CURRENT CVS HEAD Version RELEASE 2.9.0 pending ------------- (2011-??-??) General ------- New bug fixes in this release since 2.8.0 have NOT been applied to the 1.x.x series. Version 2.8.0 may be the last release for some older Windows systems. Some changes post 2011-02-26 in CVS may not be compatible with pre Windows 2000 systems. Version 1 no longer maintained ------------------------------ The 1.x.x series is no longer maintained. However, if you really need a version 1, the differences between 1.11.0 and 2.7.0 are very small, mainly revolving around the pthread_once_t_ struct. Those differences applied as a patch to the current 2.x.x should work. Don't forget to change the version numbering in pthread.h before building. If you distribute such a version 1.x.x please bear in mind that your numbers may clash with those of others doing the same thing. Please consider also making identifying changes in version.rc to differentiate your build. Testing and verification ------------------------ The current CVS head version has been tested on an SMP architecture (AMD Phenom 9750 Quad Core) by running the MinGW32 (GCC) builds against the full test suite, stress tests and benchmarks. New Features ------------ (MSC and GNU builds) The statically linked library now automatically initialises and cleans up on program start/exit, i.e. statically linked applications need not call the routines pthread_win32_process_attach_np() and pthread_win32_process_detach_np() explicitly. The per-thread routine pthread_win32_thread_detach_np() is also called at program exit to cleanup POSIX resources acquired by the primary Windows native thread (if I (RJ) understand the process correctly). Other Windows native threads that call POSIX API routines may need to call the thread detach routine on thread exit if the application depends on reclaimed POSIX resources or running POSIX TSD (TLS) destructors. See README.NONPORTABLE for descriptions of these routines. - Ramiro Polla Robust mutexes are implemented within the PROCESS_PRIVATE scope. Robust mutexes are mandatory (SUSv3 2008) for implementations that support PSHARED objects and discretionary for those that don't. We still don't have PSHARED support but wanted to be source code compatible and they will hopefully nevertheless be useful for some applications. NOTE that pthread_mutex_*() functions may return different values for robust mutexes than they otherwise do in normal usage, e.g. pthread_mutex_unlock() is required to check ownership for all mutex types when the mutex is robust, whereas this does not occur for the "normal" non-robust mutex type. - Ross Johnson pthread_getsequence_np is implemented for source level compatibility with some other implementations. This routine returns a 64 bit sequence number that is uniquely associated with a thread. It can be used by applications to order or hash POSIX thread handles. - Ross Johnson Bug fixes --------- Many more changes for 64 bit systems. - Kai Tietz Various modifications and fixes to build and test for WinCE. - Marcel Ruff, Sinan Kaya Fix pthread_cond_destroy() - should not be a cancellation point. Other minor build problems fixed. - Romano Paolo Tenca Remove potential deadlock condition from pthread_cond_destroy(). - Eric Berge Various modifications to build and test for Win64. - Kip Streithorst Various fixes to the QueueUserAPCEx async cancellation helper DLL (this is a separate download) and pthreads code cleanups. - Sebastian Gottschalk Removed potential NULL pointer reference. - Robert Kindred Removed the requirement that applications restrict the number of threads calling pthread_barrier_wait() [on the same barrier] to just the barrier count in order to avoid contention and dead lock. Also reduced the contention between barrier_wait and barrier_destroy. This change will have slowed barriers down slightly but halves the number of semaphores consumed per barrier to one. - Ross Johnson Fixed a handle leak in sched_[gs]etscheduler. - Mark Pizzolato Removed all of the POSIX re-entrant function compatibility macros from pthread.h. Some were simply not semanticly correct. - Igor Lubashev Threads no longer attempt to pass uncaught exceptions out of thread scope (C++ and SEH builds only). Uncaught exceptions now cause the thread to exit with the return code PTHREAD_CANCELED. - Ross Johnson Other changes ------------- Dependence on the winsock library is now discretionary via #define RETAIN_WSALASTERROR in config.h. It is undefined by default unless WINCE is defined (because RJ is unsure of the dependency there). - Ramiro Polla Several static POSIX mutexes used for internal management were replaced by MCS queue-based locks to reduce resource overheads, particularly Win32 objects. - Ross Johnson New tests --------- robust[1-5].c - Robust mutexes sequence1.c - per-thread unique sequence numbers Modified tests and benchtests ----------------------------- All mutex*.c tests wherever appropriate have been modified to also test robust mutexes under the same conditions. Added robust mutex benchtests to benchtest*.c wherever appropriate. RELEASE 2.8.0 ------------- (2006-12-22) General ------- New bug fixes in this release since 2.7.0 have not been applied to the version 1.x.x series. It is probably time to drop version 1. Testing and verification ------------------------ This release has not yet been tested on SMP architechtures. All tests pass on a uni-processor system. Bug fixes --------- Sem_destroy could return EBUSY even though no threads were waiting on the semaphore. Other races around invalidating semaphore structs (internally) have been removed as well. New tests --------- semaphore5.c - tests the bug fix referred to above. RELEASE 2.7.0 ------------- (2005-06-04) General ------- All new features in this release have been back-ported in release 1.11.0, including the incorporation of MCS locks in pthread_once, however, versions 1 and 2 remain incompatible even though they are now identical in performance and functionality. Testing and verification ------------------------ This release has been tested (passed the test suite) on both uni-processor and multi-processor systems. - Tim Theisen Bug fixes --------- Pthread_once has been re-implemented to remove priority boosting and other complexity to improve robustness. Races for Win32 handles that are not recycle-unique have been removed. The general form of pthread_once is now the same as that suggested earlier by Alexander Terekhov, but instead of the 'named mutex', a queue-based lock has been implemented which has the required properties of dynamic self initialisation and destruction. This lock is also efficient. The ABI is unaffected in as much as the size of pthread_once_t has not changed and PTHREAD_ONCE_INIT has not changed, however, applications that peek inside pthread_once_t, which is supposed to be opaque, will break. - Vladimir Kliatchko New features ------------ * Support for Mingw cross development tools added to GNUmakefile. Mingw cross tools allow building the libraries on Linux. - Mikael Magnusson RELEASE 2.6.0 ------------- (2005-05-19) General ------- All of the bug fixes and new features in this release have been back-ported in release 1.10.0. Testing and verification ------------------------ This release has been tested (passed the test suite) on both uni-processor and multi-processor systems. Thanks to Tim Theisen at TomoTherapy for exhaustively running the MP tests and for providing crutial observations and data when faults are detected. Bugs fixed ---------- * pthread_detach() now reclaims remaining thread resources if called after the target thread has terminated. Previously, this routine did nothing in this case. New tests --------- * detach1.c - tests that pthread_detach properly invalidates the target thread, which indicates that the thread resources have been reclaimed. RELEASE 2.5.0 ------------- (2005-05-09) General ------- The package now includes a reference documentation set consisting of HTML formatted Unix-style manual pages that have been edited for consistency with Pthreads-w32. The set can also be read online at: http://sources.redhat.com/pthreads-win32/manual/index.html Thanks again to Tim Theisen for running the test suite pre-release on an MP system. All of the bug fixes and new features in this release have been back-ported in release 1.9.0. Bugs fixed ---------- * Thread Specific Data (TSD) key management has been ammended to eliminate a source of (what was effectively) resource leakage (a HANDLE plus memory for each key destruct routine/thread association). This was not a true leak because these resources were eventually reclaimed when pthread_key_delete was run AND each thread referencing the key had exited. The problem was that these two conditions are often not met until very late, and often not until the process is about to exit. The ammended implementation avoids the need for the problematic HANDLE and reclaims the memory as soon as either the key is deleted OR the thread exits, whichever is first. Thanks to Richard Hughes at Aculab for identifying and locating the leak. * TSD key destructors are now processed up to PTHREAD_DESTRUCTOR_ITERATIONS times instead of just once. PTHREAD_DESTRUCTOR_ITERATIONS has been defined in pthread.h for some time but not used. * Fix a semaphore accounting race between sem_post/sem_post_multiple and sem_wait cancellation. This is the same issue as with sem_timedwait that was fixed in the last release. * sem_init, sem_post, and sem_post_multiple now check that the semaphore count never exceeds _POSIX_SEM_VALUE_MAX. * Although sigwait() is nothing more than a no-op, it should at least be a cancellation point to be consistent with the standard. New tests --------- * stress1.c - attempts to expose problems in condition variable and semaphore timed wait logic. This test was inspired by Stephan Mueller's sample test code used to identify the sem_timedwait bug from the last release. It's not a part of the regular test suite because it can take awhile to run. To run it: nmake clean VC-stress * tsd2.c - tests that key destructors are re-run if the tsd key value is not NULL after the destructor routine has run. Also tests that pthread_setspecific() and pthread_getspecific() are callable