//C- -*- C++ -*- //C- ------------------------------------------------------------------- //C- DjVuLibre-3.5 //C- Copyright (c) 2002 Leon Bottou and Yann Le Cun. //C- Copyright (c) 2001 AT&T //C- //C- This software is subject to, and may be distributed under, the //C- GNU General Public License, Version 2. The license should have //C- accompanied the software or you may obtain a copy of the license //C- from the Free Software Foundation at http://www.fsf.org . //C- //C- This program is distributed in the hope that it will be useful, //C- but WITHOUT ANY WARRANTY; without even the implied warranty of //C- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the //C- GNU General Public License for more details. //C- //C- DjVuLibre-3.5 is derived from the DjVu(r) Reference Library //C- distributed by Lizardtech Software. On July 19th 2002, Lizardtech //C- Software authorized us to replace the original DjVu(r) Reference //C- Library notice by the following text (see doc/lizard2002.djvu): //C- //C- ------------------------------------------------------------------ //C- | DjVu (r) Reference Library (v. 3.5) //C- | Copyright (c) 1999-2001 LizardTech, Inc. All Rights Reserved. //C- | The DjVu Reference Library is protected by U.S. Pat. No. //C- | 6,058,214 and patents pending. //C- | //C- | This software is subject to, and may be distributed under, the //C- | GNU General Public License, Version 2. The license should have //C- | accompanied the software or you may obtain a copy of the license //C- | from the Free Software Foundation at http://www.fsf.org . //C- | //C- | The computer code originally released by LizardTech under this //C- | license and unmodified by other parties is deemed "the LIZARDTECH //C- | ORIGINAL CODE." Subject to any third party intellectual property //C- | claims, LizardTech grants recipient a worldwide, royalty-free, //C- | non-exclusive license to make, use, sell, or otherwise dispose of //C- | the LIZARDTECH ORIGINAL CODE or of programs derived from the //C- | LIZARDTECH ORIGINAL CODE in compliance with the terms of the GNU //C- | General Public License. This grant only confers the right to //C- | infringe patent claims underlying the LIZARDTECH ORIGINAL CODE to //C- | the extent such infringement is reasonably necessary to enable //C- | recipient to make, have made, practice, sell, or otherwise dispose //C- | of the LIZARDTECH ORIGINAL CODE (or portions thereof) and not to //C- | any greater extent that may be necessary to utilize further //C- | modifications or combinations. //C- | //C- | The LIZARDTECH ORIGINAL CODE is provided "AS IS" WITHOUT WARRANTY //C- | OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED //C- | TO ANY WARRANTY OF NON-INFRINGEMENT, OR ANY IMPLIED WARRANTY OF //C- | MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. //C- +------------------------------------------------------------------ // // $Id: GThreads.h,v 1.10 2003/11/07 22:08:21 leonb Exp $ // $Name: release_3_5_15 $ #ifndef _GTHREADS_H_ #define _GTHREADS_H_ #ifdef HAVE_CONFIG_H #include "config.h" #endif #if NEED_GNUG_PRAGMAS # pragma interface #endif /** @name GThreads.h Files #"GThreads.h"# and #"GThreads.cpp"# implement common entry points for multithreading on multiple platforms. Each execution thread is represented by an instance of class \Ref{GThread}. Synchronization is provided by class \Ref{GMonitor} which implements a monitor (C.A.R Hoare, Communications of the ACM, 17(10), 1974). The value of compiler symbol #THREADMODEL# selects an appropriate implementation for these classes. The current implementation supports the following values: \begin{description} \item[-DTHREADMODEL=NOTHREADS] Dummy implementation. This is a good choice when the multithreading features are not required, because it minimizes the portability problems. This is currently the default when compiling under Unix. \item[-DTHREADMODEL=WINTHREADS] Windows implementation. This is the default when compiling under Windows. \item[-DTHREADMODEL=MACTHREADS] Macintosh implementation, which is based on the MacOS cooperative model. The current implementation does not yet fully support synchronization. This is the default when compiling under MacOS. \item[-DTHREADMODEL=POSIXTHREADS] Posix implementation. This implementation also supports DCE threads. The behavior of the code is subject to the quality of the system implementation of Posix