pybind11/include/pybind11/gil.h
Ralf W. Grosse-Kunstleve 0e2c3e5db4
Add pybind11/gil_safe_call_once.h (to fix deadlocks in pybind11/numpy.h) (#4877)
* LazyInitializeAtLeastOnceDestroyNever v1

* Go back to using `union` as originally suggested by jbms@. The trick (also suggested by jbms@) is to add empty ctor + dtor.

* Revert "Go back to using `union` as originally suggested by jbms@. The trick (also suggested by jbms@) is to add empty ctor + dtor."

This reverts commit e7b8c4f0fc.

* Remove `#include <stdalign.h>`

* `include\pybind11/numpy.h(24,10): fatal error C1083: Cannot open include file: 'stdalign.h': No such file or directory`

* @tkoeppe wrote: this is a C interop header (and we're not writing C)

* Suppress gcc 4.8.5 (CentOS 7) warning.

```
include/pybind11/eigen/../numpy.h:63:53: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing]
         return *reinterpret_cast<T *>(value_storage_);
                                                     ^
```

* Replace comments:

Document PRECONDITION.

Adopt comment suggested by @tkoeppe: https://github.com/pybind/pybind11/pull/4877#discussion_r1350356093

* Adopt suggestion by @tkoeppe:

* https://github.com/pybind/pybind11/pull/4877#issuecomment-1752969127

* https://godbolt.org/z/Wa79nKz6e

* Add `PYBIND11_CONSTINIT`, but it does not work for the current use cases:

```
g++ -o pybind11/tests/test_numpy_array.os -c -std=c++20 -fPIC -fvisibility=hidden -O0 -g -Wall -Wextra -Wconversion -Wcast-qual -Wdeprecated -Wundef -Wnon-virtual-dtor -Wunused-result -Werror -isystem /usr/include/python3.11 -isystem /usr/include/eigen3 -DPYBIND11_STRICT_ASSERTS_CLASS_HOLDER_VS_TYPE_CASTER_MIX -DPYBIND11_ENABLE_TYPE_CASTER_ODR_GUARD_IF_AVAILABLE -DPYBIND11_TEST_BOOST -Ipybind11/include -I/usr/local/google/home/rwgk/forked/pybind11/include -I/usr/local/google/home/rwgk/clone/pybind11/include /usr/local/google/home/rwgk/forked/pybind11/tests/test_numpy_array.cpp
```

```
In file included from /usr/local/google/home/rwgk/forked/pybind11/tests/test_numpy_array.cpp:10:
/usr/local/google/home/rwgk/forked/pybind11/include/pybind11/numpy.h: In static member function ‘static pybind11::detail::npy_api& pybind11::detail::npy_api::get()’:
/usr/local/google/home/rwgk/forked/pybind11/include/pybind11/numpy.h:258:82: error: ‘constinit’ variable ‘api_init’ does not have a constant initializer
  258 |         PYBIND11_CONSTINIT static LazyInitializeAtLeastOnceDestroyNever<npy_api> api_init;
      |                                                                                  ^~~~~~~~
```

```
In file included from /usr/local/google/home/rwgk/forked/pybind11/tests/test_numpy_array.cpp:10:
/usr/local/google/home/rwgk/forked/pybind11/include/pybind11/numpy.h: In static member function ‘static pybind11::object& pybind11::dtype::_dtype_from_pep3118()’:
/usr/local/google/home/rwgk/forked/pybind11/include/pybind11/numpy.h:697:13: error: ‘constinit’ variable ‘imported_obj’ does not have a constant initializer
  697 |             imported_obj;
      |             ^~~~~~~~~~~~
```

* Revert "Add `PYBIND11_CONSTINIT`, but it does not work for the current use cases:"

This reverts commit f07b28bda9.

* Reapply "Add `PYBIND11_CONSTINIT`, but it does not work for the current use cases:"

This reverts commit 36be645758.

* Add Default Member Initializer on `value_storage_` as suggested by @tkoeppe:

https://github.com/pybind/pybind11/pull/4877#issuecomment-1753201342

This fixes the errors reported under commit f07b28bda9.

* Fix copy-paste-missed-a-change mishap in commit 88cec1152a.

* Semi-paranoid placement new (based on https://github.com/pybind/pybind11/pull/4877#discussion_r1350573114).

* Move PYBIND11_CONSTINIT to detail/common.h

* Move code to the right places, rename new class and some variables.

* Fix oversight: update tests/extra_python_package/test_files.py

* Get the name right first.

* Use `std::call_once`, `std::atomic`, following a pattern developed by @tkoeppe

* Make the API more self-documenting (and possibly more easily reusable).

