Sun Jun 30 15:28:43 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (cond_timedwait): Fixed the implementation of the ACE_OS::cond_timedwait() wrapper such that if the caller-specified timeout elapses without the condition variable being signaled that errno is always set to ETIME. This fixes some inconsistencies that occurred with POSIX pthreads and Win32 threads. Thanks to Ross Dargahi for pointing this out. * ace/SOCK.cpp: Changed SOCK::close() to use the new ACE_OS::closesocket() call. This should fix some weird bugs that have been lurking in the code for some time now! * ace/OS: Added a new method to ACE_OS called closesocket(). This handles the differences between Win32 and UNIX in their treatment of sockets (e.g., NT requires the use of closesocket(), whereas UNIX requires the use of close(). Thanks to Irfan, Prashant, and Tim for figuring this one out! Sat Jun 29 21:23:04 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/SPIPE_Acceptor.cpp (create_new_instance): Added the flag (FILE_FLAG_OVERLAPPED) in call to CreateNamedPipe. Fri Jun 28 01:31:24 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Svc_Handler.cpp (open): Fixed a small typo that caused a compile error if the DEBUGGING macro was enabled. Thanks to Irfan for finding this. Wed Jun 26 03:19:27 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Synch_T.cpp (ts_get): The tss object allocated off the heap should be released if setspecific fails. Thanks to John Lu for reporting this. * ace/CORBA_Handler.cpp (handle_input): changed ACE_OS::send and ACE_OS::recv to ACE::send and ACE::recv respectively in CORBA_Handler.cpp. Thanks to Irfan for pointing this out. * ace/Service_Config: slightly modified the Service_Config.[h,cpp] in order to make it possible to create an ACE_Service_Config with an external ACE_Reactor as a parameter. On creation I was handling this partly correct (i.e., not creating a new reactor). However, there was no flag for remembering this for a later deletion. Thus, on deletion I was doing a delete on the Reactor regardless of who created it. This is now fixed. Thanks to Karlheinz for pointing this out. * examples/ASX/CCM_App/CCM_App.cpp: Changed all uses of init (int, char **) to init (int, char *[]) to work around a "feature" with MSVC++ 4.x... Thanks to Karlheinz for pointing this out. * ace/OS.h: Added a new typedef of TCHAR to be compatible with Win32 UNICODE type names... * ace/{Mem_Map,DEV_Addr,SPIPE_Addr,FILE_Addr}: Added UNICODE support for Win32 to all interfaces that require filenames. * ace/{SPIPE_Stream,SOCK_IO,FILE_IO,DEV_IO}: Added a new pair of send()/recv() methods that take ACE_OVERLAPPED pointers in order to make it possible to integrate seamlessly with Win32 overlapped I/O. Naturally, these methods are simply "callthroughs" to the ACE_OS versions... * ace/OS: Added a new pair of send()/recv() methods that take ACE_OVERLAPPED pointers in order to make it possible to integrate seamlessly with Win32 overlapped I/O. * ace/SPIPE_Acceptor: Factored out common code in the Win32 implementation of ACE SPIPES (which uses Win32 Named Pipes, of course ;-)). * ace/SPIPE_Acceptor.h: Removed a vestige of the past -- ACE_SPIPE_LISTENER_H should be ACE_SPIPE_ACCEPTOR_H... Wed Jun 19 19:35:12 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/SPIPE_{Acceptor,Connector}: Modified SPIPE_Acceptor and SPIPE_Connector to implement Named Pipes on NT. The public interface remains the same. Sun Jun 16 00:45:41 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Service_Config: Provided public accessor methods for the following data memebers of the ACE_Service_Config: reconfig_occurred_ end_event_loop_ Thanks to Steve Warwick for suggesting this. * ace/Synch*: Added UNICODE support for the ACE synchronization classes (e.g., ACE_Mutex, ACE_RW_Mutex, ACE_Semaphore, etc.). * ace/OS: Added UNICODE support for the ACE_OS::dl_open() function, as well as the ACE_OS synchronization functions. * ace/CORBA_Handler: Added Seth's changes for Orbix 2.0. Sat Jun 1 13:30:55 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Memory_Pool.cpp (remap): Fixed a bug in ACE_MMAP_Memory_Pool::remap (). The test for whether or not the addr falls within the range had a '!' (not) missing, that is, it was failing when it should be succeeding and vice versa. Sun Jun 9 00:01:44 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.24 for testing. * ace: Changed all uses of ACE_Guard m (lock_) to ACE_GUARD (LOCK, lock_), which is a macro that is defined as follows: ACE_Guard ace_mon (LOCK); \ if (m.locked () == 0) return; The reason we need this is (1) detect locking failures, rather than have things fail silently and (2) to automatically detect deadlock in the Reactor. In addition, I also added ACE_GUARD_RETURN, which is similar to ACE_GUARD, except that it returns a "failure" status when the lock is not acquired. Thanks to Karlheinz for suggesting this. * ace/Log_Msg: Added the new thr_state() accessors to Log_Msg. * ace/Thread_Manager.cpp: factored out some common code by having the public interfaces (like resume() and suspend()) utilize the protected interfaces (like resume_thr() and suspend_thr()). * ace/Thread_Manager.cpp: Added sanity checks for suspend(), resume(), cancel(), etc. so that we don't blow up if someone tries to perform an operation on an invalid thread id. * ace/Thread_Manager: Added a suite of new methods for (1) cooperatively "canceling" threads and (2) testing if threads are cancelled (also added similar checks to test if threads are suspended or resumed). The cooperative cancellation mechanism is based on a design fleshed out with Detlef and Karlheinz. It's essentially a compromise between the powerful mechanisms available via POSIX pthreads vs. the totally lame mechanisms available in Win32. Here's how it all works: 1. Added several new methods to ACE_Thread_Manager: cancel(thr_id); -- cancels thr_id cancel_all(); -- cancels all the threads in a Thread_Manager cancel_grp(grp_id); -- cancels a group of threads in a Thread_Manager testcancel(thr_id); -- returns "true" if thr_id has been cancelled 2. Updated ACE_Log_Msg to maintain the current state of a thread in thread-specific-storage (TSS). Actually, it's more clever than that since I really keep a *pointer* to the state of a thread in TSS. This pointer actually points *back* to the ACE_Thread_State field in the ACE_Thread_Manager! I use it as a cache as follows: ACE_Thread_Manager::testcancel (thread_t t_id) { ACE_MT (ACE_Thread_Mutex_Guard m (this->lock_)); // Try to get the cached value out of TSS to avoid lookup. ACE_Thread_State *thr_state = ACE_LOG_MSG->thr_state (); if (thr_state == 0) { // We need to init the cache. int i = this->find (t_id); if (i == -1) return -1; // Update the TSS cache. ACE_LOG_MSG->thr_state (thr_state = &this->thr_table_[i].thr_state_); } return *thr_state == ACE_THR_CANCELLED; } Note that this allows me to avoid searching the Thread_Manager on every Thread_Manager::cancel() access except the first one! 3. I've updated the examples/Threads/test_thread_manager.cpp test file to exercise the new cooperative thread cancellation scheme. It basically spawns a bunch of threads that go into their own event loops doing if (thr_mgr ()->testcancel (ACE_Thread::self ()) != 0) break; every so often. Naturally, the main thread cancels them by saying thr_mgr ()->cancel_grp (grp_id); when it wants to inform them to shut down. * ace/Thread_Manager: Moved the Thread_State enum from the ACE_Thread_Descriptor class to OS.h and renamed it to be ACE_Thread_State. This will make it easier to integrate the state of a thread in thread-specific storage... Sat Jun 8 13:35:17 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (sema_post): revised the implementation of the POSIX Pthreads semaphore implementation just a bit in order to make it slightly more efficient (it now releases the mutex before signaling the condition variable). * ace/OS.i and config-aix.4.1.h: added a fix that separates the decision about whether optlen is a pointer from the decision of the type of optlen. This fixes a problem on AIX. Thanks to Bob Olson for this fix. * ace/Proactor.h: Fixed a small glitch that was causing some compilers to break due to the fact that they can't grok fully qualified destructor syntax in the class definition. Thanks to Alfred Keller for reporting this. * ace/OS.h: Added some Win32 macros (e.g., GENERIC_READ) to the UNIX side of the house in order to compile the Proactor on Win32 and UNIX. Fri Jun 7 19:36:27 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Added a new Bourne shell script called ACE-install.sh that will automatically download and install on UNIX machines. Thanks to Ajit Sagar for contributing this. Thu Jun 6 00:37:02 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.23 for testing. * ace/Local_Tokens: Updated the *.cpp and *.i files to remove warnings and generally improve the conformance to the ACE coding guidelines. * examples: Moved the Proactor directory into the Reactor directory since the Proactor and Reactor are really two implementations of the same basic pattern. * ace/Memory_Pool.cpp: Fixed an inconsistency caused by not updating __INLINE__ to __ACE_INLINE__. Thanks to Neil Cohen for finding this. * ace: Added support for the Florida State PTHREADS package. Thanks to Gary Salsbery and Eric Beser for helping with this. * ace: Added support for the m88k OS. Thanks to Gary Salsbery and Eric Beser for helping with this. * ace/OS.h: Added a default value of NULL to the sigwait() OS wrapper method to simplify the usecase in certain cases. * ace/Memory_Pool.cpp (commit_backing_store): Fixed what is hopefully the last typo related to the ACE_DEFAULT_BASE_ADDR macro. Thanks to Neil B. Cohen for reporting this. Thu Jun 6 15:31:40 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * ace/CORBA_Handler.h: This version of the CORBA_Handler works with Orbix 2.0. Most of the changes involve the use of C++ Exception Handling (the removal of IT_X), and the location of some CORBA system exception classes. * ace/Event_Handler.h: Added handle_*_complete methods for the Proactor. The Proactor now takes Event_Handlers and calls back the **_complete methods when overlapped I/O operations have completed. * ace/Service_Config.h: Added static accessors for the Proactor to the Service_Config object. Similar to the Reactor accessors, applications can now use the Service_Config object as the global access point to the Proactor event demultiplexor. Wed Jun 5 22:40:28 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (t_free): Added a check to avoid deleting a NULL ptr. This was causing problems for the TLI_Acceptor. Thanks to Ajit Sagar for reporting this. * ace/Makefile: Changed things back so that both static and dynamic libs are built by default... Thanks to Brad Brown for pointing this out... Sat Jun 1 13:49:51 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.22 for testing. * Added Tim Harrison's latest tests for Proactor. Sat Jun 1 13:30:55 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Memory_Pool.cpp (remap): Fixed a bug in ACE_MMAP_Memory_Pool::remap (). The test for whether or not the addr falls within the range had a '!' (not) missing, that is, it was failing when it should be succeeding and vice versa. Fri May 31 16:31:13 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Rereleased version 4.0.21 for testing. * Copied over new versions of the Proactor -- there was a small problem with the UNIX version that caused it not to compile... Fri May 31 00:03:41 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * examples/Proactor/test_proactor.cpp: Added a test program for the Proactor. Check examples/Proactor/README for more details on this test example. * ace/Proactor.cpp: Added first pass at the Proactor for win32. The Proactor is a Reactor-like abstraction that uses asynchronous I/O, rather than synchronous I/O. On Windows NT we implement the Proactor using overlapped I/O. We'll soon be porting the Proactor to Solaris using POSIX 4.x aio_* API for real-time programming. Unfortunately, the Proactor has not yet been integrated with Windows NT WaitForMultipleObjects since it appears that I/O Completion ports are not "waitable" objects on Windows NT. Does anyone know if this has been fixed in release 4.0? Thu May 30 05:51:23 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Synch.i: Reordered the definitions of several methods in Synch.i to avoid warnings from G++. Thanks to Marco Sommerau for noticing this. * ace/Log_Msg.cpp: There was a problem having to do with the order in which static objects by G++. In particular, G++ was calling the constructor of the lock_ mutex in Log_Msg.cpp *after* it was being used for the first time. The right solution is to make lock_ be a pointer and allocate it once in the instance() method. Thanks to Marco Sommerau for finding this problem. * ace/TLI_Acceptor.cpp (close): Make sure to only close down the TLI_Request_Queue if queue_ is non-NULL! Thanks to Ajit Sagar for spotting this. * ace: Changed all the enums in the various SysV wrappers from things like CREATE and OPEN to ACE_CREATE and ACE_OPEN to avoid name clashes with other systems. In addition, also changed NONBLOCK to ACE_NOWAIT to avoid clashes with the existing ACE_NONBLOCK macro! Thanks to Steve Warwick for suggesting this. * ChangeLog: Changed all usages of the INLINE macro to ACE_INLINE. This avoids name clashes with other systems. Thanks to Chris Eich for suggesting this. Mon May 27 13:03:58 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Map_Manager.cpp (trybind): The function trybind() doesn't return a value as it should. A return was missing from the last line of the function. Thanks to Stuart Powell for finding this problem. * ace/Acceptor.cpp (dump): The debug print referred to "scheduling_strategy_" which doesn't exist in the relevant class. Replacing it with "concurrency_strategy_" fixes the problem. Thanks to Stuart Powell for finding this problem. * ace/Mem_Map.i: Make sure to close down the file mapping correctly when we upmap the view! * ace/OS.h: Changed the default shared memory address from 16 M to 64 M in order to work around problems with AIX. * ace/Memory_Pool: Moved and renamed the enums in ACE_Shared_Memory_Pool so that they'd be in OS.h. This makes it easier to remember to change them if they are incorrect. Thanks to Lionel Mommeja for suggesting this. * ace/Stream.cpp: Removed #if defined (ACE_HAS_THREADS) * ace/OS.cpp (svc_run): Added a Win32 try/except block to Spawn_Args::svc_run so that we catch all Win32 structured exceptions in order to make sure that we clean up correctly when the thread exits. Sun May 26 11:37:08 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp: Used the new Win32 exception integration in order to ensure that the Name Server will transparently work correctly when the backing store is extended by other processes on the same machine. * ace/Memory_Pool.cpp: Revised the ACE_MMAP_Memory_Pool to export the mechanism for extending the virtual memory mapping. This can now be called by other programs (e.g., in order to integrate with Win32 Structure Exception Handling). * ace/Memory_Pool.cpp: Changed the use of ACE_OS::lseek() to ACE_OS::filesize() in order to determine the current offset when we're remapping the address space. Wed May 22 13:08:44 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/OS.h (ACE_DEFAULT_TIME_SERVER_STR): Added a new entry for ACE_DEFAULT_TIME_SERVER_STR. * ace/Malloc_T.cpp (advance): Fixed two small bugs. In ACE_Malloc::try_bind(), if we have a match we need to set pointer to node->pointer_ and not node->name_. Ina ACE_Malloc_Iterator::advance(), the continue in the for loop should be for strcmp != 0 instead of == 0. Sun May 19 12:03:11 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.20 for testing. * ace/Service_Record.cpp (remove): Fixed a very mysterious bug that was caused by deleteing an object and then trying to access it's next_ pointer. This worked on UNIX, but fortunately the MSVC++ compiler does special things to deleted memory and the bug was revealed! At long last, MSVC++ does something right ;-). * ace/Module.cpp: Revised the code a bit to use the ACE_SET_BITS and ACE_CLR_BITS macros to improve readability. * ace/SV_Semaphore_Complex.cpp (open): Fixed a race condition where we weren't correctly checking for EIDRM. Thanks to Michael Fortinsky for reporting this. Sat May 18 10:49:04 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Added a new ace.mak and ace.mdp file for Win32. * ace/Service_Record.cpp: Rewrote some code to work around a bug with MSVC++. This bug was causing problems since the ACE_Shared_Object methods were getting called, rather than the Test_Task methods used in the examples/ASX/CCM_App.cpp file. Thanks to Tom Leith for pointing out this problem! * ace/OS.h: Added a new #define for ACE_DEFAULT_SVC_CONF to deal with differences between NT and UNIX pathnames. * ace/Time_Value.h: Moved the definition of the INLINE macro from OS.h to Time_Value.h so that it will be in scope for Time_Value.i. Thanks to Neil Cohen for reporting this. * examples/Reactor/Misc/signal_tester.cpp: There was a typo in signal_test.cpp. It had mean to use handle_input() rather than handle_output() to exercise the asynchronous signal handling capabilities. This is fixed now. * apps/Synch-Benchmarks/Benchmark.cpp (thr_id): Added a new #if define (ACE_HAS_DCETHREADS) to make this work on AIX. Thanks to Greg Wilson for reporting this. * ace/Local_Name_Space: Moved ACE_NS_String and ACE_NS_Internal from the *.cpp file to the *.h file to work around a "feature" of the AIX C++ compiler. Thanks to Greg Wilson for reporting this. * ace/Reactor.h (ACE_Handler_Repository): Changed the type of cur_size_ from size_t to ssize_t to avoid type mismatches. * ace/Name_Request_Reply.cpp (decode): Fixed some inconsistencies between signed and unsigned loop counters... * ace/OS.h: Changed the typedef of pid_t on Win32 from DWORD to long to be consistent with UNIX. Thu May 16 18:49:14 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.19 for testing. * build/SunOS5.4/ace/ACE.cpp (ldfind): Fixed a bug in ldfind() that caused a segfault if we didn't resolve the filename in the LD_SEARCH_PATH. * ace/Reactor and Timer_Queue: Changed the interface of cancel() to include a const void **arg. If arg is non-NULL then it will be set to point to the ``magic cookie'' argument passed in when the Event_Handler was registered. This makes it possible to free up the memory and avoid memory leaks. * ace/Service_Record: Removed the #ifdefs that checked if ACE_HAS_THREADS because we ought to be able to build all of this stuff now that we've got the ACE_OS wrappers. * ace/Svc_Conf.y: Removed the #ifdefs that checked if ACE_HAS_THREADS because we ought to be able to build all of this stuff now that we've got the ACE_OS wrappers. * ace/OS.h (ACE_DEFAULT_BACKING_STORE): Made a different ACE_DEFAULT_BACKING_STORE for NT and for UNIX to handle the differences in directory separator characters... Wed May 15 18:45:48 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * examples/ASX/CCM_App/CCM_App.cpp: Fixed up the test application so that it should support dynamic linking on Win32. Thanks to Tom Leith for reporting this. * ace: Added the ACE_Export macro to class ACE and the various "helper" classes related to ACE_Malloc in order to have this work properly when use with ACE_Malloc outside of the ACE DLL. * ace/Svc_Conf.l: Added support for '\' and ':' in the regular expression for an ACE pathname in order to support Win32 filenames (e.g., "C:\foobar\"). Thanks to Tom Leith for reporting this. * ace/Malloc_T.cpp (ACE_Allocator_Adapter): Moved the constructor from the *.i file to the *.cpp file and added a new constructor that takes both a pool_name *and* a lock_name. This is necessary because the ACE Malloc now has this API, as well. * examples/Threads/test_barrier.cpp (main): Fixed a bug that was caused by the main() thread exiting before all the other worker threads had finished "waiting" on their Barrier. The fix is to use ACE_Thread_Manager to control the thread exits... Mon May 13 00:03:09 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * examples/Connection/non_blocking/CPP-acceptor.cpp: Modified the implementation of the Svc_Handler::open() method so that it will truly behave as an iterative server (as advertised...). Currently, it behaves as a half-iterative server (i.e., the Oneshot_Acceptor is iterative, but the Svc_Handler is "concurrently"), which is causing problems because we're wiping out the values of Svc_Handler each time through the main event loop. Thanks to Gerolf Wendland for noticing this problem. * ace/Log_Msg: Added an alternative logging mechanism that makes it possible to integrate variable argument lists from other logging mechanisms into the ACE mechanism. Thanks to Chris Lahey for proposing this. * ace/Synch.h: Moved ACE_Process_Mutex so that it appears *after* ACE_Mutex (since it depends on ACE_Mutex). Thanks to Dieter Quehl for finding this. * Released version 4.0.18 for testing. * ace/Name_Space.cpp: Added the ACE_BUILD_DLL macro at the beginning of this file so that it will build as a DLL on NT correctly. * ace/Name_Space.cpp: Added a default constructor for ACE_Name_Binding so that it will compile when used as a template argument for ACE_Unbounded_Set. Sun May 12 14:23:44 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Map_Manager: Totally reworked the Map Manager so that it can be more flexible with respect to where the allocator comes from. This is necessary for certain parts of ACE (e.g., Local_Name_Space) that must be very careful about which allocator is used to manage memory. * ace/Synch: Moved ACE_Process_Mutex and ACE_RW_Process_Mutex *outside* of the ACE_HAS_THREADS #ifdef since these are now always defined on all platforms (because the Process_Mutex stuff uses SV Semaphores, which are portable even if we don't have threads!). * ace/Naming_Context.cpp (parse_args): Removed an unnecessary call to strdup(). This was detected by Purify! Sun May 12 14:26:20 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (shared_bind): Fixed a small bug. In computing type_len in shared_bind(), we needed to add 1 to account for the NUL character. * ace/Local_Name_Space.cpp: (list_types): Fixed some potential memory leaks. In list_types() as well as list_type_entries() calling pattern.char_rep() was allocating memory which was never getting deleted. Similarly, in list_names() and list_values, call to char_rep was also allocating memory that was not getting deleted. Sat May 11 16:19:51 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Log_Msg.cpp (local_host): added a static member to ACE_Log_Msg that maintains the local host name. Now, instead of always displaying "", we can insert the actual local name into the object and then it will be automatically printed when VERBOSE mode is enabled. Thanks to Chris Lahey for suggesting this. * ace/Local_Name_Space.cpp: Used the new ACE_Name_Binding class to simplify the implementation of all the list_* methods in ACE_Local_Name_Space. * ace/Name_Space.cpp: Made a number of changes to the ACE_Name_Binding class in order to make it work more efficiently and concisely (e.g., reduce the amount of copying and eliminate the need for converting the type field back and forth to/from ACE_WStrings). * examples/Service_Configurator/IPC-tests/server/Handle_L_Dgram.cpp: Fixed a typo where #if !defined (ACE_LACKS_UNIX_DOMAIN_SOCKETS) was *before* the #include of the header file that defines this macro! Thanks to Nigel for pointing this out, as well. * ace: Added a number of fixes to make ACE compile on SCO UNIX 3.2.4 using gcc 2.7.2. Thanks to Nigel Lowe for helping with this. * netsvcs/lib/Name_Handler.cpp: operation_ needs to be declared as just LIST_OP and not ACE_Name_Handler::LIST_OP. For some strange reason NT complains otherwise. Thanks to Prashant for finding this. Fri May 10 01:09:17 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Added new versions of ace.mak and ace.mdp to fix a couple of minor typos. * ace/Malloc_T.cpp (ACE_Malloc): Added a new constructor that takes both the pool_name and the lock_name explicitly. This is more flexible. Thanks to Ramesh Nagabushnam for suggesting this. * ace/Malloc_T: Modified the constructor of ACE_Malloc so that by default the name of the memory pool (which is also used as the name of the lock...) is ACE_DEFAULT_MUTEX. * Released version 4.0.17 for testing. * ace/Connector.cpp (create_AST): Must reset errno = EWOULDBLOCK to avoid making caller's think that something else has gone awry... Thanks to Steve Warwick for chasing this down! * ace/ACE.cpp (handle_timed_complete): If you issue a non-blocking connect on a socket, you will have back a failure with errno = EINPROGRESS. Then, if for some reason, the connection could not be established, the select (which you have to issue to know about the completion of connection) will return you the fd set both in the read mask and in the write mask (infact select returns 2 in this case). The behaviour above affects the method ACE::handle_time_complete, so I changed the last part of the above method, this way: if (n <= 0) { ... ... } else if (rd_handles.is_set(h)) { char dummy; // The following recv() won't block provided that the // ACE_NONBLOCK flag has not been turned off . if (ACE::recv (h, &dummy, 1, MSG_PEEK) <= 0) return ACE_INVALID_HANDLE; } // 1. The HANDLE is ready for writing or 2. recv() returned that // there are data to be read thus the connection was successfully // established. return h; That is, I reversed the sense of the tests so that the rd_handles() is checked first for failure along with the recv(). Thanks to Antonio Tortorici for suggesting this. * examples/Threads/test_thread_manager.cpp (main): Changed argv[1] to argv[2]. Thanks to Andres Kruse for finding this problem. * examples/Connection/non-blocking: Fixed some nasty bugs that caused the non-blocking connector and acceptor test programs to crash and burn. Thanks to Steve Warwick and Gerolf Wendland for finding this problem. * ace/SV_Semaphore_Simple.cpp (open): Check for ACE_INVALID_SEM_KEY and return -1 in this case. * ace/Synch.h: Removed the default value of 0 for the name of the ACE_Process_Mutex and the ACE_RW_Process_Mutex and replaced it with a new macro called ACE_DEFAULT_MUTEX. Using 0 didn't make any sense on either NT or UNIX because process-wide Mutexes should be named! * ace/SV_Semaphore_Simple.cpp (name_2_key): Added a check for name == 0 and bail out of that's the case rather than crash! * ace/Reactor.cpp (wait_for_multiple_events): In the Reactor's wait_for_multiple_events method, the do {} while () around the select/poll system call is trying to wait until some "good" event occurs, with handle_error() taking care of unexpected problems. In the case of a bad file descriptor, however, handle_error() returns 0. This was exiting the loop because the loop exit condition was: do { /* ... */ } while (nfound == -1 && this->handle_error () > 0); which eventually