from destructors. RELEASE 2.4.0 ------------- (2005-04-26) General ------- There is now no plan to release a version 3.0.0 to fix problems in pthread_once(). Other possible implementations of pthread_once will still be investigated for a possible future release in an attempt to reduce the current implementation's complexity. All of the bug fixes and new features in this release have been back-ported for release 1.8.0. Bugs fixed ---------- * Fixed pthread_once race (failures on an MP system). Thanks to Tim Theisen for running exhaustive pre-release testing on his MP system using a range of compilers: VC++ 6 VC++ 7.1 Intel C++ version 8.0 All tests passed. Some minor speed improvements were also done. * Fix integer overrun error in pthread_mutex_timedlock() - missed when sem_timedwait() was fixed in release 2.2.0. This routine no longer returns ENOTSUP when NEED_SEM is defined - it is supported (NEED_SEM is only required for WinCE versions prior to 3.0). * Fix timeout bug in sem_timedwait(). - Thanks to Stephan Mueller for reporting, providing diagnostic output and test code. * Fix several problems in the NEED_SEM conditionally included code. NEED_SEM included code is provided for systems that don't implement W32 semaphores, such as WinCE prior to version 3.0. An alternate implementation of POSIX semaphores is built using W32 events for these systems when NEED_SEM is defined. This code has been completely rewritten in this release to reuse most of the default POSIX semaphore code, and particularly, to implement all of the sem_* routines supported by pthreads-win32. Tim Theisen also run the test suite over the NEED_SEM code on his MP system. All tests passed. * The library now builds without errors for the Borland Builder 5.5 compiler. New features ------------ * pthread_mutex_timedlock() and all sem_* routines provided by pthreads-win32 are now implemented for WinCE versions prior to 3.0. Those versions did not implement W32 semaphores. Define NEED_SEM in config.h when building the library for these systems. Known issues in this release ---------------------------- * pthread_once is too complicated - but it works as far as testing can determine.. * The Borland version of the dll fails some of the tests with a memory read exception. The cause is not yet known but a compiler bug has not been ruled out. RELEASE 2.3.0 ------------- (2005-04-12) General ------- Release 1.7.0 is a backport of features and bug fixes new in this release. See earlier notes under Release 2.0.0/General. Bugs fixed ---------- * Fixed pthread_once potential for post once_routine cancellation hanging due to starvation. See comments in pthread_once.c. Momentary priority boosting is used to ensure that, after a once_routine is cancelled, the thread that will run the once_routine is not starved by higher priority waiting threads at critical times. Priority boosting occurs only AFTER a once_routine cancellation, and is applied only to that once_control. The once_routine is run at the thread's normal base priority. New tests --------- * once4.c: Aggressively tests pthread_once() under realtime conditions using threads with varying priorities. Windows' random priority boosting does not occur for threads with realtime priority levels. RELEASE 2.2.0 ------------- (2005-04-04) General ------- * Added makefile targets to build static link versions of the library. Both MinGW and MSVC. Please note that this does not imply any change to the LGPL licensing, which still imposes psecific conditions on distributing software that has been statically linked with this library. * There is a known bug in pthread_once(). Cancellation of the init_routine exposes a potential starvation (i.e. deadlock) problem if a waiting thread has a higher priority than the initting thread. This problem will be fixed in version 3.0.0 of the library. Bugs fixed ---------- * Fix integer overrun error in sem_timedwait(). Kevin Lussier * Fix preprocessor directives for static linking. Dimitar Panayotov RELEASE 2.1.0 ------------- (2005-03-16) Bugs fixed ---------- * Reverse change to pthread_setcancelstate() in 2.0.0. RELEASE 2.0.0 ------------- (2005-03-16) General ------- This release represents an ABI change and the DLL version naming has incremented from 1 to 2, e.g. pthreadVC2.dll. Version 1.4.0 back-ports the new functionality included in this release. Please distribute DLLs built from that version with updates to applications built on pthreads-win32 version 1.x.x. The package naming has changed, replacing the snapshot date with the version number + descriptive information. E.g. this release is "pthreads-w32-2-0-0-release". Bugs fixed ---------- * pthread_setcancelstate() no longer checks for a pending async cancel event if the library is using alertable async cancel. See the README file (Prerequisites section) for info on adding alertable async cancelation. New features ------------ * pthread_once() now supports init_routine cancellability. New tests --------- * Agressively test pthread_once() init_routine cancellability. SNAPSHOT 2005-03-08 ------------------- Version 1.3.0 Bug reports (fixed) ------------------- * Implicitly created threads leave Win32 handles behind after exiting. - Dmitrii Semii * pthread_once() starvation problem. - Gottlob Frege New tests --------- * More intense testing of pthread_once(). SNAPSHOT 2005-01-25 ------------------- Version 1.2.0 Bug fixes --------- * Attempted acquisition of a recursive mutex could cause waiting threads to not be woken when the mutex was released. - Ralf Kubis * Various package omissions have been fixed. SNAPSHOT 2005-01-03 ------------------- Version 1.1.0 Bug fixes --------- * Unlocking recursive or errorcheck mutexes would sometimes unexpectedly return an EPERM error (bug introduced in snapshot-2004-11-03). - Konstantin Voronkov SNAPSHOT 2004-11-22 ------------------- Version 1.0.0 This snapshot primarily fixes the condvar bug introduced in snapshot-2004-11-03. DLL versioning has also been included to allow applications to runtime check the Microsoft compatible DLL version information, and to extend the DLL naming system for ABI and major (non-backward compatible) API changes. See the README file for details. Bug fixes --------- * Condition variables no longer deadlock (bug introduced in snapshot-2004-11-03). - Alexander Kotliarov and Nicolas at saintmac * DLL naming extended to avoid 'DLL hell' in the future, and to accommodate the ABI change introduced in snapshot-2004-11-03. Snapshot 2004-11-03 will be removed from FTP sites. New features ------------ * A Microsoft-style version resource has been added to the DLL for applications that wish to check DLL compatibility at runtime. * Pthreads-win32 DLL naming has been extended to allow incompatible DLL versions to co-exist in the same filesystem. See the README file for details, but briefly: while the version information inside the DLL will change with each release from now on, the DLL version names will only change if the new DLL is not backward compatible with older applications. The versioning scheme has been borrowed from GNU Libtool, and the DLL naming scheme is from Cygwin. Provided the Libtool-style numbering rules are honoured, the Cygwin DLL naming scheme automatcally ensures that DLL name changes are minimal and that applications will not load an incompatible pthreads-win32 DLL. Those who use the pre-built DLLs will find that the DLL/LIB names have a new suffix (1) in this snapshot. E.g. pthreadVC1.dll etc. * The POSIX thread ID reuse uniqueness feature introduced in the last snapshot has been kept as default, but the behaviour can now be controlled when the DLL is built to effectively switch it off. This makes the library much more sensitive to applications that assume that POSIX thread IDs are unique, i.e. are not strictly compliant with POSIX. See the PTW32_THREAD_ID_REUSE_INCREMENT macro comments in config.h for details. Other changes ------------- Certain POSIX macros have changed. These changes are intended to conform to the Single Unix Specification version 3, which states that, if set to 0 (zero) or not defined, then applications may use sysconf() to determine their values at runtime. Pthreads-win32 does not implement sysconf(). The following macros are no longer undefined, but defined and set to -1 (not implemented): _POSIX_THREAD_ATTR_STACKADDR _POSIX_THREAD_PRIO_INHERIT _POSIX_THREAD_PRIO_PROTECT _POSIX_THREAD_PROCESS_SHARED The following macros are defined and set to 200112L (implemented): _POSIX_THREADS _POSIX_THREAD_SAFE_FUNCTIONS _POSIX_THREAD_ATTR_STACKSIZE _POSIX_THREAD_PRIORITY_SCHEDULING _POSIX_SEMAPHORES _POSIX_READER_WRITER_LOCKS _POSIX_SPIN_LOCKS _POSIX_BARRIERS The following macros are defined and set to appropriate values: _POSIX_THREAD_THREADS_MAX _POSIX_SEM_VALUE_MAX _POSIX_SEM_NSEMS_MAX PTHREAD_DESTRUCTOR_ITERATIONS PTHREAD_KEYS_MAX PTHREAD_STACK_MIN PTHREAD_THREADS_MAX SNAPSHOT 2004-11-03 ------------------- DLLs produced from this snapshot cannot be used with older applications without recompiling the application, due to a change to pthread_t to provide unique POSIX thread IDs. Although this snapshot passes the extended test suite, many of the changes are fairly major, and some applications may show different behaviour than previously, so adopt with care. Hopefully, any changed behaviour will be due to the library being better at it's job, not worse. Bug fixes --------- * pthread_create() no longer accepts NULL as the thread reference arg. A segfault (memory access fault) will result, and no thread will be created. * pthread_barrier_wait() no longer acts as a cancelation point. * Fix potential race condition in pthread_once() - Tristan Savatier * Changes to pthread_cond_destroy() exposed some coding weaknesses in several test suite mini-apps because pthread_cond_destroy() now returns EBUSY if the CV is still in use. New features ------------ * Added for compatibility: PTHREAD_RECURSIVE_MUTEX_INITIALIZER, PTHREAD_ERRORCHECK_MUTEX_INITIALIZER, PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP, PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP * Initial support for Digital Mars compiler - Anuj Goyal * Faster Mutexes. These have been been rewritten following a model provided by Alexander