#pragma GCC diagnostic ignored "-Wunused-value"
#endif
-#ifndef NDEBUG
-/* Many tests are failing after pybind11 commit ad6bf5cd39ca64b4a9bf846b84b11c4c8df1c8e1 "Adding PyGILState_Check() in
- * object_api<>::operator(). (#2919)".
- * See https://github.com/pybind/pybind11/commit/ad6bf5cd39ca64b4a9bf846b84b11c4c8df1c8e1
- *
- * The failing tests are mostly those with boost/raw/sysv contexts. As a workaround, define NDEBUG before pybind11
- * includes.
- */
-#define NDEBUG
-#define NDEBUG_LOCALLY_DEFINED
-#endif
-
#include <pybind11/pybind11.h> // Must come before our own stuff
#include <pybind11/functional.h>
#include <pybind11/stl.h>
-#ifdef NDEBUG_LOCALLY_DEFINED
-#undef NDEBUG_LOCALLY_DEFINED
-#undef NDEBUG
-#endif
-
#if defined(__GNUG__)
#pragma GCC diagnostic pop
#endif
PyObject** get() const { return data.get(); }
};
-/* Classes GilScopedAcquire and GilScopedRelease have the same purpose as pybind11::gil_scoped_acquire and
- * pybind11::gil_scoped_release. Refer to the manual of pybind11 for details:
- * https://pybind11.readthedocs.io/en/stable/advanced/misc.html#global-interpreter-lock-gil
- *
- * The pybind11 versions are however too sophisticated (using TLS for example) and don't work well with all kinds of
- * contexts.
- * See also https://github.com/pybind/pybind11/issues/1276, which may be related.
- *
- * Briefly, GilScopedAcquire can be used on actor creation to acquire a new PyThreadState. The PyThreadState has to be
- * released for context switches (i.e. before simcalls). That's the purpose of GilScopedRelease.
- *
- * Like their pybind11 counterparts, both classes use a RAII pattern.
- */
-class XBT_PRIVATE GilScopedAcquire {
- static PyThreadState* acquire()
- {
- PyThreadState* state = PyThreadState_New(PyInterpreterState_Head());
- PyEval_AcquireThread(state);
- return state;
- }
- static void release(PyThreadState* state)
- {
- PyEval_ReleaseThread(state);
- PyThreadState_Clear(state);
- PyThreadState_Delete(state);
- }
-
- std::unique_ptr<PyThreadState, decltype(&release)> thread_state{acquire(), &release};
-};
-
-class XBT_PRIVATE GilScopedRelease {
- std::unique_ptr<PyThreadState, decltype(&PyEval_RestoreThread)> thread_state{PyEval_SaveThread(),
- &PyEval_RestoreThread};
-};
-
} // namespace
PYBIND11_DECLARE_HOLDER_TYPE(T, boost::intrusive_ptr<T>)
.def(
"error", [](const char* s) { XBT_ERROR("%s", s); }, "Display a logging message of 'error' priority.")
.def("execute", py::overload_cast<double, double>(&simgrid::s4u::this_actor::execute),
- py::call_guard<GilScopedRelease>(),
+ py::call_guard<py::gil_scoped_release>(),
"Block the current actor, computing the given amount of flops at the given priority.", py::arg("flops"),
py::arg("priority") = 1)
.def("exec_init", py::overload_cast<double>(&simgrid::s4u::this_actor::exec_init),
- py::call_guard<GilScopedRelease>())
+ py::call_guard<py::gil_scoped_release>())
.def("get_host", &simgrid::s4u::this_actor::get_host, "Retrieves host on which the current actor is located")
- .def("set_host", &simgrid::s4u::this_actor::set_host, py::call_guard<GilScopedRelease>(),
+ .def("set_host", &simgrid::s4u::this_actor::set_host, py::call_guard<py::gil_scoped_release>(),
"Moves the current actor to another host.", py::arg("dest"))
.def("sleep_for", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_for),
- py::call_guard<GilScopedRelease>(), "Block the actor sleeping for that amount of seconds.",
+ py::call_guard<py::gil_scoped_release>(), "Block the actor sleeping for that amount of seconds.",
py::arg("duration"))
.def("sleep_until", static_cast<void (*)(double)>(&simgrid::s4u::this_actor::sleep_until),
- py::call_guard<GilScopedRelease>(), "Block the actor sleeping until the specified timestamp.",
+ py::call_guard<py::gil_scoped_release>(), "Block the actor sleeping until the specified timestamp.",
py::arg("duration"))
- .def("suspend", &simgrid::s4u::this_actor::suspend, py::call_guard<GilScopedRelease>(),
+ .def("suspend", &simgrid::s4u::this_actor::suspend, py::call_guard<py::gil_scoped_release>(),
"Suspend the current actor, that is blocked until resume()ed by another actor.")