causes ACE_Service_Config::run_event_loop() to exit. Since the offending file descriptor is handled by handle_error(), the loop should continue. Therefore, I've changed the while to read: while (nfound == -1 && this->handle_error () >= 0); Thanks to Eric C. Newton for providing this fix. * ace/INET_Addr.cpp (string_to_addr): Changed the order of the parameters so that the form is now "ip-address:port". This should (finally) be consistent for both string_to_addr() and addr_to_string()... * ace/Log_Msg: Changed the char * parameter of the log() method to const char *. Thanks to Chris for suggesting this! * ace/Synch_T.cpp: Moved the #if defined (ACE_HAS_THREADS) down to the right part of the file... Thanks to Alex Karev for finding this. * ace/Malloc.h: Added a very important #else... Thanks to Alex Karev for finding this. * ace/Signal.i (ACE_Sig_Guard): Changed the sense of the #ifdef tests in the ACE_Sig_Guard constructor and destructor from #if !defined (ACE_MT_SAFE) to #if 0 // !defined (ACE_MT_SAFE) in order to get the right semantics for signals (which should be blocked "process wide"). * ace/Synch_T.h: put the frigging copy constructor in the public section of ACE_Atomic_Op because it was causing trouble for NT... Mon May 6 00:11:37 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.16 for testing. * ace/Log_Msg.cpp: Was defining ACE_Recursive_Thread_Mutex in the wrong place (i.e., outside of the #if defined (ACE_MT_SAFE). This was screwing up HP/UX. Thanks to Neil Cohen for pointing this out! * Incorported new versions of ace.mdp and ace.mak for Win32. Sun May 5 16:18:43 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * netsvcs/lib/Server_Logging_Handler.cpp (handle_input): Fixed a bug whereby handle_input() was returning the number of bytes read by handle_logging_record() rather than 0 or -1. This was causing problems for the logger since it was hanging in read()! * ace/Synch_T.cpp: Fixed some minor problems with the use of const in ACE_Atomic_Op that was causing warnings. Sat May 4 16:31:46 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace/SOCK_Dgram_Bcast: Reimplemented most of the ACE socket broadcast mechanism to (1) clean up the code and have it use other parts of ACE and (2) make it work for Windows NT. Thanks to Steve Weismuller for the basic ideas here. * ace/INET_Addr: Added a new method to set the port number without changing the host address. This is useful in the broadcast class. * ace/Log_Record.cpp (print): Changed the hack of replacing the newline with a call to flush() instead. This seems like a better fix. Thanks to Alex for suggesting it. * ace/Log_Msg.cpp (log): Moved the order of the print operations so that the ostream one goes last. This avoids a nasty problem due to the fact that it replaces the newline (if any). Thanks to Alex for pointing this out too! * ace/Log_Msg.cpp (log): Don't auto-increment bp at the end of the log() method, instead just terminate it: *bp = '\0'; // Terminate bp. This makes the length computation correct... Thanks to the ever-vigilent Alexandre Karev for reporting this. * examples/Shared_Malloc/test_malloc.cpp: Fixed a bug where a void * that was really an int was being cast incorrectly. Thanks to Raj for pointing this out. Sat May 4 12:51:25 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Reactor.cpp (find): Modified the ACE_Handler_Repository so that it doesn't crash when it isn't initialized properly. Thanks to Darrin for pointing this out. * ace/Synch_T.cpp (wait): Changed the call wait(ACE_Time_Value *) to wait(const ACE_Time_Value *) since wait() will not change the time value object. * ace/Thread_Manager.cpp (wait): Changed the call wait(ACE_Time_Value *) to wait(const ACE_Time_Value *) since wait() will not change the time value object. Thanks to Chris Lahey for suggesting this. * ace/Synch.h: Added a (const char * = 0) argument to ACE_Null_Mutex so that it would work correctly with ACE_Thread_Mutex and ACE_Process_Mutex... Fri May 3 17:26:07 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * netsvcs/lib/Name_Handler.cpp (lists_entries): Used '_' with name/value/type in ACE_Name_Binding to be consistent with the notation. Fri May 3 02:24:19 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Synch: Updated ACE_Process_Mutex to use SV_Semaphore_Complex as the UNIX implementation. This seems like the only portable and robust way to get the same semantics on UNIX and on NT! * ace/Malloc_T: It turned out there was a subtle bug with ACE_Malloc on NT when used with the ACE_Process_Mutex concurrency policy. On NT, we were storing the *HANDLE* of a mutex in shared memory. Naturally, this didn't make any sense since each HANDLE is relative to only one process!!! (duh ;-)). This worked fine on UNIX because mutex_t's created with USYNC_PROC mode can be stored in shared memory and accessed directly... We fixed this by changing ACE_Malloc so that the LOCK is allocated in non-shared memory. We then fixed ACE_Process_Mutex so that on UNIX it is implemented with SV_Semaphore_Complex. This is (a) portable and (b) it gives the right semantics. As a result, we were able to totally eliminate the storage of the lock in shared memory. Therefore, it doesn't matter of the host crashes anymore! In addition, we were able to totally remove the ugly "init_finished" lock that was previously stored in shared memory by ACE_MMAP_Memory_Pool. This is *much* cleaner!! Thanks to Karlheinz et al for pointing this problem out in the first place! * ace: Removed a stray file called Svc_Conf_tokens.h. This was causing problems on Win32 due to name clashes... Thanks to Adam Miller for pointing this out. * ace/SString.cpp (strstr): there were some for (size_t j; ...) {} if (j == x) constructs in the code that have as of recently become non-standard. Thanks to Darrin for reporting this. Tue Apr 30 00:18:46 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Thread.i (self): Commented out the ACE_Trace call in the ACE_Thread::self() methods to avoid infinite recursion problems on Win32. * ace/Token.cpp (ACE_Token): Commented out the ACE_Trace call in the constructor of ACE_Token to avoid infinite recursion problems on Win32. * ace/Log_Msg.cpp (instance): Moved the definition of the static variable keylock_ into the static instance() method in order to avoid "order of initialization" problems on Win32. Thanks to Tim for figuring this out! Sun Apr 28 17:07:58 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * netsvcs: Rearranged and renamed the subdirectories so that they are now called ./netsvcs/{lib,clients,servers}. In addition, moved the Logger, Naming, and Tokens examples from ./examples into the ./netsvcs/clients directory in order to make the relationships more clear. Sat Apr 27 14:23:43 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Space.cpp (operator ==): Created a new file Name_Space.cpp and moved the definitions for ACE_Name_Binding::operator== and ~ACE_Name_Space in it. Also modified code to use '_' at the end of name/value/type in ACE_Name_Binding. Sat Apr 27 16:00:03 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/INET_Addr.i (get_host_addr): There was an error where sprintf(s,"%d:%s" ...) should have been %s:%d. Thanks to Raj for pointing this out. Mon Apr 22 01:24:45 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Dump_T.h: #included "ace/Dump.h" so that this file will compile on HP/UX. Thanks to Neil Cohen for reporting this problem. * Released version 4.0.15 for testing. * ace/Synch_T: Added the appropriate "const" qualifiers to certain operators in ACE_Atomic_Op. Sun Apr 21 12:54:18 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Fixed a problem that Dieter was having with the Name Server when toggling between Naming_Contexts. Basically, what I've done is 1. Moved the unmap() call into the ACE_Mem_Map::close() method (this ensures that the region is correctly unmapped when the Mem_Map is deleted). 2. I've removed the call to this->allocator_->remove () within ~ACE_Local_Name_Space. This ensures that we don't remove the backing store file. * ace/Mem_Map.cpp: Modified the ACE_Mem_Map::remove method to call ACE_Mem_Map::close() in order to share code. * ace/Mem_Map.cpp: Modified the ACE_Mem_Map::close() method so that it unmaps the mapped region before closing down the backing store. This prevents "dangling mapping." * ace/Local_Name_Space.cpp (ACE_Local_Name_Space): Initialized all the pointers to NULL. * ace/Synch_T.cpp (ACE_Atomic_Op): Enhanced the Atomic_Op implementation by adding an assignment operator and disallowing the copy constructor (forcing objects of ACE_Atomic_Op to be passed by reference). * examples/Naming/Client/Client_Test: Reorganized the code for Client_Test so that it will run correctly as a Win32 netsvc (e.g., added the ACE_Svc_Export macro). Sun Apr 21 20:23:40 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Naming_Context.cpp (close): Added a new method to Naming_Context called close() that deletes the instance of name_space_. * ace/Local_Name_Space.cpp (ACE_Local_Name_Space): Added stuff to the destructor of Local_Name_Space so that it calls remove on the allocator_ to ensure we unmap the file. Also, we delete the allocator_. * examples/Naming/Client/Client_Test.cpp (set_proc_local): Changed set_proc_local (), set_node_local() and set_host() so that before we change name space, we do a close() on Naming_Context. The close ensures that we unmap the file as well as delete the instance of the name space. Sat Apr 20 12:39:20 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Removed the Name_Options.* files and merged the ACE_Name_Options class into Naming_Context.*. This simplifies some order of include problems... * ace/Naming_Context.cpp (init): Switched the code so that rather than always opening with ACE_Naming_Context::PROC_LOCAL, we use ACE_Name_Options::context(), which can be overridden by using the new -c command-line option (e.g., -c NODE_LOCAL). Thanks to Dieter for suggesting this. * ace/OS.i: Modified the Win32 implementation of all the ACE_OS::flock_*lock() methods so that they'd have the same behavior as the UNIX ones with respect to a length of 0 meaning "lock the entire length of the file." This works, of course, by having each method check if len == 0, and if so, calling GetFileSize() to set the size of the file. * ace/Synch.h: Changed the default len value for all the File_Lock::*acquire* methods to 1 rather than 0 to work around a problem with Win32. Thanks to Detlef for reporting this. * ace/ACE.cpp (ldfind): Added a strdup() of the LD_SEARCH_PATH returned by getenv() so that we don't overwrite the environment variable by using strtok(). Thanks very much to Prashant for figuring this out! Thu Apr 18 22:13:43 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * examples/Naming/Client/Client_Test.cpp (list_value_entries): Added check in list_name_entries and list_value_entries to see if type actually exists before trying to print it out. Wed Apr 17 16:40:42 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.14 for testing. * Added Tim Harrison's new instructions for building and using ACE as a DLL and SLL on Win32. * ace/Svc_Conf.h: Due to the new changes to the yacc code, I had to move the definition of YYSTYPE into the Svc_Conf.h file rather than have it in the Svc_Conf.y file. This solves a variety of multiple-inclusion problems... * ace: Modified all the yacc token symbols so that they will be prefixed with ACE_. This avoids namespace pollution problems. * Added the ACE_STATIC_SVC* macro support to a number of files so that the static versions of all the ACE services will be registered with the ACE Service Repository correctly. Thanks to Jesper for pointing this out! * ace/OS.h: the first occurence of typedef thread_t tid_t; in the OS.h file should read typedef pthread_t tid_t; Thanks to Jan Rychter for reporting this. * ace/Read_Buffer.cpp: Fixed rec_read() so that it will fail correctly when new fails! Wed Apr 17 19:05:42 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * netsvcs/lib/Name_Handler.cpp (type_request): Here is a brief description of the changes made to ACE_Name_Handler to simplify the dispatch method and make it more elegant: 1) I created a table called op_table_. It contains pointers to member functions of ACE_Name_Handler. This table can be indexed using the msg_type. However, I went one step further to have all the list methods be grouped under two methods. That is, requests for list_names, list_values, and list_types are handled by one method called lists() and similarly, requests for list_name_entries, list_value_entries, and list_type_entries are handled by another method called lists_entries(). A MASK is used to have the op_table_ index to the same method for these requests. 2) I also created another table called list_table_. This is primarily used by lists() to keep track of a couple of things: + pointers to member functions of Name_Handler that handle the actual request. + pointers to member functions of Name_Handler that act as factories to create the appropriate request to send back. + description of the message type. A different MASK is used to index into the list_table_ to invoke the appropriate method or get the appropriate description. 3) Within the method lists_entries(), I once again make use of the pointers to member functions technique. This time, however, I use pointers to member functions of Naming_Context which I assign in the switch statement. Tue Apr 16 13:03:49 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.13 for testing. * ace/INET_Addr.i: moved the get_port_number routine up to the top of the file so it gets defined before it is used. Linux did not like it where it was... Thanks to Neil Cohen for reporting this. * ace: Split the Dump.