* google-clang-tidy IWYU fixes

* Rewrite comment as suggested by @tkoeppe

* Update test_exceptions.cpp and exceptions.rst

* Fix oversight in previous commit: add `PYBIND11_CONSTINIT`

* Make `get_stored()` non-const for simplicity.

As suggested by @tkoeppe: not seeing any reasonable use in which `get_stored` has to be const.

* Add comment regarding `KeyboardInterrupt` behavior, based heavily on information provided by @jbms.

* Add `assert(PyGILState_Check())` in `gil_scoped_release` ctor (simple & non-simple implementation) as suggested by @EthanSteinberg.

* Fix oversight in previous commit (missing include cassert).

* Remove use of std::atomic, leaving comments with rationale, why it is not needed.

* Rewrite comment re `std:optional` based on deeper reflection (aka 2nd thoughts).

* Additional comment with the conclusion of a discussion under PR #4877.

* https://github.com/pybind/pybind11/pull/4877#issuecomment-1757363179

* Small comment changes suggested by @tkoeppe.
2023-10-11 21:05:31 -07:00

248 lines
8.3 KiB
C++

/*
pybind11/gil.h: RAII helpers for managing the GIL
Copyright (c) 2016 Wenzel Jakob <wenzel.jakob@epfl.ch>
All rights reserved. Use of this source code is governed by a
BSD-style license that can be found in the LICENSE file.
*/
#pragma once
#include "detail/common.h"
#include <cassert>
#if defined(WITH_THREAD) && !defined(PYBIND11_SIMPLE_GIL_MANAGEMENT)
# include "detail/internals.h"
#endif
PYBIND11_NAMESPACE_BEGIN(PYBIND11_NAMESPACE)
PYBIND11_NAMESPACE_BEGIN(detail)
// forward declarations
PyThreadState *get_thread_state_unchecked();
PYBIND11_NAMESPACE_END(detail)
#if defined(WITH_THREAD)
# if !defined(PYBIND11_SIMPLE_GIL_MANAGEMENT)
/* The functions below essentially reproduce the PyGILState_* API using a RAII
* pattern, but there are a few important differences:
*
* 1. When acquiring the GIL from an non-main thread during the finalization
* phase, the GILState API blindly terminates the calling thread, which
* is often not what is wanted. This API does not do this.
*
* 2. The gil_scoped_release function can optionally cut the relationship
* of a PyThreadState and its associated thread, which allows moving it to
* another thread (this is a fairly rare/advanced use case).
*
* 3. The reference count of an acquired thread state can be controlled. This
* can be handy to prevent cases where callbacks issued from an external
* thread would otherwise constantly construct and destroy thread state data
* structures.
*
* See the Python bindings of NanoGUI (http://github.com/wjakob/nanogui) for an
* example which uses features 2 and 3 to migrate the Python thread of
* execution to another thread (to run the event loop on the original thread,
* in this case).
*/
class gil_scoped_acquire {
public:
PYBIND11_NOINLINE gil_scoped_acquire() {
auto &internals = detail::get_internals();
tstate = (PyThreadState *) PYBIND11_TLS_GET_VALUE(internals.tstate);
if (!tstate) {
/* Check if the GIL was acquired using the PyGILState_* API instead (e.g. if
calling from a Python thread). Since we use a different key, this ensures
we don't create a new thread state and deadlock in PyEval_AcquireThread
below. Note we don't save this state with internals.tstate, since we don't
create it we would fail to clear it (its reference count should be > 0). */
tstate = PyGILState_GetThisThreadState();
}
if (!tstate) {
tstate = PyThreadState_New(internals.istate);
# if defined(PYBIND11_DETAILED_ERROR_MESSAGES)
if (!tstate) {
pybind11_fail("scoped_acquire: could not create thread state!");
}
# endif
tstate->gilstate_counter = 0;
PYBIND11_TLS_REPLACE_VALUE(internals.tstate, tstate);
} else {
release = detail::get_thread_state_unchecked() != tstate;
}
if (release) {
PyEval_AcquireThread(tstate);
}
inc_ref();
}
gil_scoped_acquire(const gil_scoped_acquire &) = delete;
gil_scoped_acquire &operator=(const gil_scoped_acquire &) = delete;
void inc_ref() { ++tstate->gilstate_counter; }
PYBIND11_NOINLINE void dec_ref() {
--tstate->gilstate_counter;
# if defined(PYBIND11_DETAILED_ERROR_MESSAGES)
if (detail::get_thread_state_unchecked() != tstate) {
pybind11_fail("scoped_acquire::dec_ref(): thread state must be current!");
}
