osb/source/core/StarThread.hpp
Kai Blaschke 431a9c00a5
Fixed a huge amount of Clang warnings
On Linux and macOS, using Clang to compile OpenStarbound produces about 400 MB worth of warnings during the build, making the compiler output unreadable and slowing the build down considerably.

99% of the warnings were unqualified uses of std::move and std::forward, which are now all properly qualified.

Fixed a few other minor warnings about non-virtual destructors and some uses of std::move preventing copy elision on temporary objects.

Most remaining warnings are now unused parameters.
2024-02-19 16:55:19 +01:00

426 lines
11 KiB
C++

#ifndef STAR_THREAD_HPP
#define STAR_THREAD_HPP
#include "StarException.hpp"
#include "StarString.hpp"
namespace Star {
STAR_STRUCT(ThreadImpl);
STAR_STRUCT(ThreadFunctionImpl);
STAR_STRUCT(MutexImpl);
STAR_STRUCT(ConditionVariableImpl);
STAR_STRUCT(RecursiveMutexImpl);
template <typename Return>
class ThreadFunction;
class Thread {
public:
// Implementations of this method should sleep for at least the given amount
// of time, but may sleep for longer due to scheduling.
static void sleep(unsigned millis);
// Sleep a more precise amount of time, but uses more resources to do so.
// Should be less likely to sleep much longer than the given amount of time.
static void sleepPrecise(unsigned millis);
// Yield this thread, offering the opportunity to reschedule.
static void yield();
static unsigned numberOfProcessors();
template <typename Function, typename... Args>
static ThreadFunction<decltype(std::declval<Function>()(std::declval<Args>()...))> invoke(String const& name, Function&& f, Args&&... args);
Thread(String const& name);
Thread(Thread&&);
// Will not automatically join! ALL implementations of this class MUST call
// join() in their most derived constructors, or not rely on the destructor
// joining.
virtual ~Thread();
Thread& operator=(Thread&&);
// Start a thread that is currently in the joined state. Returns true if the
// thread was joined and is now started, false if the thread was not joined.
bool start();
// Wait for a thread to finish and re-join with the thread, on completion
// isJoined() will be false. Returns true if the thread was joinable, and is
// now joined, false if the thread was already joined.
bool join();
// Returns false when this thread been started without being joined. This is
// subtlely different than "!isRunning()", in that the thread could have
// completed its work, but a thread *must* be joined before being restarted.
bool isJoined() const;
// Returns false before start() has been called, true immediately after
// start() has been called, and false once the run() method returns.
bool isRunning() const;
String name();
protected:
virtual void run() = 0;
private:
unique_ptr<ThreadImpl> m_impl;
};
// Wraps a function call and calls in another thread, very nice lightweight
// one-shot alternative to deriving from Thread. Handles exceptions in a
// different way from Thread, instead of logging the exception, the exception
// is forwarded and re-thrown during the call to finish().
template <>
class ThreadFunction<void> {
public:
ThreadFunction();
ThreadFunction(ThreadFunction&&);
// Automatically starts the given function, ThreadFunction can also be
// constructed with Thread::invoke, which is a shorthand.
ThreadFunction(function<void()> function, String const& name);
// Automatically calls finish, though BEWARE that often times this is quite
// dangerous, and this is here mostly as a fallback. The natural destructor
// order for members of a class is often wrong, and if the function throws,
// since this destructor calls finish it will throw.
~ThreadFunction();
ThreadFunction& operator=(ThreadFunction&&);
// Waits on function finish if function is assigned and started, otherwise
// does nothing. If the function threw an exception, it will be re-thrown
// here (on the first call to finish() only).
void finish();
// Returns whether the ThreadFunction::finish method been called and the
// ThreadFunction has stopped. Also returns true when the ThreadFunction has
// been default constructed.
bool isFinished() const;
// Returns false if the thread function has stopped running, whether or not
// finish() has been called.
bool isRunning() const;
// Equivalent to !isFinished()
explicit operator bool() const;
String name();
private:
unique_ptr<ThreadFunctionImpl> m_impl;
};
template <typename Return>
class ThreadFunction {
public:
ThreadFunction();
ThreadFunction(ThreadFunction&&);
ThreadFunction(function<Return()> function, String const& name);
~ThreadFunction();
ThreadFunction& operator=(ThreadFunction&&);
// Finishes the thread, moving and returning the final value of the function.
// If the function threw an exception, finish() will rethrow that exception.
// May only be called once, otherwise will throw InvalidMaybeAccessException.
Return finish();
bool isFinished() const;
bool isRunning() const;
explicit operator bool() const;
String name();
private:
ThreadFunction<void> m_function;
shared_ptr<Maybe<Return>> m_return;
};
// *Non* recursive mutex lock, for use with ConditionVariable
class Mutex {
public:
Mutex();
Mutex(Mutex&&);
~Mutex();
Mutex& operator=(Mutex&&);
void lock();
// Attempt to acquire the mutex without blocking.
bool tryLock();
void unlock();
private:
friend struct ConditionVariableImpl;
unique_ptr<MutexImpl> m_impl;
};
class ConditionVariable {
public:
ConditionVariable();
ConditionVariable(ConditionVariable&&);
~ConditionVariable();
ConditionVariable& operator=(ConditionVariable&&);