* files into Dump_T.* files. This is necessary to support the template policies of various C++ compilers. * ace/Malloc_T.cpp (dump): Fixed an erroneous use of -> since guard_ is a non-pointer... Thanks to Neil Cohen for reporting this. * ace/Log_Record.cpp (print): Finally figured out how to do extensible ostream logging so that it will seamlessly work with ACE_Log_Msg. This required one change to ACE_Log_Record, however. In Log_Record.cpp: In ACE_Log_Record::print(char *, int, ostream, size_t len), added char *t = this->msg_data_[len - 1]; if (t == '\n') *t = '\0'; s << this->msg_data_ << endl; This is required because the virtual function overflow() in streambuf does not get called until endl is called to flush the buffer. Note that (at least on AIX), '\n' is treated separately from endl and will not cause the overflow() function to be called. Thanks to Chris Lahey for this idea. * ace/OS: Added a new wrapper for strrchr(). * ace/Token_Collection.cpp (renew): Added a cast of (const char *) to token_name so that the conversion operator will get called correctly. Thanks to Gonzalo Diethelm for reporting this. Tue Apr 16 13:29:32 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Request_Reply.cpp (decode): Fixed a small bug in encode() and decode() of ACE_Name_Request. When doing ntohs and htons, we only ought to do it for the name and value part of data. type needed to be left alone since it is not a short. Mon Apr 15 02:31:00 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * include/makeinclude/platform_hpux_orbix.GNU: Added David.L.Sames.1@gsfc.nasa.gov (David Sames) config*.h file and platform_*.GNU file for HP/UX 9.x using Orbix. * ace/OS.i: Fixed all uses of pthread_cleanup_{pop,push} so that they will work correctly if (1) the OS implements the as macros and (2) if these macros must obey strict nesting rules... Thanks to Reginald S. Perry for pointing out how to do this effectively. * ace/Thread_Manager.h: Fixed a typo where the typedef typedef (ACE_Thread_Manager::*THR_FUNC)(int, int); was lacking a return value. Thanks to Reginald S. Perry for reporting this. * netsvcs/bin/main.cpp (main): Fixed the main program so that it passes the options correctly for the statically linked service invocations. Thanks to Jesper for reporting this. * examples/Naming/Client/Client_Test: Updated the client test program so that it will work on Window NT, where it's not valid to select() on non-socket HANDLEs (ugh). Thanks to Jesper for pointing this out... * Released version 4.0.11 for testing. Mon Apr 15 00:20:02 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (strstr): Fixed ACE_NS_String::strstr to function properly. It was assuming that length of the pattern as well as the string was in units of ACE_USHORT16 when in reality the length is number of bytes. * ace/Local_Name_Space.cpp (value_): Modified ACE_NS_Internal so that instead of taking an ACE_NS_String for type, it now take a char*. It therefore keeps type around as a char*. This should help solve some problems we were encountering with byte ordering when sending data between NT machines and Sun. * netsvcs/lib/Name_Handler.cpp: Cleaned up lists_entries by making use of pointer to member functions. A single call to the appropriate list method takes care of everything. Sun Apr 14 16:21:32 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/INET_Addr.i (addr_to_string): Swapped the order that the hostname and port number are printed to make this consistent with the way that ACE_INET_Addr::string_to_addr works. Thanks to Ashish Singhai for pointing this out. * ace/TLI_Acceptor: Changed the methods to take an ACE_Addr rather than an ACE_INET_Addr. This allows other address families (such as DECnet) to be used with the TLI wrappers. Thanks to Ajit Sagar for suggesting this. * ace/Local_Name_Space.cpp (list_types): Removed the use of regular expression matching for the name and value types of the name server and replaced these with substring matching instead. This is necessary since we're working with wide characters here, and it doesn't make any sense to perform regular expressions on these puppies... * ace/Local_Name_Space.cpp: Added a new strstr() method on all the ACE_NS_String class so that we can compare substrings for wide character types. * ace/SString: Added a new strstr() method on all the ACE_[WSC]String classes so that we can compare substrings, even for wide character types! * apps/gperf: Incorporated the GNU gperf program from the FSF. We're going to make some improvements to this for a subsequent paper. Sun Apr 14 15:30:05 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Request_Reply.h: Changed the enum values for the Constants such as BIND, REBIND, etc. The new values allow us to do bitwise operations using MASK to be able to dispatch the same method for all the list operations. For use case of this, please see Name_Handler::dispatch(). * netsvcs/lib/Name_Handler.cpp (dispatch): Completely changed dispatch() so that now it uses a table of pointer to member functions to call the appropriate method when a request arrives. In addition, requests for LIST_NAMES, LIST_VALUES, and LIST_ENTRIES are now handled by one method called lists() in which I factored out a lot of common code. Similarly, requests for LIST_NAME_ENTRIES, LIST_VALUE_ENTRIES, and LIST_TYPE_ENTRIES are handled by lists_entries(). This has really cleaned up the code. Sat Apr 13 15:26:51 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Request_Reply.cpp: Made changes so that all byte-ordering computations take place in encode and decode methods of ACE_Name_Request_Reply and ACE_Name_Reply only. Previously some of these computations were taking place in Get/Set methods such as name_len() which was highly error prone. (init): Added new methods called init() to both ACE_Name_Request and ACE_Name_Reply that initialize length to size of transfer_. This is needed since the length gets set only once in the constructor and after that each call to encode() switches the byte ordering causing problems. Sat Apr 13 11:44:16 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.10 for testing. * ace: Removed the ACE_HAS_THREAD_T macro from various places. This was unnecessary given the new scheme for dealing with threading. * ace/config-linux-pthread.h: Fixed a typo that was preventing compilation from working. Thanks to Jan Rychter for finding this. Fri Apr 12 13:17:47 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * apps/Gateway/Gateway: The gateway application appears to be working again! * ace/OS.cpp (thr_create): Based on conversations with Chris Lahey, modified the pthreads ACE_OS::thr_create implementation so that it doesn't try to assign the hthread_t thr_handle since this is meaningless in pthreads... * apps/Gateway/Gateway/Gateway.cpp: Fixed a typo that was causing the system to go into an infinite loop when the config file was empty! * apps/Gateway/Gateway/Gateway.cpp: I was forgetting to set the line_number to 0 before calling read_entry()... * ace/Parse_Node.cpp: Fixed a dumb error that arose because I forgot that you can't call dlerror() twice and get the same result (it returns NULL the second time). As usual, RTFM. * ace: Removed ACE_HAS_TLI from all the SunOS 4.x config*.h files and platform_*.GNU files. There are just too many bugs with TLI on SunOS 4.x to support it by default... * examples: Made some minor fixes to make the examples all compile with GCC. * apps/Gateway/Gateway/Gateway: Reimplemented the Gateway application as to be an ACE network service. This should make life much easier on platforms with broken C++ templates... * apps/Gateway/Gateway: Revised the Config_Parser.* files so that templates would be split from the non-templates. This avoids bugs with some C++ compilers... * ChangeLog: added the ACE_TEMPLATES_REQUIRE_SPECIALIZATION flag to the config-linux-pthreads.h file. Thanks to Jean-Francois.Ripouteau@netsurf.org for reporting this, and also for archiving the ACE mailing list. Thu Apr 11 01:37:25 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Released version 4.0.9 for testing. * ace: regenerated all the manual pages and html pages. * Built ACE successfully on SGI IRIX 5.3 using GNU G++ 2.7.2. * ace/Thread_Manager: renamed the descriptor() accessor methods to hthread_descriptor() and thread_descriptor() to avoid ambiguities due to the fact that hthread_t and thread_t can now be the same type. * ace/Thread_Manager: revised the hthread_t so that it is always large enough to hold valid thread ids. * ace: Moved the config-osf1.h file to be named config-osf1-3.2.h in anticipation of OSF/1 4.0... * ace: Changed all occurrences of wchar_t to ACE_USHORT16, which is a more portable way of representing wchar_t's so that we can pass them across the network... Unfortunately, wchar_t tends to be different sizes on different machines, so we can't use the binary form!!! * ace: Added many, many small changes to get ACE to compile on OSF/1 3.2D using DEC C++ 5.3. Thanks to Tom Marrs <0002104588@mcimail.com> for slogging through all of this stuff! * ace/Thread_Manager.h: Fixed a stupid oversight where I forgot to add a -1 to the trailing argument of spawn*. Thanks to Neil Cohen for spotting this. * ace: Added a new ACE config file for AIX 3.2.5 courtesy of Bob Olson . In addition, I changed the name of the config-aix.h file to be config-aix-4.1.x.h and called the new file config-aix-3.2.5.h. * apps/Synch-Benchmarks/Benchmark.cpp: Added template specializations for ACE_TSS and ACE_Atomic_Op. Thanks to Matt Stevens for pointing out the need for this. * ace/CORBA_Handler: Added a number of fixes to get the ACE CORBA_Handler to compile on Win32 with Orbix 2.0. Thanks to Rich Ryan for these fixes. * ace/OS.cpp (thr_create): Fixed a small bug in the pthreads thr_create() code by dereferencing p_thr before casting it to hthread_t and assigning it to *thr_handle. * ace/OS.i: Backed out the previous changes of pthread_cleanup*. It turns out that on SunOS 5.5 these macros force a certain style of programming this is hard to integrate with the existing implementations of other ACE wrapper methods. Fortunately, none of this affects existing ACE code since we just use the default solaris threads API in this case... * ace/OS.i: Modified the order of #ifdefs in various ACE_OS::thr_* methods to take advantage of the pthreads features on Solaris 2.5. * ace: Removed the last few typos that prevented ACE from building on SunOS 4.x with G++... * ace/Thread_Manager: Changed spawn() and spawn_n() so that they optionally take a group parameter and automatically add this to an existing thread group. That way we don't need to spawn the thread(s) and then reassign them after the fact. The new API automatically assigns the thread(s) to the group you specify. Thanks to Chris Lahey for this idea. * ace: Fully integrated POSIX pthreads into the ACE build for Solaris 2.5. * netsvcs/lib/Server_Logging_Handler.cpp: Rearranged the Server Logging files so that things will compile and link correctly using stock GNU G++. * ace/Log_Msg.cpp (ACE_Log_Msg): Forgot to initialize the ostream * to 0. This was causing problems for Win32... * ace: *Finally* got the static and dynamic ACE libraries to compile on SunOS 4.x using stock G++! * ace/Synch_T.h: Added a new macro called ACE_SYNCH that will be ACE_NULL_SYNCH if !ACE_HAS_THREADS and ACE_MT_SYNCH if ACE_HAS_THREADS. * ace/Service_Record.cpp: Changed this to use ACE_SYNCH. This will fix template problems with G++ on SunOS 4.x. * build/SunOS5.5/ace/OS.i (sema_wait): Changed all uses of ::pthread_cleanup_{push,pop} to pthread_cleanup_{push,pop} since these are implemented as macros on Solaris.... Thu Apr 11 19:43:33 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * examples/Naming/Dump_Restore/Dump_Restore.cpp (init): Modified Dump_Restore in order to work correctly without having Name_Options around as a Singleton. * examples/Naming/Client/Client_Test.cpp (set_host): Modified Client_Test in order to work correctly without having Name_Options around as a Singleton. It simply uses the accessor provided by Naming_Context to get to Name_Options. * ace/Name_Options.cpp: Changed Name_Options so that it is no longer a Singleton. As a result, now there is an instance of Name_Options per Naming_Context. Note that for an application to change Name_Options, it can use the accessor function provided in Naming_Context. As a consequence of all these changes, we can now have multiple Naming_Contexts per application. Wed Apr 10 20:19:50 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.h: Incorporated a mapping for Solaris 2.5 pthreads so that it fits in nicely with the earlier ACE pthreads support. From now on, the pthreads API will be the default threading scheme for ACE on Solaris 2.5... * ace: Added the file name to every *.cpp and *.i file in ACE. This makes it easier to figure out what's what when looking at the source code. Thanks to Chris Lahey for suggesting this. * ace/CORBA_Handler.cpp (ACE_CORBA_Handler): Added bodies for the copy constructor and assignment operator of ACE_CORBA_Handler (which are private) since some compilers generate link errors for these! Thanks to Rich Ryan for pointing this out. * ace/Synch_T.cpp (ts_get): at line 270 (within ACE_TSS::ts_get (void) const) there was a keycreate() call that did not have an instance