Terekhov that reduces kernel space checks, and eliminates some additional critical sections used to manage a race between timedlock expiration and unlock. Please be aware that the new mutexes do not enforce strict absolute FIFO scheduling of mutexes, however any out-of-order lock acquisition should be very rare. * Faster semaphores. Following a similar model to mutexes above, these have been rewritten to use preliminary users space checks. * sem_getvalue() now returns the number of waiters. * The POSIX thread ID now has much stronger uniqueness characteristics. The library garrantees not to reuse the same thread ID for at least 2^(wordsize) thread destruction/creation cycles. New tests --------- * semaphore4.c: Tests cancelation of the new sem_wait(). * semaphore4t.c: Likewise for sem_timedwait(). * rwlock8.c: Tests and times the slow execution paths of r/w locks, and the CVs, mutexes, and semaphores that they're built on. SNAPSHOT 2004-05-16 ------------------- Attempt to add Watcom to the list of compilers that can build the library. This failed in the end due to it's non-thread-aware errno. The library builds but the test suite fails. See README.Watcom for more details. Bug fixes --------- * Bug and memory leak in sem_init() - Alex Blanco * ptw32_getprocessors() now returns CPU count of 1 for WinCE. - James Ewing * pthread_cond_wait() could be canceled at a point where it should not be cancelable. Fixed. - Alexander Terekhov * sem_timedwait() had an incorrect timeout calculation. - Philippe Di Cristo * Fix a memory leak left behind after threads are destroyed. - P. van Bruggen New features ------------ * Ported to AMD64. - Makoto Kato * True pre-emptive asynchronous cancelation of threads. This is optional and requires that Panagiotis E. Hadjidoukas's QueueUserAPCEx package be installed. This package is included in the pthreads-win32 self-unpacking Zip archive starting from this snapshot. See the README.txt file inside the package for installation details. Note: If you don't use async cancelation in your application, or don't need to cancel threads that are blocked on system resources such as network I/O, then the default non-preemptive async cancelation is probably good enough. However, pthreads-win32 auto-detects the availability of these components at run-time, so you don't need to rebuild the library from source if you change your mind later. All of the advice available in books and elsewhere on the undesirability of using async cancelation in any application still stands, but this feature is a welcome addition with respect to the library's conformance to the POSIX standard. SNAPSHOT 2003-09-18 ------------------- Cleanup of thread priority management. In particular, setting of thread priority now attempts to map invalid Win32 values within the range returned by sched_get_priority_min/max() to useful values. See README.NONPORTABLE under "Thread priority". Bug fixes --------- * pthread_getschedparam() now returns the priority given by the most recent call to pthread_setschedparam() or established by pthread_create(), as required by the standard. Previously, pthread_getschedparam() incorrectly returned the running thread priority at the time of the call, which may have been adjusted or temporarily promoted/demoted. * sched_get_priority_min() and sched_get_priority_max() now return -1 on error and set errno. Previously, they incorrectly returned the error value directly. SNAPSHOT 2003-09-04 ------------------- Bug fixes --------- * ptw32_cancelableWait() now allows cancelation of waiting implicit POSIX threads. New test -------- * cancel8.c tests cancelation of Win32 threads waiting at a POSIX cancelation point. SNAPSHOT 2003-09-03 ------------------- Bug fixes --------- * pthread_self() would free the newly created implicit POSIX thread handle if DuplicateHandle failed instead of recycle it (very unlikely). * pthread_exit() was neither freeing nor recycling the POSIX thread struct for implicit POSIX threads. New feature - Cancelation of/by Win32 (non-POSIX) threads --------------------------------------------------------- Since John Bossom's original implementation, the library has allowed non-POSIX initialised threads (Win32 threads) to call pthreads-win32 routines and therefore interact with POSIX threads. This is done by creating an on-the-fly POSIX thread ID for the Win32 thread that, once created, allows fully reciprical interaction. This did not extend to thread cancelation (async or deferred). Now it does. Any thread can be canceled by any other thread (Win32 or POSIX) if the former thread's POSIX pthread_t value is known. It's TSD destructors and POSIX cleanup handlers will be run before the thread exits with an exit code of PTHREAD_CANCELED (retrieved with GetExitCodeThread()). This allows a Win32 thread to, for example, call POSIX CV routines in the same way that POSIX threads would/should, with pthread_cond_wait() cancelability and cleanup handlers (pthread_cond_wait() is a POSIX cancelation point). By adding cancelation, Win32 threads should now be able to call all POSIX threads routines that make sense including