- .def("yield_", &simgrid::s4u::this_actor::yield, py::call_guard<GilScopedRelease>(), "Yield the actor")
- .def("exit", &simgrid::s4u::this_actor::exit, py::call_guard<GilScopedRelease>(), "kill the current actor")
+ .def("yield_", &simgrid::s4u::this_actor::yield, py::call_guard<py::gil_scoped_release>(), "Yield the actor")
+ .def("exit", &simgrid::s4u::this_actor::exit, py::call_guard<py::gil_scoped_release>(), "kill the current actor")
.def(
"on_exit",
[](py::object fun) {
fun.inc_ref(); // FIXME: why is this needed for tests like actor-kill and actor-lifetime?
simgrid::s4u::this_actor::on_exit([fun](bool /*failed*/) {
try {
- GilScopedAcquire py_context; // need a new context for callback
+ py::gil_scoped_acquire py_context; // need a new context for callback
fun();
} catch (const py::error_already_set& e) {
xbt_die("Error while executing the on_exit lambda: %s", e.what());
}
});
},
- py::call_guard<GilScopedRelease>(), "");
+ py::call_guard<py::gil_scoped_release>(), "");
/* Class Engine */
py::class_<Engine>(m, "Engine", "Simulation Engine")
.def("get_all_hosts", &Engine::get_all_hosts, "Returns the list of all hosts found in the platform")
.def("load_platform", &Engine::load_platform, "Load a platform file describing the environment")
.def("load_deployment", &Engine::load_deployment, "Load a deployment file and launch the actors that it contains")
- .def("run", &Engine::run, py::call_guard<GilScopedRelease>(), "Run the simulation")
+ .def("run", &Engine::run, py::call_guard<py::gil_scoped_release>(), "Run the simulation")
.def(
"register_actor",
[](Engine* e, const std::string& name, py::object fun_or_class) {
e->register_actor(name, [fun_or_class](std::vector<std::string> args) {
try {
- GilScopedAcquire py_context;
+ py::gil_scoped_acquire py_context;
/* Convert the std::vector into a py::tuple */
py::tuple params(args.size() - 1);
for (size_t i = 1; i < args.size(); i++)
.def_property(
"pstate", &Host::get_pstate,
[](Host* h, int i) {
- GilScopedRelease gil_guard;
+ py::gil_scoped_release gil_guard;
h->set_pstate(i);
},
"The current pstate")
- .def("current", &Host::current, py::call_guard<GilScopedRelease>(),
+ .def("current", &Host::current, py::call_guard<py::gil_scoped_release>(),
"Retrieves the host on which the running actor is located.")
.def_property_readonly(
"name",
/* Class Disk */
py::class_<simgrid::s4u::Disk, std::unique_ptr<simgrid::s4u::Disk, py::nodelete>> disk(m, "Disk", "Simulated disk");
- disk.def("read", &simgrid::s4u::Disk::read, py::call_guard<GilScopedRelease>(), "Read data from disk")
- .def("write", &simgrid::s4u::Disk::write, py::call_guard<GilScopedRelease>(), "Write data in disk")
+ disk.def("read", &simgrid::s4u::Disk::read, py::call_guard<py::gil_scoped_release>(), "Read data from disk")
+ .def("write", &simgrid::s4u::Disk::write, py::call_guard<py::gil_scoped_release>(), "Write data in disk")
.def("read_async", &simgrid::s4u::Disk::read_async, "Non-blocking read data from disk")
.def("write_async", &simgrid::s4u::Disk::write_async, "Non-blocking write data in disk")
.def("set_sharing_policy", &simgrid::s4u::Disk::set_sharing_policy, "Set sharing policy for this disk",
.def(
"__str__", [](const Mailbox* self) { return std::string("Mailbox(") + self->get_cname() + ")"; },
"Textual representation of the Mailbox`")
- .def("by_name", &Mailbox::by_name, py::call_guard<GilScopedRelease>(), "Retrieve a Mailbox from its name")
+ .def("by_name", &Mailbox::by_name, py::call_guard<py::gil_scoped_release>(), "Retrieve a Mailbox from its name")
.def_property_readonly(
"name",
[](const Mailbox* self) {
data.inc_ref();
self->put(data.ptr(), size);
},
- py::call_guard<GilScopedRelease>(), "Blocking data transmission")
+ py::call_guard<py::gil_scoped_release>(), "Blocking data transmission")
.def(
"put_async",
[](Mailbox* self, py::object data, int size) {
data.inc_ref();
return self->put_async(data.ptr(), size);
},
- py::call_guard<GilScopedRelease>(), "Non-blocking data transmission")
+ py::call_guard<py::gil_scoped_release>(), "Non-blocking data transmission")
.def(
"get",
[](Mailbox* self) {
// data.dec_ref(); // FIXME: why does it break python-actor-create?