pointer pass in with it. The symptom has been that if the last active task has been finished all active tasks being created afterwards won't get the close() hook invoked. I've fixed this in order to solve a mysterious bug with ACE_TSS. Thanks to the ever-vigilant Detlef for reporting this bug and fix. Wed Apr 10 01:56:52 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace: Added the ACE_TEMPLATES_REQUIRE_SPECIALIZATION to config-sunos5.5-g++.h file. This should allow ACE to build correctly. Thanks to Adam Miller for pointing this out. * ace: Installed SunOS 5.5, so now we can finally test pthreads! * ace/OS: Modified the implementation of mutex_t for Win32 so that it automatically selects the CRITICAL_SECTION or the HANDLE form of Mutex depending on whether the type argument is USYNC_THREAD or USYNC_PROCESS, respectively. This now means that all the existing ACE code that used ACE_Condition will continue to work correctly on Win32 and UNIX! Tue Apr 9 23:04:30 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace/Synch: Somehow, the definitions for ACE_Thread_Semaphore and ACE_Process_Semaphore were MIA. I've added the implementations. Thanks to Bruce Worden for noticing this... Tue Apr 9 02:16:02 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Service_Config.cpp: reordered the #includes in this file so that it works correctly on Win32. Thanks to Rich Ryan for help with this. * ace/Auto_Ptr.cpp (remove): added the implementations of the auto_ptr::remove() and auto_array_ptr::remove() static methods. Thanks to Chris Lahey for noticing their absence... * ace/Auto_Ptr: Added the #pragma implementation ("Auto_Ptr.cpp") statement if defined ACE_TEMPLATES_REQUIRE_PRAGMA to work on AIX. Thanks to Chris Lahey for this. * ace/Makefile: Moved Auto_Ptr from FILES to TEMPLATE_FILES to work on AIX. Thanks to Chris Lahey for pointing this out. * Modified a bunch of the apps and example Makefiles so that things will build better using G++. * ace/Name_Options.cpp (parse_args): Made the "database" name the same as the process name by default... * ace/Reactor.h: Changed the type of current_ in ACE_Handler_Repository_Iterator from size_t to ssize_t so that this can handle negative numbers. Thanks to Mark Zusman for reporting this. Mon Apr 8 23:33:15 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * netsvcs/lib/Name_Handler.cpp (recv_request): Fixed a small bug. In recv_request() ntohl was being called again on length which was causing problems since the length was already in host byte order. The ACE Name Server should now be working on NT. Mon Apr 8 02:14:30 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Fixed a number of minor problems that caused duplicate symbols when linking ACE with G++ (no-repo) on SunOS 5.x and SunOS 4.x. Thanks to Andy Gokhale for help with this. * ace: Added Jan Rychter very nice contribution of pthreads and DCE threads to ACE. This should make life easier for Linux users and users of other pthreads packages. * ace/Service_Config.cpp (process_directives): Changed the code to use an auto_ptr to make sure that we always release the ace_obstack memory, regardless of how we exit. * ace/OS.i (strerror): Changed sys_errlist to _sys_errlist to work around bugs with SunOS 4.x. This will undoubted break some other twisted UNIX system. * ace/Local_Tokens.cpp (acquire): Added a return 0 at the very end... Thanks to Chris Lahey for reporting this (yet again... ;-)). * ace/Parse_Node: Made a bunch of small changes to deal with the fact that there's no dlerror() equivalent on NT... * ace/Task.h: The static function instance() should return ACE_Task_Exit* (or ), not ACE_Task_Exit *. Thanks to Chris Lahey for spotting this. * ace/Thread_Manager.cpp: added an argument to the exit() method that allows applications to exit without calling thr_exit. This would allow the method to be called from within a pthread cancellation cleanup routine, without calling thr_exit a 2nd time. Thanks to Chris Lahey for suggesting this. * ace/Log_Record.cpp (operator <<): Changed the verbose option so that it is an attribute, this will allow the ostream to print verbose information if necessary. Thanks to Chris Lahey for suggesting this change. * ace/config-win32-msvc*.h: Added the ACE_LACKS_STATIC_DATA_MEMBER_TEMPLATES flag since it appears that this causes problem with some versions of MSVC++ 4.0... * Released version 4.0.6 for testing. * netsvcs/lib/Server_Logging_Handler.h: #included the Svc_Handler.h file, that absence of which was causing problems for HP/UX. Thanks to Richard Orr for reporting this. * Made a few minor changes to the source to make sure that it compiles correctly on SunOS 5.x with G++. * ace/Thread_Manager.h: Added the insert() method for the non-MT version of Thread_Manager. Thanks to Alexandre Karev for reporting this. * ace/Task: Renamed the static double-check lock_ to ace_task_lock_ to avoid a name conflict with the existing this->lock_ instance in each class. Thanks to Prashant for tracking this down... Sun Apr 7 14:40:05 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * examples/Service_Configurator/IPC-tests/server/Handle_Thr_Stream.cpp: Rolled back a change that replaced #defines with typedefs. It turned out the #defines were there for a reason... Thanks Neil B. Cohen for pointing this out... * ace/OS.h: Added a new check to deal with the fact that many versions of Pthreads don't support tid_t (which seems to be an AIXism...). Thu Apr 4 01:19:19 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Officially released version 4.0.5... * ace/OS.h (ACE_TSS_TYPE): Somehow the ACE_TSS_TYPE macros ended up *before* the inclusion of Time_Value.h, which meant that these macros weren't being expanded properly. This may help explain odd behavior with ACE TSS. * ace/Thread_Manager.cpp (insert): Added a new method that allows us to insert a thread into an ACE_Thread_Control after it's been constructed. * ace/Message_Block: Fixed a typo in the init() method. Thanks to Ross Dargahi for noticing this. * ChangeLog: Finally got all of ACE built on SunOS 5.x and SunOS 4.x using only the stock GNU GCC compiler (i.e., no template repository...). This should make it easier to port to other versions of G++. * ace/Task: Modified the ACE_Task and ACE_Task_Exit implemenation so that it only uses 1 TSS key per ACE_Task template instantation, rather than 1 TSS key per ACE_Task instance. This works around horrible limitations with Win32... * ace/Thread_Manager: Added new methods to set and get the Thread_Manager used in a Thread_Control. * ace/Pipe.cpp (open): Fixed a bug in where ACE_Pipe::open did not set this->handles_, thus a garbage handle gets registered. Inserted the following two lines at line 53 of Pipe.cpp: this->handles_[0] = reader.get_handle (); this->handles_[1] = writer.get_handle (); Thanks to Kirk Sinnard for this fix. * ace/OS.h: Added a couple of fixes for the SCO port. Thanks again to Matt Newhook * ace/OS.cpp: Integrated Detlef's clever scheme for freeing up thread-specific storage correctly on Win32... * ace/Task.cpp (ACE_Task): Made sure to initialize the next_-pointer of ACE_TASK and Message_Queue point to NULL. This fixes a problem on NT. Thanks to Karlheinz for noticing this... * include/makeinclude/rules.lib.GNU (VLIB): Fixed up the ACE makefiles so that we can now build on SunOS 4.x correctly without using the template repositories... * ace/Service_Config.cpp: Added a bunch of template specializations so that GNU G++ can be used to compile ACE on SunOS 4.x *without* requiring the template repository hacks... Thanks to Mark Zusman for helping with this. Wed Apr 3 00:55:12 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * examples/Reactor/Misc/test_reactors.cpp: Added a new torture test of the ACE_Reactor and ACE_Task. Thanks to Detlef for contributing this... * netsvcs/bin/Makefile: Removed a stray -lACE_svcs from the Makefile. Thanks to Matt Stevens for reporting this. * ace/Synch.cpp: Fixed two mistakes related to keeping INLINE on the get_thread_id() and get_nesting_level() when I moved them into the *.cpp file... Thanks to Matt Stevens for finding this. * ace/Reactor.cpp (owner): modified owner() so that it returns the original owner when setting a new owner. This makes it possible to write code like this: thread_t t; reactor->owner (ACE_Thread::self (), &t); reactor->handle_events (); reactor->owner (t); * ace/SOCK_Connector.cpp (connect): Added an additional check for errno == EWOULDBLOCK for non-blocking connects due to screwy semantics of Win32 non-blocking sockets... * netsvcs/lib/Client_Logging_Handler: Fixed a very obscure bug that arose due to the way that UNIX select() interacts with SVR4 MSG_BAND data. It turns out that you must use the ACE_Event_Handler::EXCEPT_MASK to get this to work properly (gag). This stuff is much easier with SVR4 poll(). Tue Apr 2 13:57:05 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i: Fixed an error compiling the new Log_Msg.cpp because when it inlined OS.i, there were references to ACE_OS::thread_mutex_*() before the functions were declared inline. The references were assumed to be external, and then when the linkage was actually declared as inline, the compiler choked. To fix this, in OS.i, all the thread_mutex_* calls where moved above the first reference, which was in ACE_OS::cond_broadcast(). Also moved the mutex*() calls since they are referenced by the thread_mutex calls. So the order is mutex*(), then thread_mutex*(), and then cond*(). Thanks to Chris Lahey for reporting this. * netsvcs/lib/Client_Logging_Handler.cpp: Added missing return values in some of the methods. Thanks to Tim Harrison for spotting this... * ace/Map_Manager.cpp (shared_unbind): It was possible that INT_ID could be assigned a value from a previously unbound map entry. The shared_unbind matches the ext_id, but does not see if the entry had previously been unbound. This causes a problem if the INT_ID type is a pointer type, and the user of the map deletes objects as a result of the unbind returning a 0. The correct solution is to include a test for is_free in the shared_unbind before checking the equality of the ext_id. Thanks to Phil Mesnier for reporting this. * ace/Reactor.cpp (next): Fixed a bug that prevented the iterator's next() method from detecting the end of the iteration. Thanks to Mark Zusman for reporting this. * ace/Reactor.cpp (advance): Fixed a bug that prevented the iterator from advancing. Thanks to Mark Zusman for reporting this. * ace/Log_Msg.cpp (open): Added a check for logger_key being NULL, in which case we log to STDERR rather than segfault... * netsvcs/lib/Server_Logging_Handler.h: Fixed a typo in the declaration of ACE_SVC_FACTORY_DECLARE (ACE_Server_Logging_Acceptor) Thanks to Neil Cohen for finding this. Mon Apr 1 00:17:21 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace/Time_Value.i: Fixed two stupid bugs in the Time_Value relational operators. Thanks to Mathew Newhook for pointing this out. * ace/OS.h: Added an extern "C" block around the netdb, net/if.h, netinet/in.h, arpa/inet.h to work around problems with SCO. Thanks to Mathew Newhook for this. * ace/Reactor: Merged the ACE_Pipe into the Reactor implementation. This cleans up some nasty OS-specific code in a clean way. * ace/CORBA_Handler.cpp (ACE_MT_CORBA_Handler): Fixed some typos that arose during the transition to the ACE_Pipe. * ace/Pipe: Added a new open() method that doesn't return the handles, it just stashes them away for safe keeping. This is useful for places like the Reactor. * ace/Local_Name_Space.h: Added ACE_Export to the front of ACE_NS_String. Thanks to Detlef for suggesting this. * From now on, I'll be numbering each new release of ACE with a different minor number. However, the latest version of ACE will also always be available at http://www.cs.wustl.edu/~schmidt/ACE.tar.gz This is useful if you just want to get the latest one without having to bother with keeping track of minor numbers. For instance, that way you can still keep that reference in my bookmark list and you don't have to remodify it with every release. Thanks to the ever-vigilant Chris Lahey for suggesting this. * ace: added new a ACE project file (ace.mdp) and Makefile (ace.mak) to create ACE as a DLL on Win32. This should greatly simplify the Win32 build process... * INSTALL: Added new INSTALL file explaining how to build for Win32. * ace/Thread_Manager.cpp: Fixed a horrible bug with Win32. On reasonable systems ACE_Thread::exit() should not return. However, due to horrible semantics with Win32 thread-specific storage this call can return (don't ask...). Therefore, we need to reacquire the mutex so that we don't get burned when the Guard automatically releases it when this method returns. Thanks to Tim for helping me figure this out. * ace: Fixed some problems with errno in Remote_Tokens and Local_Tokens. * ace/Reactor.cpp: Fixed an annoying preponderance of useless ACE_MT_SAFE #ifdefs in the Reactor implementation. Thanks to Gerolf Wendland for pointing this out! Sun Mar 31 13:09:27 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace/OS: Changed the implementation of cond_t and rwlock_t for Win32 to use thread_mutex_t (i.e., CRITICAL_SECTIONS) rather than mutex_t (i.e., Win32 Mutexes). This should improve performance without sacrificing generality (of which there is none at the moment since we don't have process-wide condition variables or readers/writer locks on Win32 anyway.. * ace/Thread_Manager.cpp (remove_thr): Fixed bugs in the ACE_Thread_Manager class. The wait() member