semaphores, mutexes, condition variables, read/write locks, barriers, spinlocks, tsd, cleanup push/pop, cancelation, pthread_exit, scheduling, etc. Note that these on-the-fly 'implicit' POSIX thread IDs are initialised as detached (not joinable) with deferred cancelation type. The POSIX thread ID will be created automatically by any POSIX routines that need a POSIX handle (unless the routine needs a pthread_t as a parameter of course). A Win32 thread can discover it's own POSIX thread ID by calling pthread_self(), which will create the handle if necessary and return the pthread_t value. New tests --------- Test the above new feature. SNAPSHOT 2003-08-19 ------------------- This snapshot fixes some accidental corruption to new test case sources. There are no changes to the library source code. SNAPSHOT 2003-08-15 ------------------- Bug fixes --------- * pthread.dsp now uses correct compile flags (/MD). - Viv * pthread_win32_process_detach_np() fixed memory leak. - Steven Reddie * pthread_mutex_destroy() fixed incorrect return code. - Nicolas Barry * pthread_spin_destroy() fixed memory leak. - Piet van Bruggen * Various changes to tighten arg checking, and to work with later versions of MinGW32 and MsysDTK. * pthread_getschedparam() etc, fixed dangerous thread validity checking. - Nicolas Barry * POSIX thread handles are now reused and their memory is not freed on thread exit. This allows for stronger thread validity checking. New standard routine -------------------- * pthread_kill() added to provide thread validity checking to applications. It does not accept any non zero values for the signal arg. New test cases -------------- * New test cases to confirm validity checking, pthread_kill(), and thread reuse. SNAPSHOT 2003-05-10 ------------------- Bug fixes --------- * pthread_mutex_trylock() now returns correct error values. pthread_mutex_destroy() will no longer destroy a recursively locked mutex. pthread_mutex_lock() is no longer inadvertantly behaving as a cancelation point. - Thomas Pfaff * pthread_mutex_timedlock() no longer occasionally sets incorrect mutex ownership, causing deadlocks in some applications. - Robert Strycek and Alexander Terekhov SNAPSHOT 2002-11-04 ------------------- Bug fixes --------- * sem_getvalue() now returns the correct value under Win NT and WinCE. - Rob Fanner * sem_timedwait() now uses tighter checks for unreasonable abstime values - that would result in unexpected timeout values. * ptw32_cond_wait_cleanup() no longer mysteriously consumes CV signals but may produce more spurious wakeups. It is believed that the sem_timedwait() call is consuming a CV signal that it shouldn't. - Alexander Terekhov * Fixed a memory leak in ptw32_threadDestroy() for implicit threads. * Fixed potential for deadlock in pthread_cond_destroy(). A deadlock could occur for statically declared CVs (PTHREAD_COND_INITIALIZER), when one thread is attempting to destroy the condition variable while another is attempting to dynamically initialize it. - Michael Johnson SNAPSHOT 2002-03-02 ------------------- Cleanup code default style. (IMPORTANT) ---------------------------------------------------------------------- Previously, if not defined, the cleanup style was determined automatically from the compiler/language, and one of the following was defined accordingly: __CLEANUP_SEH MSVC only __CLEANUP_CXX C++, including MSVC++, GNU G++ __CLEANUP_C C, including GNU GCC, not MSVC These defines determine the style of cleanup (see pthread.h) and, most importantly, the way that cancelation and thread exit (via pthread_exit) is performed (see the routine ptw32_throw() in private.c). In short, the exceptions versions of the library throw an exception when a thread is canceled or exits (via pthread_exit()), which is caught by a handler in the thread startup routine, so that the the correct stack unwinding occurs regardless of where the thread is when it's canceled or exits via pthread_exit(). In this and future snapshots, unless the build explicitly defines (e.g. via a compiler option) __CLEANUP_SEH, __CLEANUP_CXX, or __CLEANUP_C, then the build NOW always defaults to __CLEANUP_C style cleanup. This style uses setjmp/longjmp in the cancelation and pthread_exit implementations, and therefore won't do stack unwinding even when linked to applications that have it (e.g. C++ apps). This is for consistency with most current commercial Unix POSIX threads implementations. Compaq's TRU64 may be an exception (no pun intended) and possible future trend. Although it was not clearly documented before, it is still necessary to build your application using the same __CLEANUP_* define as was used for the version of the library that you link with, so that the correct parts of pthread.h are included. That is, the possible defines require the following library versions: __CLEANUP_SEH pthreadVSE.dll __CLEANUP_CXX pthreadVCE.dll or pthreadGCE.dll __CLEANUP_C pthreadVC.dll or pthreadGC.dll E.g. regardless of whether your app is C or C++, if you link with pthreadVC.lib or libpthreadGC.a, then you must define __CLEANUP_C. THE POINT OF ALL THIS IS: if you have not been defining