern that this test is a bit unwieldly because of the use
of the raw <Python.h> functions. Removing for now.

* Add a test that validates const references propagation.

This test verifies that cast_op may be used to correctly detect
const reference types when used with std::reference_wrapper.

* mend

* Review comments based changes.

1. std::add_lvalue_reference<type> -> type&
2. Simplify the test a little more; we're never returning the ConstRefCaster
type so the class_ definition can be removed.

* formatted files again.

* Move const_ref_caster test to builtin_casters

* Review comments: use cast_op and adjust some comments.

* Simplify ConstRefCasted test

I like this version better as it moves the assertion that matters
back into python.

This is nice enough to be mentioned explicitly in the docs.

docs: add warning about FindPython's Development component when libraries don't exist (e.g. on manylinux) (#2689)

* Add waring about FindPython's Development component when libraries don't exist (e.g. on manylinux)

* Minor wording update (thanks, @henryiii!)

`git submodule add` needs the branch before the repository or else it is ignored. The previous code checked out the `master` branch, not the `stable` branch.