threads. \item[-DTHREADMODEL=COTHREADS] Custom cooperative threads. These custom threads do not redefine system calls. Before executing a potentially blocking system function, each thread must explicitly check whether it is going to block and yield control explicitly if this is the case. This code must be compiled with a patched version of egcs-1.1.1 \URL{http://egcs.cygnus.com}. The patch addresses exception thread-safety and is provided in #"@Tools/libgcc2.c.diff"#. Once you get the right compiler, this implementation is remarkably compact and portable. A variety of processors are supported, including mips, intel, sparc, hppa, and alpha. \item[-DTHREADMODEL=JRITHREADS] Java implementation hooks. Multi-threading within a Netscape plugin can be tricky. A simple idea however consists of implementing the threading primitives in Java and to access them using JRI. The classes just contain a JRIGlobalRef. This is not a real implementation since everything (Java code, native functions, stubs, exception thread safety) must be addressed by the plugin source code. Performance may be a serious issue. \end{description} {\bf Portability}: The simultaneous use of threads and exceptions caused a lot of portability headaches under Unix. We eventually decided to implement the COTHREADS cooperative threads (because preemptive threads have more problems) and to patch EGCS in order to make exception handling COTHREAD-safe. @memo Portable threads @author L\'eon Bottou -- initial implementation.\\ Praveen Guduru -- mac implementation. // From: Leon Bottou, 1/31/2002 // Almost unchanged by Lizardtech. // GSafeFlags should go because it not as safe as it claims. @version #$Id: GThreads.h,v 1.10 2003/11/07 22:08:21 leonb Exp $# */ //@{ #include "DjVuGlobal.h" #include "GException.h" #define NOTHREADS 0 #define COTHREADS 1 #define JRITHREADS 2 #define POSIXTHREADS 10 #define WINTHREADS 11 #define MACTHREADS 12 // Known platforms #ifndef THREADMODEL #if defined(WIN32) #define THREADMODEL WINTHREADS #endif #if defined(macintosh) #define THREADMODEL MACTHREADS #endif #endif // Exception emulation is not thread safe #ifdef USE_EXCEPTION_EMULATION #undef THREADMODEL #define THREADMODEL NOTHREADS #endif // Default is nothreads #ifndef THREADMODEL #define THREADMODEL NOTHREADS #endif // ---------------------------------------- // INCLUDES #if THREADMODEL==WINTHREADS #ifndef _WINDOWS_ #define WIN32_LEAN_AND_MEAN #include "windows.h" #endif #endif #if THREADMODEL==MACTHREADS #include #endif #if THREADMODEL==POSIXTHREADS #include #include #include #undef TRY #undef CATCH #define _CMA_NOWRAPPERS_ #include #endif #if THREADMODEL==JRITHREADS #include "jri.h" #endif #if THREADMODEL==COTHREADS #include #include #include #endif // ---------------------------------------- // PORTABLE CLASSES #ifdef HAVE_NAMESPACES namespace DJVU { # ifdef NOT_DEFINED // Just to fool emacs c++ mode } #endif #endif /** Thread class. A multithreaded process is composed of a main execution thread and of several secondary threads. Each secondary thread is represented by a #GThread# object. The amount of memory required for the stack of a secondary thread is defined when the #GThread# object is constructed. The execution thread is started when function \Ref{GThread::create} is called. The destructor of class GThread waits until the thread terminanes. Note that the execution can be terminated at any time (with possible prejudice) by calling \Ref{GThread::terminate}. Several static member functions control the thread scheduler. Function \Ref{GThread::yield} relinquishes the processor to another thread. Function \Ref{GThread::select} (#COTHREADS# only) provides a thread-aware replacement for the well-known unix system call #select#. {\bf Note} --- Both the copy constructor and the copy operator are declared as private members. It is therefore not possible to make multiple copies of instances of this class, as implied by the class semantic. */ class GThread { public: /** Constructs a new thread object. Memory is allocated for the thread, but the thread is not started. Argument #stacksize# is used by the #COTHREADS# model only for specifying the amount of memory needed for the processor stack. A negative value will be replaced by a suitable default value of 128Kb. A minimum value of 32Kb is silently enforced. */ GThread(int stacksize = -1); /** Destructor. Destroying the thread object