one of these explicitly, then the defaults as described at the top of this section were being used. THIS NOW CHANGES, as has been explained above, but to try to make this clearer here's an example: If you were building your application with MSVC++ i.e. using C++ exceptions and not explicitly defining one of __CLEANUP_*, then __CLEANUP_C++ was automatically defined for you in pthread.h. You should have been linking with pthreadVCE.dll, which does stack unwinding. If you now build your application as you had before, pthread.h will now automatically set __CLEANUP_C as the default style, and you will need to link with pthreadVC.dll. Stack unwinding will now NOT occur when a thread is canceled, or the thread calls pthread_exit(). Your application will now most likely behave differently to previous versions, and in non-obvious ways. Most likely is that locally instantiated objects may not be destroyed or cleaned up after a thread is canceled. If you want the same behaviour as before, then you must now define __CLEANUP_C++ explicitly using a compiler option and link with pthreadVCE.dll as you did before. WHY ARE WE MAKING THE DEFAULT STYLE LESS EXCEPTION-FRIENDLY? Because no commercial Unix POSIX threads implementation allows you to choose to have stack unwinding. Therefore, providing it in pthread-win32 as a default is dangerous. We still provide the choice but unless you consciously choose to do otherwise, your pthreads applications will now run or crash in similar ways irrespective of the threads platform you use. Or at least this is the hope. WHY NOT REMOVE THE EXCEPTIONS VERSIONS OF THE LIBRARY ALTOGETHER? There are a few reasons: - because there are well respected POSIX threads people who believe that POSIX threads implementations should be exceptions aware and do the expected thing in that context. (There are equally respected people who believe it should not be easily accessible, if it's there at all, for unconditional conformity to other implementations.) - because pthreads-win32 is one of the few implementations that has the choice, perhaps the only freely available one, and so offers a laboratory to people who may want to explore the effects; - although the code will always be around somewhere for anyone who wants it, once it's removed from the current version it will not be nearly as visible to people who may have a use for it. Source module splitting ----------------------- In order to enable smaller image sizes to be generated for applications that link statically with the library, most routines have been separated out into individual source code files. This is being done in such a way as to be backward compatible. The old source files are reused to congregate the individual routine files into larger translation units (via a bunch of # includes) so that the compiler can still optimise wherever possible, e.g. through inlining, which can only be done within the same translation unit. It is also possible to build the entire library by compiling the single file named "pthread.c", which just #includes all the secondary congregation source files. The compiler may be able to use this to do more inlining of routines. Although the GNU compiler is able to produce libraries with the necessary separation (the -ffunction-segments switch), AFAIK, the MSVC and other compilers don't have this feature. Finally, since I use makefiles and command-line compilation, I don't know what havoc this reorganisation may wreak amongst IDE project file users. You should be able to continue using your existing project files without modification. New non-portable functions -------------------------- pthread_num_processors_np(): Returns the number of processors in the system that are available to the process, as determined from the processor affinity mask. pthread_timechange_handler_np(): To improve tolerance against operator or time service initiated system clock changes. This routine can be called by an application when it receives a WM_TIMECHANGE message from the system. At present it broadcasts all condition variables so that waiting threads can wake up and re-evaluate their conditions and restart their timed waits if required. - Suggested by Alexander Terekhov Platform dependence ------------------- As Win95 doesn't provide one, the library now contains it's own InterlockedCompareExchange() routine, which is used whenever Windows doesn't provide it. InterlockedCompareExchange() is used to implement spinlocks and barriers, and also in mutexes. This routine relies on the CMPXCHG machine instruction which is not available on i386 CPUs. This library (from snapshot 20010712 onwards) is therefore no longer supported on i386 processor platforms. New standard routines --------------------- For source code portability only - rwlocks cannot be process shared yet. pthread_rwlockattr_init() pthread_rwlockattr_destroy() pthread_rwlockattr_setpshared() pthread_rwlockattr_getpshared() As defined in the new POSIX standard, and the Single Unix Spec version 3: sem_timedwait() pthread_mutex_timedlock() - Alexander Terekhov and Thomas Pfaff pthread_rwlock_timedrdlock() - adapted