grabbed the Thread_Mutex lock_, then (if the current_count is not zero) grabs the Condition lock zero_cond_. Doing so, however implicitly released lock_, which meant that another thread could get into wait(), and wait on zero_cond_. zero_cond_ was only signaled from ACE_Thread_Manager::remove_thr() when the current_count_ is zero, but it signaled with ACE_Condition_Thread_Mutex::signal(), which only releases one thread that is waiting on the Mutex. Thus, any other threads waiting on zero_cond_ would never be resumed. The fix was to use ACE_Condition_Thread_Mutex::broadcast() rather than ACE_Condition_Thread_Mutex::signal() in ACE_Thread_Manager::remove_thr(). This fix is only reliable since remove_thr() is called only when ACE_Thread_Manager::lock_ is held by the calling thread. Thank to Bruce Worden for reporting problem and suggesting this fix. * ace/Token: Modified this class so that it only works with ACE_Thread_Mutex (which is more precise than what was going on before...). * ace/Synch.h: Modified ACE_RW_Process_Mutex so that on Win32 it uses ACE_Process_Mutex until we've got a working RW_Process_Mutex for Win32... * ace/Synch: Removed the ACE_Process_Barrier until we get a working implementation... * ace/Synch: Changed the ACE_Barrier so that it only tries to work within a single process. * ace: Changed all uses of ACE_Condition_Mutex to ACE_Condition_Thread_Mutex to reflect what's really going on here... * Changed all uses of ACE_Mutex to ACE_Thread_Mutex throughout ACE. This is *much* more meaningful and makes it possible to do some great optimizations on Win32! * ace: Changed all uses of ACE_Mutex_Guard to ACE_Thread_Mutex_Guard. This is a more accurate name for how this is used in ACE. * netsvcs/lib/Client_Logging_Server: Reengineered the ACE Client_Logging service so that fits into the ACE network service format. This version is particularly interesting since it illustrates a "Connector-driven" service. In contrast, all the other ACE network services are "Acceptor-driven" services. * netsvcs: Merged the implementation of all the ACE network services into a single ./netsvcs/lib directory. Each of these is now a fully dynamically linkable service. Created a single main.cpp program in the ./netsvcs/bin directory. This main illustrates how to dynamically link an or all of the ACE network services to form complete applications. Thanks to Prashant for help with this. * ace/OS.cpp: Changed readv() and writev() so that they will once again compile for Win32. Note, however, that if you try to readv() and writev() on the same descriptor you will lose since they are not atomic! However, this stuff is stuff useful if you're *not* reading/writing to a common descriptor... * ace/Synch_T.cpp: Removed the ACE_Null_Condition<> template. This didn't really make any sense... * ace/OS: Changed the test of lock_.get_nesting_level() in ACE_TSS_Cleanup::exit () to account for the fact that a ACE_Recursive_Thread_Mutex now starts out with a nesting level of 1 when it is first acquired. * ace/Synch_T: Removed the implementation of ACE_Recursive_Lock. It just doesn't generalize correctly to other types of synchronization mechanisms... * ace/Synch: Reimplemented ACE_Recursive_Thread_Mutex using Dave Butenhof's strategy. This fixes some latent race conditions in the original implementation. * ace: Changed the name of ACE_Recursive_Mutex to ACE_Recursive_Thread_Mutex since the current implementation really only works for Thread_Mutexes or Thread_RW_Mutexes... * ace: Removed the "Assert.h" file and moved its functionality into Log_Msg.h. This is more consistent with the rest of the error reporting and logging in ACE and also avoids some nasty circular include problems. * ace/Local_Tokens.cpp: Changed the use of ACE_RETURN so that it passes in errno. * ace/Log_Msg: Changed the arguments to the ACE_RETURN macro so that errno can be returned explicitly. Sun Mar 31 15:52:58 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (create_manager): Added three new methods -- list_name_entries, list_name_values, and list_name_types. They work similar to list_names, list_values, and list_types (respectively) except they return the entire tuple associated with a name binding. Sat Mar 30 16:46:32 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/SPIPE_Stream.i (send_handle): There was a typo -- the parameter shouldn't be "const ACE_HANDLE handle", it should just be ACE_HANDLE handle. Sat Mar 30 16:09:25 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * INSTALL: Added new Win32 installation instructions to build ACE as a DLL. Sat Mar 30 14:42:02 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Carefully tested ACE on SGI and SunOS 5.x using G++ and CC. Everything seems to work now... Released version 4.0.2. * ace/Acceptor.cpp (ACE_Strategy_Acceptor): Added default initializations of NULL for service_name_ and service_description_ in the ACE_Strategy_Acceptor's default constructor. * ace: Fixed a slew of warnings for HP/UX. Thanks to John Morey for reporting these. * ace/FILE.h (ACE_FILE_Info): Changed the type of field size_ from size_t to off_t to avoid warnings from C++ compilers... * ace/OS.cpp (ftruncate): Added a new function for platforms like SCO that lack ftruncate(). Thanks to Matthew Newhook for contributing this. Sat Mar 30 12:41:16 1996 Douglas C. Schmidt * ace: Removed the ACE_svcs library for the time being and merged everything back into libACE. There were two reasons for this change: 1. There is a subtle dependency between the ACE_Reactor and the ACE_Local_Tokens* stuff when deadlock detection is enabled. This was causing problems. 2. This was driving GNU G++ nuts because of the need to do the "prelink." Sooo, for the time being, there's just one ACE library. If anyone wants to take the time to split everything up so that it works for all different platforms and send me the fixes I'll be glad to include this in ACE. * ace/Thread.i: Made the ACE_Thread::thr_self() methods call down to the ACE_OS::thr_self() methods for the case where the platform doesn't support threads. This means that we only have to set the default thread id in one place... Sat Mar 30 11:53:31 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * bin/class2info.awk: Modified the handleClass() AWK function so that it can correctly deal with the new ACE_Export and ACE_Svc_Export tags needed to build ACE DLLs... Here's the trick: sub( "^[\t ]*class[\t ]+(ACE_[.]*Export[\t ]+)?", "", $0 ) completely intuitive, eh? ;-) * apps/Gateway: Reimplemented the Gateway prototype so that it uses the new Reactor cancel_wakeup() and schedule_wakeup() methods instead of mask_ops(). * ace/Reactor: Added new high-level "scheduling" operations called schedule_wakeup() and cancel_wakeup(). These methods utilize the lower-level Reactor mask_ops() methods to schedule and cancel Event_Handlers for subsequent dispatching by the Reactor. I added these new interfaces because they are more intuitive than calling mask_ops(), which is a non-descriptive name unless you understand the Reactor implementation... * ace/config-irix5.*.h: It turns out that on IRIX 5.x the bi-directional SVR4 pipe semantics are *disabled* by default. Therefore, I've changed the config-irix4.*.h files to comment out ACE_HAS_STREAM_PIPES. Fri Mar 29 08:25:33 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/ACE.cpp (ldfind): Changed the character given to strtok() to account for the differences in UNIX and Win32 described in the follow bullet. * ace/OS.h: UNIX and Win32 use different characters to separate pathnames in their runtime LD search path (':' and ';' respectively). Therefore, I've added new macros to ACE OS to account for this stuff portably. * ace/OS.i: Changed the behavior of the ACE_OS::thr_self() methods so that they always return 1 if ACE_HAS_THREADS is *not* enabled. This makes application code work correctly across threaded and non-threaded platforms. * ace/OS.i: Changed the return value of the Win32 arm of the ACE_OS::dlopen() and ACE_OS::dlsym() functions so that they are now identical to the way that UNIX behaves. * ace/Svc_Conf.y: added a #define for ACE_BUILD_DLL to make Win32 happy... * ace/Synch.i: Fixed a number of bugs in the ACE_Mutex and ACE_Thread_Mutex tryacquire_read() and tryacquire_write() methods. I was calling mutex_lock() rather than mutex_trylock()! Thanks to Prashant for finding this. * ace/OS.cpp (exit): Fixed a bug caused by the fact that the test for the nesting level of the recursive mutex should have been for > 0 rather than > 1. Thanks to Kirk Sinnard <1764@mn.lawson.lawson.com> for tracking this down! * ace/Log_Msg: Enhanced ACE_Log_Msg so that it automatically caches the process id. This makes it more robust even if users don't call open()! * examples/Connection/non_blocking/CPP-connector.cpp Fixed a bunch of typos that caused templates to fail on HP/UX. Thanks to Neil Cohen for spotting this. * ace/Shared_Memory_SV: Change all uses of "int id" to "key_t id" to reflect the new use of key_t in the Memory_Pool. * ace/Memory_Pool.cpp (ACE_MMAP_Memory_Pool): Fixed a potential but in the construction of an MMAP memory pool. Because NT uses strings rather than integers to name its semaphores we were passing in junk to the NT Process_Mutex initializer. Arrgh! I fixed this problem by typedef'ing key_t to be char * on Win32 and then updating the ACE_DEFAULT_SEM_KEY to be "C:\\temp\ace.sem" rather than 1234.... Also added a new macro called ACE_INVALID_SEM_KEY, which is NULL on NT and -1 on UNIX. * ace/Log_Msg.cpp (instance): Fixed a typo for HP/UX and other non-threaded platforms! * ace/Synch_T.cpp (ACE_TSS_Guard): Added another small fix to ACE_TSS_Guard to make sure we free up the key when we go away. Thu Mar 28 15:28:44 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.h: changed the names of macros ACE_DYNAMIC_SVC* to new names that reflect their new, more general nature now that we've got the ACE_Static_Svcs stuff... * ace/Service_Config.cpp (load_defaults): Added a very nice mechanism that generalizes the configuration of statically linked services. This makes it possible to insert statically linked services into a program without changing any of the existing ACE_Service_Config code. These static services can then be enabled via the svc.conf file using the "static" directive. Thanks to Matthew Newhook for suggesting this and for doing the bulk of the implementation. * ace: Commented out the ACE_TRACE calls in ACE_Unbounded_Set, ACE_OS, and all the ACE synchronization wrappers to work around some horrible dependencies with Win32 and thread-specific storage... The moral here is that you have to be careful what you trace because circualar dependencies abound... My new policy is that low-level routines are not automatically traced because they can cause infinite recursion. * ace/Synch_T.cpp: Fixed a nasty bug with ACE_TSS_Guard. This was causing programs that uses ACE_TSS_Guard to segfault... Thanks to Kirk Sinnard for reporting this. * ace/Synch: Added "lock()" methods to all the ACE synchronization wrappers to return the underlying C level object... This is necessary at some points in ACE. * ace/Synch: Fixed a horrible bug in ACE_Recursive_Mutex and ACE_Recursive_Lock that was causing instant deadlock because the non-recursive Solaris/Pthread mutex was being acquired too eagerly... Thu Mar 28 12:14:57 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Map_Manager.cpp (rebind): Added a new method called resize() which increases the size of the map. Both open() and shared_bind() make use of this method to dynamically increase the size of the map. * ace/Local_Name_Space.cpp (shared_bind): Modified shared_bind() in accordance to the changes to Map_Manager. shared_bind() no longer has to check if the map runs out of room since Map_Manager dynamically grows the map. * ace/Map_Manager.cpp (shared_bind): Changed shared_bind() so that when the map reaches max_size_, we grow the map by DEFAULT_SIZE by making a call to open(). Wed Mar 27 20:00:47 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (shared_bind): Fixed a bug which was allocating memory for a name binding which would fail on a bind since there already existed a binding. This was causing the mapped-file to grow even when binds would fail. Tue Mar 26 13:49:24 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/INET_Addr.i (addr_to_string): Moved the order of get_host_addr() so that it comes before addr_to_string(). Thanks to Neil Cohen for finding this. * ace/Synch: Added a new implementation of ACE_Thread_Mutex that takes advantage of the new ACE_OS::thread_mutex_* mechanisms. * ace/OS: Added a new set of ACE_OS::thread_mutex_* mechanisms that map efficiently to either Win32 CRITICAL_SECTIONs or UNIX mutex_t's with type set to USYNC_THREAD. This allows ACE to take advantage of the lightweight Win32 synchronization mechanisms. When Windows NT 4.0 comes out, we'll at last have a perfect mapping since WinNT 4.0 supports TryEnterCriticalSection! Tue Mar 26 17:35:31 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Map_Manager.cpp: Added two new methods -- total_size() and current_size() which return the max_size and current_size of the map respectively. Mon Mar 25 20:22:25 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Service_Config.cpp (initialize): Updated the use of ACE_ARGV to use the new method names. * ace/ARGV.cpp (ACE_ARGV): Changed the name of the two main methods of ACE_ARGC from count() and operator&() to argc() and argv(). This seems much more consistent with UNIX and C/C++ naming