while the thread is running is perfectly ok since it only destroys the thread identifier. Execution will continue without interference. */ ~GThread(); /** Starts the thread. The new thread executes function #entry# with argument #arg#. The thread terminates when the function returns. A thread cannot be restarted after its termination. You must create a new #GThread# object. */ int create(void (*entry)(void*), void *arg); /** Terminates a thread with extreme prejudice. The thread is removed from the scheduling list. Execution terminates regardless of the execution status of the thread function. Automatic variables may or may not be destroyed. This function must be considered as a last resort since memory may be lost. */ void terminate(); /** Causes the current thread to relinquish the processor. The scheduler selects a thread ready to run and transfers control to that thread. The actual effect of #yield# heavily depends on the selected implementation. Function #yield# usually returns zero when the execution of the current thread is resumed. It may return a positive number when it can determine that the current thread will remain the only runnable thread for some time. You may then call function \Ref{get_select} to obtain more information. */ static int yield(); /** Returns a value which uniquely identifies the current thread. */ static void *current(); #if THREADMODEL==WINTHREADS private: HANDLE hthr; DWORD thrid; #elif THREADMODEL==MACTHREADS private: unsigned long thid; static pascal void *start(void *arg); #elif THREADMODEL==POSIXTHREADS private: pthread_t hthr; static void *start(void *arg); #elif THREADMODEL==JRITHREADS private: JRIGlobalRef obj; #elif THREADMODEL==COTHREADS friend class GMonitor; public: class cotask; class cotask *task; /** Replaces system call #select# (COTHREADS only). The #COTHREADS# model does not redefine system function. System functions therefore can potentially block the whole process (instead of blocking the current thread only) because the system is not aware of the #COTHREADS# scheduler. The function #GThread::select# is a #COTHREADS#-aware replacement for the well known system function #select#. You can also use #GThread::select# for making sure that calls to system functions will not block the entire process, as demonstrated below: \begin{verbatim} int gthread_read(int fd, void *buffer, size_t len) { fd_set rdset; FD_ZERO(&rdset); FD_SET(fd, &rdset); GThread::select(fd+1, &rdset, 0, 0, 0); return read(fd, buffer, len); } \end{verbatim} */ static int select(int nfds, fd_set*, fd_set*, fd_set*, struct timeval*); /** Provide arguments for system call #select# (COTHREADS only). It may be appropriate to call the real system call #select# if the current thread is the only thread ready to run. Other threads however may wake up when certain file descriptors are ready or when a certain delay expires. Function #get_select# returns this information by filling the three usual file descriptor sets (similar to the arguments of system call #select#). It also returns a timeout #timeout# expressed in milliseconds. Note that this timeout is zero is the current thread is not the sole thread ready to run. */ static void get_select(int &nfds, fd_set*, fd_set*, fd_set*, unsigned long &timeout); /** Install hooks in the scheduler (COTHREADS only). The hook function #call# is called when a new thread is created (argument is #GThread::CallbackCreate#), when a thread terminates (argument is #GThread::CallbackTerminate#), or when thread is unblocked (argument is #GThread::CallbackUnblock#). This callback can be useful in certain GUI toolkits where the most convenient method for scheduling the threads consists in setting a timer event that calls \Ref{GThread::yield}. */ static void set_scheduling_callback(void (*call)(int)); enum { CallbackCreate, CallbackTerminate, CallbackUnblock }; #endif public: // Should be considered as private void (*xentry)(void*); void *xarg; private: // Disable default members GThread(const GThread&); GThread& operator=(const GThread&); }; /** Monitor class. Monitors have been first described in (C.A.R Hoare, Communications of the ACM, 17(10), 1974). This mechanism provides the basic mutual exclusion (mutex) and thread notification facilities (condition variables). Only one thread can own the monitor at a given time. Functions \Ref{enter} and \Ref{leave} can be used to acquire