// Atomically unlocks the mutex argument and waits on the condition. On
// acquiring the condition, atomically returns and re-locks the mutex. Must
// lock the mutex before calling. If millis is given, waits for a maximum of
// the given milliseconds only.
void wait(Mutex& mutex, Maybe<unsigned> millis = {});
// Wake one waiting thread. The calling thread for is allowed to either hold
// or not hold the mutex that the threads waiting on the condition are using,
// both will work and result in slightly different scheduling.
void signal();
// Wake all threads, policy for holding the mutex is the same for signal().
void broadcast();
private:
unique_ptr<ConditionVariableImpl> m_impl;
};
// Recursive mutex lock. lock() may be called many times freely by the same
// thread, but unlock() must be called an equal number of times to unlock it.
class RecursiveMutex {
public:
RecursiveMutex();
RecursiveMutex(RecursiveMutex&&);
~RecursiveMutex();
RecursiveMutex& operator=(RecursiveMutex&&);
void lock();
// Attempt to acquire the mutex without blocking.
bool tryLock();
void unlock();
private:
unique_ptr<RecursiveMutexImpl> m_impl;
};
// RAII for mutexes. Locking and unlocking are always safe, MLocker will never
// attempt to lock the held mutex more than once, or unlock more than once, and
// destruction will always unlock the mutex *iff* it is actually locked.
// (Locked here refers to one specific MLocker *itself* locking the mutex, not
// whether the mutex is locked *at all*, so it is sensible to use with
// RecursiveMutex)
template <typename MutexType>
class MLocker {
public:
// Pass false to lock to start unlocked
MLocker(MutexType& ref, bool lock = true);
~MLocker();
MLocker(MLocker const&) = delete;
MLocker& operator=(MLocker const&) = delete;
MutexType& mutex();
void unlock();
void lock();
bool tryLock();
private:
MutexType& m_mutex;
bool m_locked;
};
typedef MLocker<Mutex> MutexLocker;
typedef MLocker<RecursiveMutex> RecursiveMutexLocker;
class ReadersWriterMutex {
public:
ReadersWriterMutex();
void readLock();
bool tryReadLock();
void readUnlock();
void writeLock();
bool tryWriteLock();
void writeUnlock();
private:
Mutex m_mutex;
ConditionVariable m_readCond;
ConditionVariable m_writeCond;
unsigned m_readers;
unsigned m_writers;
unsigned m_readWaiters;
unsigned m_writeWaiters;
};
class ReadLocker {
public:
ReadLocker(ReadersWriterMutex& rwlock, bool startLocked = true);
~ReadLocker();
ReadLocker(ReadLocker const&) = delete;
ReadLocker& operator=(ReadLocker const&) = delete;
void unlock();
void lock();
bool tryLock();
private:
ReadersWriterMutex& m_lock;
bool m_locked;
};
class WriteLocker {
public:
WriteLocker(ReadersWriterMutex& rwlock, bool startLocked = true);
~WriteLocker();
WriteLocker(WriteLocker const&) = delete;
WriteLocker& operator=(WriteLocker const&) = delete;
void unlock();
void lock();
bool tryLock();
private:
ReadersWriterMutex& m_lock;
bool m_locked;
};
class SpinLock {
public:
SpinLock();
void lock();
bool tryLock();
void unlock();
private:
atomic_flag m_lock;
};
typedef MLocker<SpinLock> SpinLocker;
template <typename MutexType>
MLocker<MutexType>::MLocker(MutexType& ref, bool l)
: m_mutex(ref), m_locked(false) {
if (l)
lock();
}
template <typename MutexType>
MLocker<MutexType>::~MLocker() {
unlock();
}
template <typename MutexType>
MutexType& MLocker<MutexType>::mutex() {
return m_mutex;
}
template <typename MutexType>
void MLocker<MutexType>::unlock() {
if (m_locked) {
m_mutex.unlock();
m_locked = false;
}
}
template <typename MutexType>
void MLocker<MutexType>::lock() {
if (!m_locked) {
m_mutex.lock();
m_locked = true;
}
}
template <typename MutexType>
bool MLocker<MutexType>::tryLock() {
if (!m_locked) {
if (m_mutex.tryLock())
m_locked = true;
}
return m_locked;
}
template <typename Function, typename... Args>
ThreadFunction<decltype(std::declval<Function>()(std::declval<Args>()...))> Thread::invoke(String const& name, Function&& f, Args&&... args) {
return {bind(std::forward<Function>(f), std::forward<Args>(args)...), name};
}
template <typename Return>
ThreadFunction<Return>::ThreadFunction() {}
template <typename Return>
ThreadFunction<Return>::ThreadFunction(ThreadFunction&&) = default;
template <typename Return>
ThreadFunction<Return>::ThreadFunction(function<Return()> function, String const& name) {
m_return = make_shared<Maybe<Return>>();
m_function = ThreadFunction<void>([function = std::move(function), retValue = m_return]() {
*retValue = function();
}, name);
}
template <typename Return>
ThreadFunction<Return>::~ThreadFunction() {
m_function.finish();
}
template <typename Return>
ThreadFunction<Return>& ThreadFunction<Return>::operator=(ThreadFunction&&) = default;
template <typename Return>
Return ThreadFunction<Return>::finish() {
m_function.finish();
return m_return->take();
}
template <typename Return>
bool ThreadFunction<Return>::isFinished() const {
return m_function.isFinished();
}
template <typename Return>
bool ThreadFunction<Return>::isRunning() const {
return m_function.isRunning();
}
template <typename Return>
ThreadFunction<Return>::operator bool() const {
return !isFinished();
}
template <typename Return>
String ThreadFunction<Return>::name() {
return m_function.name();
}
inline SpinLock::SpinLock() {
m_lock.clear();
}
inline void SpinLock::lock() {
while (m_lock.test_and_set(std::memory_order_acquire))
;
}
inline void SpinLock::unlock() {
m_lock.clear(std::memory_order_release);
}
inline bool SpinLock::tryLock() {
return !m_lock.test_and_set(std::memory_order_acquire);
}
}
#endif