conventions. * ace/ARGV.cpp (ACE_ARGV): Added environment variable substitution to the ACE_ARGV class using the new ACE::strenvdup() method. This is transparently available to the ACE Service Configurator now, which enables the svc.conf file to include things like this: dynamic Name_Service Service_Object * name_svc.so:make_Name_Service() "-p $PORT -h $HOST" * ace/ACE: Added a new function called strenvdup() that returns a dynamically allocated duplicate of its argument, substituting the environment variable if str[0] == '$'. * ace/Naming_Context: Modified this class so that it now inherits from ACE_Service_Config. This enables application programmers to dynamically link instances of ACE_Naming_Context into their applications via the svc.conf file. * ace/OS.i (access): Added a new wrapper for the access() method. * ace/Local_Name_Space.cpp: Fixed an error in each list_xxx methods. If there's an error in the for-loop we must still call ACE_OS::free() to release resources. Thanks to the ever-vigilant Karlheinz Dorn for spotting this. * ace/Strategies: added virtual methods with name dump() to ACE_Scheduling_Strategy, ACE_Schedule_All_Reactive_Strategy, and ACE_Schedule_All_Threaded_Strategy. These methods are referenced in ACE_Strategy_Acceptor::dump() method (Acceptor.cpp). Thanks to Alexandre Karev for reporting this. * include/makeinclude: Added a number of changes to allow ACE to build shared libraries correctly on AIX. Thanks to Chris Lahey for these fixes. * ace/Makefile: added tempinc to the list of directories to delete for "clean". Also, between the call to the two sub make files, added a $(RM) -rf tempinc. Since we make both libs from the same directory, we need to delete the tempinc directory to be sure we don't get the templates generated for libACE included in libACE_svcs. * ace/OS.i: Improved the mapping between ACE Win32 GetLastError() values and POSIX-like errnos. This is now centralized in the ACE_FAIL_RETURN macro. * examples/Threads/test_thread_specific.cpp: Added many new tests to exercise the ACE thread-specific storage mechanisms on Win32 and UNIX. Mon Mar 25 4:00:01 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Options.cpp (parse_args): Added a new data variable called database_ that keeps the name of the name server database. Also added accessor methods for this. * ace/Local_Name_Space.cpp (create_manager): Changed create_manager() to use ACE_Name_Options::instance ()->database () instead of ACE_Name_Options::instance ()->process_name(). * ace/ACE.cpp (ldopen): Added a new method called ldopen that finds a file either using absolute path or using LD_LIBRARY_PATH. If the file is found, it opens the file and returns a pointer to the file. * ace/Parse_Node.cpp: used the new ACE::ldfind() method to locate the shared object file. This makes it possible to put "relative" names into svc.conf files and then allow ACE to locate the appropriate shared object DLL. * ace/ACE: added new methods called ldopen() and ldfind() that find the file either using absolute path or using LD_LIBRARY_PATH. If found, ldopen() opens the file and returns a pointer to the file. Sun Mar 24 10:41:12 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Compiled ACE successfully on Windows '95. * ace/OS: Totally redid the ACE thread-specific storage cleanup mechanism in OS.cpp to avoid bugs with the previous implementation. * ace/Synch_T: Modified ACE_TSS_Guard so that it will call ACE_OS::thr_keyfree () when it's finished. This should help to eliminate the unbounded use of thread-specific storage slots within a thread. * ace/Thread: added a new method keyfree() that calls down to the ACE_OS::thr_keyfree (). * ace/Thread_Manager.cpp: Modified the semantics of ACE_Thread_Manager::exit () so that it "always" calls ACE_Thread::exit () even if the user hasn't registered this thread with the thread manager. This avoids some subtle problems that arise typically when I forget to put an ACE_Thread_Control around the main() function. * examples/Threads/test_thread_specific.cpp: Added new tests to ensure that the Win32 TSS cleanup logic is working correctly! * ace/Log_Msg: Moved the definition of the ACE_FIFO_Send_Msg queue from the Log_Msg.h file to the Log_Msg.cpp file. The MSVC++ compiler was having problems with this when building a DLL. Besides, this needs to be replaced with Sockets for Win32 anyway... * ace/OS.cpp: Put a safe guard in the readv() and writev() emulations so that they return ACE_NOTSUP_RETURN if ACE_HAS_THREADS is enabled. This is necessary because the readv() and writev() emulations are *not* atomic if multiple threads are used. If someone can provide an atomic implementation I'll enhance the code to remove the existing restrictions. * ace/OS: Created a new macro called ACE_SYSCALL_FAILED, which is mapped to 0xFFFFFFFF on Win32 and -1 on UNIX. * ace/Synch and Synch_T: cleaned up the implementation of the Recursive_Lock and Recursive_Mutex to use the ACE_Guard classes. This simplifies the code. Sat Mar 23 16:53:14 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace: Updated all ACE classes to use the new ACE_Export macro. This will enable them to transparently be used to create a Win32 DLL. Also added ACE_BUILD_DLL to all the *.cpp files. * tests/CPP-inserver.cpp (main): Modified CPP-inclient.cpp and CPP-inserver.cpp to use non-blocking I/O. In particular, CPP-inserver times out on select and then checks to see if any connections have arrived. Both applications are also working on NT. Sat Mar 23 15:02:47 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Local_Tokens.cpp (proxy_): Modified the print statement for local tokens to account for the fact that thread ids can be unsigned on some platforms. * ace/Log_Msg.cpp (log): Modified the output format of the %t directive to ACE_Log_Msg::log() so that it prints values in unsigned form rather than signed form. This fixes a bug with Win32 on Windows '95. * ace/ACE: Added two new methods ACE::read_n() and ACE::write_n(). These are now necessary since Win32 distinguishes between operations on SOCKETs and operations on other forms of HANDLEs. * ace/ACE: Changed all uses of ACE::set_fl() and ACE::clr_fl() to ACE::set_flags() and ACE::clr_flags(). * ace/ACE: Changed the name of ACE function handle_timed_connect() to handle_timed_open() since this is reall what it does... Fri Mar 22 00:11:19 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Stack: Added a "Node of last resort" to the ACE_Unbounded_Stack. This ensures we are graceful in the face of memory failures. * ace/OS.i (sleep): Fixed a bug with the Win32 ACE_OS::sleep mapping. This should have had an "ACE_OSCALL_RETURN" in it... * ace/ACE.h: Had forgotten to put "static" in front of "basename". Thanks to Neil Cohen for reporting this... * ace/Name_Options.cpp (process_name): Changed char * to const char * to avoid compile error. Thanks to Neil Cohen for this... * ace/IPC_SAP: Implemented the enable()/disable() methods to work with non-blocking I/O for SOCKETs on Windows NT. * ace/OS.i: Modified ioctl() to support the ioctlsocket() call on Windows NT. * ace/Log_Msg.cpp (log): Fixed a bug that was caused by not NUL-terminating the logging string properly when a %a is given. Thanks to Tim Harrison for finding this. * ace/OS: Added a new overloaded "ACE_OS::sleep" method that takes an ACE_Time_Value rather than a u_int. This is much more useful for fine-grained timers than the horrible UNIX sleep() call. * ace/OS.h: #included "sys\timeb.h" in the Win32 version of ACE. Thu Mar 21 22:18:50 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * apps: Fixed all incorrect usages of ACE_HANDLE = -1 to use ACE_HANDLE = ACE_INVALID_HANDLE in order to work with Win32... * ace/OS.i (gettimeofday): Replaced the existing implementation of gettimeofday with a new version that is shorter, more efficient, and most importantly, correct... Thanks to Todd Montgomery. and Mike Flinn for this stuff. * ace/ACE: added a new method called basename() that returns the filename portion of a pathname. Thu Mar 21 21:51:48 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/OS.h (ACE_DEFAULT_GLOBALNAME): Added some new #defines for NT: ACE_DEFAULT_NAMESPACE_DIR, ACE_DEFAULT_LOCALNAME, and ACE_DEFAULT_GLOBALNAME. * ace/Local_Name_Space.cpp (ACE_Local_Name_Space): Fixed a subtle bug. Instead of doing a new char [xxx], I was doing a new char (xxx). Also, added some #if defined (ACE_WIN32) to support special cases for NT (for example determining the context file name). Wed Mar 20 02:03:39 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Log_Msg.cpp (log): If we're writing to stdout or to an iostream make sure the lock is help to avoid race conditions and scrambled output. Tue Mar 19 00:12:25 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Thread_Manager.h: Added new default values to spawn() and spawn_n(). * ace/OS.i (sigaction): Added support for signals to the ACE NT mapping. * ace/ACE: Added 4 new methods to class ACE: send(), send(), recv(), recv(). These calls mask some differences between UNIX and Win32 IPC mechanisms. Then, reimplemented the Reactor's notification mechanism to use this. * ace/OS.i (ACE_OSCALL_RETURN): fixed zillions of warnings on HP/UX by making a small change to line 151 of the OS.i file. Thanks to Neil Cohen for reporting this. * ace/Reactor.cpp: Rewrote the Reactor::notify() mechanism to avoid using readv() and writev(). These work fine on UNIX, but don't work correctly on Windows NT due to race conditions. Basically, we need to reimplement this stuff on Win32 to avoid the problem. * ace/OS.h: Added SIGQUIT, SIG_BLOCK, SIG_UNBLOCK, and SIG_SETMASK definitions to NT. * ace/Time_Value: added a new set() method to be consistent with the ACE_Time_Value (long, long) constructor. * ace/Service_Config.cpp (open): Fixed a bug where the ACE_Service_Repository and ACE_Reactor weren't being initialized correctly if the ACE_Service_Config::ACE_Service_Config (const char *) method was called. * ace/Service_Config.cpp: Changed the constructor for Service_Config so that it doesn't barf if the svc.conf file isn't found. Mon Mar 18 00:34:45 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Reactor.h: Hid the inclusion of the Local_Tokens.h file in the Reactor so that by default there will be no dependency between libACE and libACE_svcs. This should prevent compilation problems on some platforms. * bin: Included the html-windex shell script from Vincent Pommier . * man: Included a shell script to automatically generate an acewindex.html file. This file is useful just after running man2html. Thanks to Vincent Pommier for contributing this. * man/html: I've just used the man2html programs to generate the html documentation from the man3 directory. It seems to work pretty well. * ace/Assert.h: Added a forward declaration for class ACE_Log_Msg to fix problems NuMega preprocessing. Thanks to Mike Flinn for finding this. * ace: Implemented about 1/3rd of the "dump" methods for the various ACE classes. * ace/Acceptor.cpp (info): Fixed a bug -- should use PR_AD rather than ACE_INET_Addr since this is used for SPIPEs as well... * ace/Synch_T.cpp (ACE_TSS): If ACE_Thread::keycreate () fails then we do an fprintf() and return at the moment. It doesn't do any good to do an ACE_Log_Msg operation since those all require thread-specific storage and this will just cause a recursive problem... * ace/OS.cpp: Removed a memory leak in ACE_OS::thr_destory_tss() that was caused by forgetting to delete each ACE_TSS_Key_Info object when we no longer needed it. Thanks to Mike Flinn for finding this. * ace/OS.cpp: Added a new method to class ACE_OS called thr_keyfree(). This enables us to free up a thread-specific storage key on Win32 (it isn't implemented on the UNIX platforms). * ace/ACE.cpp (timestamp): In order to get ACE_OS::timestamp to work on Win32, the wMonth, wDay, wYear structure members were added to the sprintf statement. Thanks to Mike Flinn for this fix. * examples: Changed all uses of ACE::send_n (1, ...) to ACE_OS::write (ACE_STDOUT, ...) to avoid problems with NT's hacked support for sockets and HANDLEs. Thanks to Bernd Hofner for noticing this. Sun Mar 17 00:43:14 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/CORBA_Handler.cpp (handle_input): Fixed the CORBA_Handler to use send()/recv() rather than read()/write(). * ace/Process_Manager: put the hooks in for the forthcoming ACE Process Manager, which will provide a service similar to the ACE_Thread_Manager, except that it will work for processes, not threads. * netsvcs: all the ACE network services appear to be working again... * man: Completely regenerated all the manual pages to reflect all the most recent changes. * ace/OS.h: Created a new macro called ACE_DEFAULT_REACTOR_SIZE. This is useful for two reasons (1) it centralizes this value in one place in ACE and (2) removes a horrible circular dependency between the Service_Config.h and Reactor.h... * ace/Strategies: Added new strategies to handle "scheduling" of Svc_Handlers launched by a Strategy_Acceptor. In this context, scheduling refers to "suspending" and "resuming" Svc_Handlers. There are several alternative strategies ranging from (1) suspending/resuming a single handler, (2) suspending/resuming all the handlers in a Reactor, and (3) suspending/resuming all handlers controlled by a Thread_Manager. This very powerful feature makes it simple to write ACE_Network_Services that can control all their handlers in one fell swoop... * ace/Task: Added new accessors/mutators to get/set the Thread_Manager and the Message_Queue