if (tstate->gilstate_counter < 0) {
pybind11_fail("scoped_acquire::dec_ref(): reference count underflow!");
}
# endif
if (tstate->gilstate_counter == 0) {
# if defined(PYBIND11_DETAILED_ERROR_MESSAGES)
if (!release) {
pybind11_fail("scoped_acquire::dec_ref(): internal error!");
}
# endif
PyThreadState_Clear(tstate);
if (active) {
PyThreadState_DeleteCurrent();
}
PYBIND11_TLS_DELETE_VALUE(detail::get_internals().tstate);
release = false;
}
}
/// This method will disable the PyThreadState_DeleteCurrent call and the
/// GIL won't be acquired. This method should be used if the interpreter
/// could be shutting down when this is called, as thread deletion is not
/// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
/// protect subsequent code.
PYBIND11_NOINLINE void disarm() { active = false; }
PYBIND11_NOINLINE ~gil_scoped_acquire() {
dec_ref();
if (release) {
PyEval_SaveThread();
}
}
private:
PyThreadState *tstate = nullptr;
bool release = true;
bool active = true;
};
class gil_scoped_release {
public:
// PRECONDITION: The GIL must be held when this constructor is called.
explicit gil_scoped_release(bool disassoc = false) : disassoc(disassoc) {
assert(PyGILState_Check());
// `get_internals()` must be called here unconditionally in order to initialize
// `internals.tstate` for subsequent `gil_scoped_acquire` calls. Otherwise, an
// initialization race could occur as multiple threads try `gil_scoped_acquire`.
auto &internals = detail::get_internals();
// NOLINTNEXTLINE(cppcoreguidelines-prefer-member-initializer)
tstate = PyEval_SaveThread();
if (disassoc) {
// Python >= 3.7 can remove this, it's an int before 3.7
// NOLINTNEXTLINE(readability-qualified-auto)
auto key = internals.tstate;
PYBIND11_TLS_DELETE_VALUE(key);
}
}
gil_scoped_release(const gil_scoped_release &) = delete;
gil_scoped_release &operator=(const gil_scoped_release &) = delete;
/// This method will disable the PyThreadState_DeleteCurrent call and the
/// GIL won't be acquired. This method should be used if the interpreter
/// could be shutting down when this is called, as thread deletion is not
/// allowed during shutdown. Check _Py_IsFinalizing() on Python 3.7+, and
/// protect subsequent code.
PYBIND11_NOINLINE void disarm() { active = false; }
~gil_scoped_release() {
if (!tstate) {
return;
}
// `PyEval_RestoreThread()` should not be called if runtime is finalizing
if (active) {
PyEval_RestoreThread(tstate);
}
if (disassoc) {
// Python >= 3.7 can remove this, it's an int before 3.7
// NOLINTNEXTLINE(readability-qualified-auto)
auto key = detail::get_internals().tstate;
PYBIND11_TLS_REPLACE_VALUE(key, tstate);
}
}
private:
PyThreadState *tstate;
bool disassoc;
bool active = true;
};
# else // PYBIND11_SIMPLE_GIL_MANAGEMENT
class gil_scoped_acquire {
PyGILState_STATE state;
public:
gil_scoped_acquire() : state{PyGILState_Ensure()} {}
gil_scoped_acquire(const gil_scoped_acquire &) = delete;
gil_scoped_acquire &operator=(const gil_scoped_acquire &) = delete;
~gil_scoped_acquire() { PyGILState_Release(state); }
void disarm() {}
};
class gil_scoped_release {
PyThreadState *state;
public:
// PRECONDITION: The GIL must be held when this constructor is called.
gil_scoped_release() {
assert(PyGILState_Check());
state = PyEval_SaveThread();
}
gil_scoped_release(const gil_scoped_release &) = delete;
gil_scoped_release &operator=(const gil_scoped_release &) = delete;
~gil_scoped_release() { PyEval_RestoreThread(state); }
void disarm() {}
};
# endif // PYBIND11_SIMPLE_GIL_MANAGEMENT
#else // WITH_THREAD
class gil_scoped_acquire {
public:
gil_scoped_acquire() {
// Trick to suppress `unused variable` error messages (at call sites).
(void) (this != (this + 1));
}
gil_scoped_acquire(const gil_scoped_acquire &) = delete;
gil_scoped_acquire &operator=(const gil_scoped_acquire &) = delete;
void disarm() {}
};
class gil_scoped_release {
public:
gil_scoped_release() {
// Trick to suppress `unused variable` error messages (at call sites).
(void) (this != (this + 1));
}
gil_scoped_release(const gil_scoped_release &) = delete;
gil_scoped_release &operator=(const gil_scoped_release &) = delete;
void disarm() {}
};
#endif // WITH_THREAD
PYBIND11_NAMESPACE_END(PYBIND11_NAMESPACE)