return data;
},
- py::call_guard<GilScopedRelease>(), "Blocking data reception")
+ py::call_guard<py::gil_scoped_release>(), "Blocking data reception")
.def(
"get_async",
[](Mailbox* self) -> std::tuple<simgrid::s4u::CommPtr, PyGetAsync> {
auto comm = self->get_async(wrap.get());
return std::make_tuple(std::move(comm), std::move(wrap));
},
- py::call_guard<GilScopedRelease>(),
+ py::call_guard<py::gil_scoped_release>(),
"Non-blocking data reception. Use data.get() to get the python object after the communication has finished")
.def(
"set_receiver", [](Mailbox* self, ActorPtr actor) { self->set_receiver(actor); },
- py::call_guard<GilScopedRelease>(), "Sets the actor as permanent receiver");
+ py::call_guard<py::gil_scoped_release>(), "Sets the actor as permanent receiver");
/* Class PyGetAsync */
py::class_<PyGetAsync>(m, "PyGetAsync", "Wrapper for async get communications")
/* Class Comm */
py::class_<simgrid::s4u::Comm, simgrid::s4u::CommPtr>(m, "Comm", "Communication")
- .def("test", &simgrid::s4u::Comm::test, py::call_guard<GilScopedRelease>(),
+ .def("test", &simgrid::s4u::Comm::test, py::call_guard<py::gil_scoped_release>(),
"Test whether the communication is terminated.")
- .def("wait", &simgrid::s4u::Comm::wait, py::call_guard<GilScopedRelease>(),
+ .def("wait", &simgrid::s4u::Comm::wait, py::call_guard<py::gil_scoped_release>(),
"Block until the completion of that communication.")
// use py::overload_cast for wait_all/wait_any, until the overload marked XBT_ATTRIB_DEPRECATED_v332 is removed
- .def_static("wait_all",
- py::overload_cast<const std::vector<simgrid::s4u::CommPtr>&>(&simgrid::s4u::Comm::wait_all),
- py::call_guard<GilScopedRelease>(), "Block until the completion of all communications in the list.")
+ .def_static(
+ "wait_all", py::overload_cast<const std::vector<simgrid::s4u::CommPtr>&>(&simgrid::s4u::Comm::wait_all),
+ py::call_guard<py::gil_scoped_release>(), "Block until the completion of all communications in the list.")
.def_static(
"wait_any", py::overload_cast<const std::vector<simgrid::s4u::CommPtr>&>(&simgrid::s4u::Comm::wait_any),
- py::call_guard<GilScopedRelease>(),
+ py::call_guard<py::gil_scoped_release>(),
"Block until the completion of any communication in the list and return the index of the terminated one.");
/* Class Io */
py::class_<simgrid::s4u::Io, simgrid::s4u::IoPtr>(m, "Io", "I/O activities")
- .def("test", &simgrid::s4u::Io::test, py::call_guard<GilScopedRelease>(), "Test whether the I/O is terminated.")
- .def("wait", &simgrid::s4u::Io::wait, py::call_guard<GilScopedRelease>(),
+ .def("test", &simgrid::s4u::Io::test, py::call_guard<py::gil_scoped_release>(),
+ "Test whether the I/O is terminated.")
+ .def("wait", &simgrid::s4u::Io::wait, py::call_guard<py::gil_scoped_release>(),
"Block until the completion of that I/O operation")
.def_static(
- "wait_any_for", &simgrid::s4u::Io::wait_any_for, py::call_guard<GilScopedRelease>(),
+ "wait_any_for", &simgrid::s4u::Io::wait_any_for, py::call_guard<py::gil_scoped_release>(),
"Block until the completion of any I/O in the list (or timeout) and return the index of the terminated one.")
- .def_static("wait_any", &simgrid::s4u::Io::wait_any, py::call_guard<GilScopedRelease>(),
+ .def_static("wait_any", &simgrid::s4u::Io::wait_any, py::call_guard<py::gil_scoped_release>(),
"Block until the completion of any I/O in the list and return the index of the terminated one.");
/* Class Exec */
.def_property_readonly(
"remaining",
[](simgrid::s4u::ExecPtr self) {
- GilScopedRelease gil_guard;
+ py::gil_scoped_release gil_guard;
return self->get_remaining();
},
"Amount of flops that remain to be computed until completion.")