and release the monitor. This mutual exclusion provides an efficient way to protect segment of codes ({\em critical sections}) which should not be simultaneously executed by two threads. Class \Ref{GMonitorLock} provides a convenient way to do this effectively. When the thread owning the monitor calls function \Ref{wait}, the monitor is released and the thread starts waiting until another thread calls function \Ref{signal} or \Ref{broadcast}. When the thread wakes-up, it re-acquires the monitor and function #wait# returns. Since the signaling thread must acquire the monitor before calling functions #signal# and #broadcast#, the signaled thread will not be able to re-acquire the monitor until the signaling thread(s) releases the monitor. {\bf Note} --- Both the copy constructor and the copy operator are declared as private members. It is therefore not possible to make multiple copies of instances of this class, as implied by the class semantic. */ class GMonitor { public: GMonitor(); ~GMonitor(); /** Enters the monitor. If the monitor is acquired by another thread this function waits until the monitor is released. The current thread then acquires the monitor. Calls to #enter# and #leave# may be nested. */ void enter(); /** Leaves the monitor. The monitor counts how many times the current thread has entered the monitor. Function #leave# decrement this count. The monitor is released when this count reaches zero. An exception is thrown if this function is called by a thread which does not own the monitor. */ void leave(); /** Waits until the monitor is signaled. The current thread atomically releases the monitor and waits until another thread calls function #signal# or #broadcast#. Function #wait# then re-acquires the monitor and returns. An exception is thrown if this function is called by a thread which does not own the monitor. */ void wait(); /** Waits until the monitor is signaled or a timeout is reached. The current thread atomically releases the monitor and waits until another thread calls function #signal# or #broadcast# or a maximum of #timeout# milliseconds. Function #wait# then re-acquires the monitor and returns. An exception is thrown if this function is called by a thread which does not own the monitor. */ void wait(unsigned long timeout); /** Signals one waiting thread. Function #signal# wakes up at most one of the waiting threads for this monitor. An exception is thrown if this function is called by a thread which does not own the monitor. */ void signal(); /** Signals all waiting threads. Function #broadcast# wakes up all the waiting threads for this monitor. An exception is thrown if this function is called by a thread which does not own the monitor. */ void broadcast(); private: #if THREADMODEL==WINTHREADS int ok; int count; DWORD locker; CRITICAL_SECTION cs; struct thr_waiting *head; struct thr_waiting *tail; #elif THREADMODEL==MACTHREADS int ok; int count; unsigned long locker; int wlock; int wsig; #elif THREADMODEL==POSIXTHREADS int ok; int count; pthread_t locker; pthread_mutex_t mutex; pthread_cond_t cond; #elif THREADMODEL==COTHREADS int ok; int count; void *locker; int wlock; int wsig; #elif THREADMODEL==JRITHREADS JRIGlobalRef obj; #endif private: // Disable default members GMonitor(const GMonitor&); GMonitor& operator=(const GMonitor&); }; // ---------------------------------------- // NOTHREADS INLINES #if THREADMODEL==NOTHREADS inline GThread::GThread(int) {} inline GThread::~GThread(void) {} inline void GThread::terminate() {} inline int GThread::yield() { return 0; } inline void* GThread::current() { return 0; } inline GMonitor::GMonitor() {} inline GMonitor::~GMonitor() {} inline void GMonitor::enter() {} inline void GMonitor::leave() {} inline void GMonitor::wait() {} inline void GMonitor::wait(unsigned long) {} inline void GMonitor::signal() {} inline void GMonitor::broadcast() {} #endif // NOTHREADS // ---------------------------------------- // SCOPE LOCK /** Wrapper for mutually exclusive code. This class locks a specified critical section (see \Ref{GCriticalSection}) at construction time and unlocks it at destruction time. It provides a convenient way to take advantage of the C++ implicit destruction of automatic variables in order to make sure that the monitor is released when exiting the protected code. The following code will release the monitor when the execution thread leaves