from pthread_rwlock_rdlock() pthread_rwlock_timedwrlock() - adapted from pthread_rwlock_wrlock() pthread.h no longer includes windows.h -------------------------------------- [Not yet for G++] This was done to prevent conflicts. HANDLE, DWORD, and NULL are temporarily defined within pthread.h if they are not already. pthread.h, sched.h and semaphore.h now use dllexport/dllimport -------------------------------------------------------------- Not only to avoid the need for the pthread.def file, but to improve performance. Apparently, declaring functions with dllimport generates a direct call to the function and avoids the overhead of a stub function call. Bug fixes --------- * Fixed potential NULL pointer dereferences in pthread_mutexattr_init, pthread_mutexattr_getpshared, pthread_barrierattr_init, pthread_barrierattr_getpshared, and pthread_condattr_getpshared. - Scott McCaskill * Removed potential race condition in pthread_mutex_trylock and pthread_mutex_lock; - Alexander Terekhov * The behaviour of pthread_mutex_trylock in relation to recursive mutexes was inconsistent with commercial implementations. Trylock would return EBUSY if the lock was owned already by the calling thread regardless of mutex type. Trylock now increments the recursion count and returns 0 for RECURSIVE mutexes, and will return EDEADLK rather than EBUSY for ERRORCHECK mutexes. This is consistent with Solaris. - Thomas Pfaff * Found a fix for the library and workaround for applications for the known bug #2, i.e. where __CLEANUP_CXX or __CLEANUP_SEH is defined. See the "Known Bugs in this snapshot" section below. This could be made transparent to applications by replacing the macros that define the current C++ and SEH versions of pthread_cleanup_push/pop with the C version, but AFAIK cleanup handlers would not then run in the correct sequence with destructors and exception cleanup handlers when an exception occurs. * Cancelation once started in a thread cannot now be inadvertantly double canceled. That is, once a thread begins it's cancelation run, cancelation is disabled and a subsequent cancel request will return an error (ESRCH). * errno: An incorrect compiler directive caused a local version of errno to be used instead of the Win32 errno. Both instances are thread-safe but applications checking errno after a pthreads-win32 call would be wrong. Fixing this also fixed a bad compiler option in the testsuite (/MT should have been /MD) which is needed to link with the correct library MSVCRT.LIB. SNAPSHOT 2001-07-12 ------------------- To be added SNAPSHOT 2001-07-03 ------------------- To be added SNAPSHOT 2000-08-13 ------------------- New: - Renamed DLL and LIB files: pthreadVSE.dll (MS VC++/Structured EH) pthreadVSE.lib pthreadVCE.dll (MS VC++/C++ EH) pthreadVCE.lib pthreadGCE.dll (GNU G++/C++ EH) libpthreadw32.a Both your application and the pthread dll should use the same exception handling scheme. Bugs fixed: - MSVC++ C++ exception handling. Some new tests have been added. SNAPSHOT 2000-08-10 ------------------- New: - asynchronous cancelation on X86 (Jason Nye) - Makefile compatible with MS nmake to replace buildlib.bat - GNUmakefile for Mingw32 - tests/Makefile for MS nmake replaces runall.bat - tests/GNUmakefile for Mingw32 Bugs fixed: - kernel32 load/free problem - attempt to hide internel exceptions from application exception handlers (__try/__except and try/catch blocks) - Win32 thread handle leakage bug (David Baggett/Paul Redondo/Eyal Lebedinsky) Some new tests have been added. SNAPSHOT 1999-11-02 ------------------- Bugs fixed: - ctime_r macro had an incorrect argument (Erik Hensema), - threads were not being created PTHREAD_CANCEL_DEFERRED. This should have had little effect as deferred is the only supported type. (Ross Johnson). Some compatibility improvements added, eg. - pthread_setcancelstate accepts NULL pointer for the previous value argument. Ditto for pthread_setcanceltype. This is compatible with Solaris but should not affect standard applications (Erik Hensema) Some new tests have been added. SNAPSHOT 1999-10-17 ------------------- Bug fix - Cancelation of threads waiting on condition variables now works properly (Lorin Hochstein and Peter Slacik) SNAPSHOT 1999-08-12 ------------------- Fixed exception stack cleanup if calling pthread_exit() - (Lorin Hochstein and John Bossom). Fixed bugs in condition variables - (Peter Slacik): - additional contention checks - properly adjust number of waiting threads after timed condvar timeout. SNAPSHOT 1999-05-30 ------------------- Some minor bugs have been fixed. See the ChangeLog file for details. Some more POSIX 1b functions are now included but ony return an error (ENOSYS) if called. They are: sem_open sem_close sem_unlink sem_getvalue SNAPSHOT 1999-04-07 ------------------- Some POSIX 1b functions which were internally supported are now available as exported functions: sem_init sem_destroy sem_wait sem_trywait sem_post sched_yield sched_get_priority_min sched_get_priority_max Some minor bugs have been fixed. See the ChangeLog file for details. SNAPSHOT 1999-03-16 ------------------- Initial release.