.def_property_readonly(
"remaining_ratio",
[](simgrid::s4u::ExecPtr self) {
- GilScopedRelease gil_guard;
+ py::gil_scoped_release gil_guard;
return self->get_remaining_ratio();
},
"Amount of work remaining until completion from 0 (completely done) to 1 (nothing done "
"yet).")
.def_property("host", &simgrid::s4u::Exec::get_host, &simgrid::s4u::Exec::set_host,
"Host on which this execution runs. Only the first host is returned for parallel executions.")
- .def("test", &simgrid::s4u::Exec::test, py::call_guard<GilScopedRelease>(),
+ .def("test", &simgrid::s4u::Exec::test, py::call_guard<py::gil_scoped_release>(),
"Test whether the execution is terminated.")
- .def("cancel", &simgrid::s4u::Exec::cancel, py::call_guard<GilScopedRelease>(), "Cancel that execution.")
- .def("start", &simgrid::s4u::Exec::start, py::call_guard<GilScopedRelease>(), "Start that execution.")
- .def("wait", &simgrid::s4u::Exec::wait, py::call_guard<GilScopedRelease>(),
+ .def("cancel", &simgrid::s4u::Exec::cancel, py::call_guard<py::gil_scoped_release>(), "Cancel that execution.")
+ .def("start", &simgrid::s4u::Exec::start, py::call_guard<py::gil_scoped_release>(), "Start that execution.")
+ .def("wait", &simgrid::s4u::Exec::wait, py::call_guard<py::gil_scoped_release>(),
"Block until the completion of that execution.");
/* Class Actor */
args.inc_ref(); // FIXME: why is this needed for tests like actor-migrate?
return simgrid::s4u::Actor::create(name, h, [fun, args]() {
try {
- GilScopedAcquire py_context;
+ py::gil_scoped_acquire py_context;
fun(*args);
} catch (const py::error_already_set& ex) {
if (ex.matches(pyForcefulKillEx)) {
}
});
},
- py::call_guard<GilScopedRelease>(), "Create an actor from a function or an object.")
+ py::call_guard<py::gil_scoped_release>(), "Create an actor from a function or an object.")
.def_property(
"host", &Actor::get_host,
[](Actor* a, Host* h) {
- GilScopedRelease gil_guard;
+ py::gil_scoped_release gil_guard;
a->set_host(h);
},
"The host on which this actor is located")
.def_property_readonly("ppid", &Actor::get_ppid,
"The PID (unique identifier) of the actor that created this one.")
.def("by_pid", &Actor::by_pid, "Retrieve an actor by its PID")
- .def("daemonize", &Actor::daemonize, py::call_guard<GilScopedRelease>(),
+ .def("daemonize", &Actor::daemonize, py::call_guard<py::gil_scoped_release>(),
"This actor will be automatically terminated when the last non-daemon actor finishes (more info in the C++ "
"documentation).")
.def("is_daemon", &Actor::is_daemon,
"Returns True if that actor is a daemon and will be terminated automatically when the last non-daemon actor "
"terminates.")
- .def("join", py::overload_cast<double>(&Actor::join, py::const_), py::call_guard<GilScopedRelease>(),
+ .def("join", py::overload_cast<double>(&Actor::join, py::const_), py::call_guard<py::gil_scoped_release>(),
"Wait for the actor to finish (more info in the C++ documentation).", py::arg("timeout"))
- .def("kill", &Actor::kill, py::call_guard<GilScopedRelease>(), "Kill that actor")
- .def("kill_all", &Actor::kill_all, py::call_guard<GilScopedRelease>(), "Kill all actors but the caller.")
+ .def("kill", &Actor::kill, py::call_guard<py::gil_scoped_release>(), "Kill that actor")
+ .def("kill_all", &Actor::kill_all, py::call_guard<py::gil_scoped_release>(), "Kill all actors but the caller.")
.def("self", &Actor::self, "Retrieves the current actor.")
.def("is_suspended", &Actor::is_suspended, "Returns True if that actor is currently suspended.")
- .def("suspend", &Actor::suspend, py::call_guard<GilScopedRelease>(),
+ .def("suspend", &Actor::suspend, py::call_guard<py::gil_scoped_release>(),
"Suspend that actor, that is blocked until resume()ed by another actor.")
- .def("resume", &Actor::resume, py::call_guard<GilScopedRelease>(),
+ .def("resume", &Actor::resume, py::call_guard<py::gil_scoped_release>(),
"Resume that actor, that was previously suspend()ed.");
}