the protected scope, either because the protected code has executed successfully, or because an exception was thrown. \begin{verbatim} { -- protected scope static GMonitor theMonitor; GMonitorLock lock(&theMonitor) ... -- protected code } \end{verbatim} This construct will do nothing when passed a null pointer. */ class GMonitorLock { private: GMonitor *gsec; public: /** Constructor. Enters the monitor #gsec#. */ GMonitorLock(GMonitor *gsec) : gsec(gsec) { if (gsec) gsec->enter(); }; /** Destructor. Leaves the associated monitor. */ ~GMonitorLock() { if (gsec) gsec->leave(); }; }; // ---------------------------------------- // GSAFEFLAGS (not so safe) /** A thread safe class representing a set of flags. The flags are protected by \Ref{GMonitor}, which is attempted to be locked whenever somebody accesses the flags. One can modify the class contents using one of two functions: \Ref{test_and_modify}() and \Ref{wait_and_modify}(). Both of them provide atomic operation of testing (first) and modification (second). The flags remain locked between the moment of testing and modification, which guarantees, that their state cannot be changed in between of these operations. */ class GSafeFlags : public GMonitor { private: volatile long flags; public: /// Constructs #GSafeFlags# object. GSafeFlags(long flags=0); /** Assignment operator. Will also wake up threads waiting for the flags to change. */ GSafeFlags & operator=(long flags); /** Returns the value of the flags */ operator long(void) const; /** Modifies the flags by ORing them with the provided mask. A broadcast will be sent after the modification is done. */ GSafeFlags & operator|=(long mask); /** Modifies the flags by ANDing them with the provided mask. A broadcast will be sent after the modification is done. */ GSafeFlags & operator&=(long mask); /** If all bits mentioned in #set_mask# are set in the flags and all bits mentioned in #clr_mask# are cleared in the flags, it sets all bits from #set_mask1# in the flags, clears all flags from #clr_mask1# in the flags and returns #TRUE#. Otherwise returns #FALSE#. */ bool test_and_modify(long set_mask, long clr_mask, long set_mask1, long clr_mask1); /** Waits until all bits mentioned in #set_mask# are set in the flags and all bits mentioned in #clr_flags# are cleared in the flags. After that it sets bits from #set_mask1# and clears bits from #clr_mask1# in the flags. */ void wait_and_modify(long set_mask, long clr_mask, long set_mask1, long clr_mask1); /** Waits until all bits set in #set_mask# are set in the flags and all bits mentioned in #clr_mask# are cleared in the flags. */ void wait_for_flags(long set_mask, long clr_mask=0) const; /** Modifies the flags by setting all bits mentioned in #set_mask# and clearing all bits mentioned in #clr_mask#. If the flags have actually been modified, a broadcast will be sent. */ void modify(long set_mask, long clr_mask); }; inline GSafeFlags::GSafeFlags(long xflags) : flags(xflags) { } inline void GSafeFlags::wait_for_flags(long set_mask, long clr_mask) const { ((GSafeFlags *) this)->wait_and_modify(set_mask, clr_mask, 0, 0); } inline void GSafeFlags::modify(long set_mask, long clr_mask) { test_and_modify(0, 0, set_mask, clr_mask); } inline GSafeFlags & GSafeFlags::operator|=(long mask) { test_and_modify(0, 0, mask, 0); return *this; } inline GSafeFlags & GSafeFlags::operator&=(long mask) { test_and_modify(0, 0, 0, ~mask); return *this; } //@} // ---------------------------------------- // COMPATIBILITY CLASSES // -- these classes are no longer documented. class GCriticalSection : protected GMonitor { public: void lock() { GMonitor::enter(); }; void unlock() { GMonitor::leave(); }; }; class GEvent : protected GMonitor { private: int status; public: GEvent() : status(0) { }; void set() { if (!status) { enter(); status=1; signal(); leave(); } }; void wait() { enter(); if (!status) GMonitor::wait(); status=0; leave(); }; void wait(int timeout) { enter(); if (!status) GMonitor::wait(timeout); status=0; leave(); }; }; class GCriticalSectionLock { private: GCriticalSection *gsec; public: GCriticalSectionLock(GCriticalSection *gsec) : gsec(gsec) { if (gsec) gsec->lock(); }; ~GCriticalSectionLock() { if (gsec) gsec->unlock(); }; }; // ---------------------------------------- #ifdef HAVE_NAMESPACES } # ifndef NOT_USING_DJVU_NAMESPACE using namespace DJVU; # endif #endif #endif //_GTHREADS_H_