associated with a Task. This makes life easier and more abstract in subclasses and elsewhere. * ace/Reactor: added an iterator to the ACE_Handler_Repository. This makes it possible to implement the suspend_handlers() and resume_handlers() very efficiently on both UNIX and Windows NT. * ace/Reactor: Added two new methods called suspend_handlers() and resume_handlers(). These methods suspend and resume all the active Event_Handlers in the Reactor in one fell swoop. * ace/Reactor.i (suspend_handler): Fixed a bug in the Reactor where we were not factoring out the code for suspend_handler() appropriately with respect to locking. * netsvcs/Logging/Server_Daemon/Thr_Logging_Handler: Enhanced the threaded logging service so that all active threads can be automatically suspended and resumed via the ACE_Service_Config svc.conf file. * netsvcs/Logging/Server_Daemon/Logging_Handler.cpp: Rewrote the ACE Logging service to use the new ACE_Strategy_Acceptor implementation. This greatly reduced duplicate code. There's almost nothing left in this directory save the actual service itself! * netsvcs/Tokens: Rewrote the ACE Token service to use the new ACE_Strategy_Acceptor implementation. This greatly reduced duplicate code. There's almost nothing left in this directory save the actual service itself! * netsvcs/Naming: Rewrote the ACE Naming service to use the new ACE_Strategy_Acceptor implementation. This greatly reduced duplicate code. There's almost nothing left in this directory save the actual service itself! Sat Mar 16 20:02:08 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * netsvcs: Created a new subdirectory off of $WRAPPER_ROOT and put all the ACE network services (e.g., Time, Logging, Naming, and Tokens) into that directory. These services all instantiate the new ACE_Strategy_Acceptor implementation, which greatly simplifies their behavior and code. * ace/Acceptor: Made some enhancements to the ACE_Strategy_Acceptor in order to make it more suitable for use as a generic "Service" launcher. This class now defines common behavior for all the ACE network services (e.g., Time, Logging, Naming, and Tokens). * apps/Token_Server: Made the main Token_Server application be dynamically linked if there's a valid svc.conf file. * apps/Name_Server/Name_Server: Made the Server_Test a "well-behaved" service. Previously, it block in a private event loop within Name_Server::run(), which was called by Server_Test::init() when the service was linked dynamically. This obviously doesn't work correctly since it means that we can't dynamically link any other services after this one! The new version is "event-loop" friendly since it uses the main Service_Config::run_event_loop() method. * ACE-categories: Added a new emacs "outline" file that illustrates how the filenames in $WRAPPER_ROOT/ace cluster into class categories. Thanks to Chris Eich and Alex V. Maclinovsky for helping create this. * ace: Split ACE into two libraries: libACE and libACE_svcs. libACE contains the "core" ACE components. libACE_svcs contains the client-side layered services (e.g., naming service, token service, etc.). There are two reasons for doing this: 1. It reduces the size of ACE for many common usecases (i.e., most people aren't using the naming service or the token service). Any future layered services in ACE will be placed into the libACE_svcs library rather than libACE. 2. It works around annoying compiler bugs with lame compilers like HP/UX C++. Now, if those compiles can't compile the token service (which is very template intensive) it won't affect the core ACE library components. This improves the portability of ACE. * Added -lACE_svcs to the ./apps/{Token_Server,Name_Server} and ./examples/{Tokens,Naming} Makefiles to account for the new libraries. Note that no source code changes are required... Fri Mar 15 00:03:48 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * apps/Name_Server/server/Name_Handler.cpp (recv_request): Fixed the same problem as occurred in Token_Handler (these both come from the same source...). * apps/Token_Server/Token_Handler.cpp (recv_request): There was a subtle bug in the Token_Handler that stems from my old code. The line n = this->peer_.recv ((void *) (((char *) &this->token_request_) + sizeof (ACE_UINT32)), length); This should actually be: n = this->peer_.recv ((void *) (((char *) &this->token_request_) + sizeof (ACE_UINT32)), length - sizeof (ACE_UINT32)); since length was read already. The old way tried to read 4 bytes too much. Thanks to Jesper for noticing this. * ace/Connector.cpp (connect_svc_handler): Modified the connector so that it automatically calls the close () method of the Svc_Handler when the connection fails abortively. This makes the semantics the same for synchronous and asynchronous connection invocation modes. Thanks to Irfan for insisting on this! * apps/Token_Server/Token_Acceptor.cpp (handle_input): Changed return type from int to ACE_HANDLE for NT compatibility. Thanks to Jesper for this insight. * ace/OS.h: The defines for ACE_LACKS_MODE_MASKS (OS.h) were only defined when !ACE_WIN32. This has been fixed. Thanks to Irfan for spotting this too. * ace/OS.i (mmap): made a small change to ACE_OS::mmap to fix NT portability problem (new variable added:nt_flags). Thanks to Irfan for spotting this. * ace/Local_Tokens.h: Added #include "ace/Map_Manager.h" after line 41 in Local_Tokens.h to allow ACE to compile correctly with the horrid HP/UX compiler. Thanks to Neil Cohen for figuring this out (what a trooper!). * ace/Local_Tokens.cpp (ACE_TPQ_Entry *): Replace return ACE_TSS::operator ACE_TPQ_Entry *(); with return (ACE_TPQ_Entry *) (*((ACE_TSS *) this)); to work around problems with MSVC++ 4.0 when the browse option is enabled. Thanks to Kirk Sinnard <1764@mn.lawson.lawson.com> for this fix. * ace/config-sunx86-sunc++-4.x.h: Swapped the lines // Compiler/platform supports SVR4 signal typedef. //#define ACE_HAS_SVR4_SIGNAL_T #define ACE_HAS_CONSISTENT_SIGNAL_PROTOTYPES to // Compiler/platform supports SVR4 signal typedef. #define ACE_HAS_SVR4_SIGNAL_T //#define ACE_HAS_CONSISTENT_SIGNAL_PROTOTYPES to avoid an error using ProWorks C++ 4.0.1 (w/ patches) which blow up in the Signal.x stuff with a prototyping error. Thanks to John P. Hearn for this recommendation. * ace/Shared_Memory.h: The conditional compilation wrapper was out of date, so I renamed it. Thanks to Alex V Maclinovsky for reporting this. Thu Mar 14 23:18:59 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Thread.h: There was a bug in the ACE_Thread class. I attempted to make the class uninstantiable by making the constructor private. However, the statement was: private: Thread (void); which is not the definition of the ctor! The code compiled since it thinks Thread is a member function. As a result, you could instantiate a object of ACE_Thread class, but no longer... Also added this to the !defined (ACE_HAS_THREADS) arm of the conditional compilation. Thank to Sandeep Joshi for noticing this. * ace/Log_Record.i (decode): Fixed a stupid typo where htonl (this->length_) should have been htohl (this->length_)... Thanks to Audun Tornquist for noticing this. Tue Mar 12 14:51:39 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Name_Request_Reply.cpp (decode): Added ntohl and htonl calls wherever data was being exchanged. Sat Mar 9 17:49:51 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * ace/OS.i (cond_wait): Updated the Win32 implementation to correspond precisely with the UNIX semantics where the mutex is always reacquired even when errors occur. Mon Mar 4 23:03:37 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * INSTALL: updated the win32 to describe building test applications. Mon Mar 4 16:17:05 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (resolve): Fixed a small bug. The type field in resolve was copying extraneous stuff (because of being kept around as a wchar_t). So I added a new method to the class ACE_NS_String called len() which simply returns len_. Using this I can now do a strncpy of len bytes for type (instead of a simple strcpy) and then null-terminate the string to remove the garbage. Thu Feb 29 23:41:04 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * ace/Token_Request_Reply.cpp: Fixed a byte-order bug which was showing up on intel platforms. * ace/Remote_Tokens.cpp: Fixed a race condition in ACE_TSS_Connection which was munging mt token server clients. * examples/Tokens/mutex/*: The Token Server example applications now use ACE_Thread_Manager, which should make them portable to Win32. * apps/Token_Server/Token_Server.cpp: The Token Server is now a Service_Object which can be dynamically linked. * ace/OS.i: Win32 ACE_OS:: signal methods now return 0 instead of -1. So, calls to signal code succeed, but do nothing useful. Thu Feb 29 20:38:32 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * apps/Logger/Server_Daemon/Server_Logger.cpp (handle_logging_record): There was a typo where ACE_LACKS_STATIC_DATA_MEMBER_TEMPLATES was incorrectly spelled. This caused problems on HP/UX. Thanks to Neil Cohen for finding this. Wed Feb 28 11:41:49 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Mem_Map.cpp: Fixed a nasty bug with Win32 memory mapping... It turns out we need to be very careful when remapping a previously mapped region if the MapViewOfFile we're trying to establish *grows* beyond what was originally mapped with CreateFileMapping(). The new scheme seems to work correctly with both UNIX and Win32. * ace/OS.h: apparently when using MFC library functions it is not possible to include windows.h. Instead, users must include AFX.h. Therefore, I've modified OS.h to include the following: #if !defined (__AFX_H__) #include #endif /* __AFX_H__ */ Thanks to Patty Genualdi (genualdp@agcs.com) for pointing this out. * ace/Memory_Pool.cpp: Modified init_acquire() so that it calls ACE_Mem_Map::open() rather than ACE_Mem_Map::map() the first time. This ensures that the file is mapped correctly. * ace/Mem_Map: Added a new method called open() that creates/opens a file without actually mapping it. * ace/SOCK_Dgram_Mcast.cpp (make_multicast_address): Added new support for Windows NT. * ace/Memory_Pool: reimplemented ACE_MMAP_Memory_Pool so that uses ACE_Mem_Map. This ensures that we can leverage all the work that went into making ACE_Mem_Map work on Win32. * ace/Memory_Pool.cpp (map_file): Modified the code so that we always unmap the file before trying to remap it. This avoids problems with Win32... * ace/Mem_Map: Modified the implementation of ACE_Mem_Map so that it takes advantage of the improved features in ACE_OS::mmap. Also added new a method that returns the memory-mapped addr more easily. * ace/OS: Modified the ACE_OS::mmap() method so that it is more efficient for remapping files on Win32. * ace/OS.h: renamed the type QWORD to ACE_QWORD to avoid namespace pollution. Thanks to Patty Genualdi (genualdp@agcs.com) for pointing this out. * ace/OS.h: Changed the value of the SIGPIPE emulation for Win32 to match the UNIX value in order to avoid problems. Thanks to Jesper for noticing this. * apps/Token_Server: Fixed misuse of int for ACE_HANDLE in various places. Thanks to Jesper for noticing this. * Renamed the ./tests directory to ./examples to make way for the new ACE regression tests. These tests will go into the new ./tests directory. * ace/Mem_Map.cpp: Added a destructor and a close() method that closes down the file HANDLE if we allocated it. This is useful to prevent descriptor leaks. Thanks to Irfan for suggesting this. * ace: Made a number of changes to remove unreferenced parameters from ACE methods. This eliminates zillions of warning from the HP/UX compiler. Thanks to Neil Cohen for reporting this. * ace/Local_Name_Space.cpp (resolve): revised some code to work around bugs with the HP/UX compiler... Thanks to Neil Cohen for this workaround. Tue Feb 27 21:06:09 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i: The Win32 CreateFile does not handle the return you would expect when the file exists. You would expect EEXISTS (17 in UNIX) but the GetLastError is set to 80 (ERROR_FILE_EXISTS). Therefore, I changed ACE_OS::open to map ERROR_FILE_EXISTS onto EEXIST. This is only a partial solution, however. There must be a better way to handle this! * ace/Mem_Map.cpp (ACE_Mem_Map): Fixed a bug with one of the ACE_Mem_Map destructors that was causing problems due to the fact that base_addr_ wasn't being initialized to 0. Thanks to Karlheinz for noticing this. * ace/Signal.i: signals on AIX cannot use sigprocmask() in any multithreaded programs. Instead, the call ACE_OS::thr_sigsetmask() should be used since it is used to update the thread signal mask. This comes up only in the Sig_Guard constructor and destructor. Therefore, I #ifdef the call under the ACE_MT_SAFE macro, using sigprocmask() in the #else case. Thanks to Chris Lahey for reporting this. * tests: removed all uses of ACE_OS::join (0, ....) in ACE and replaced them with ACE_Thread_Manager::wait() calls. This ensures that the ACE code is portable to Win32 and POSIX pthreads! * ace/Dynamic_Service.cpp: Added a #include of "ace/Service_Config.h" to work around problems with HP/UX. Thanks to John Morey for reporting this. Sun Feb 25 12:10:38 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * apps/Token_Server: changed the name token_server.cpp to main.cpp to avoid a name clash on Win32... * man/man3: updated all the ACE manual pages to reflect the most recent changes. * ace/OS.i: Implemented the cond_timedwait() method correctly in ACE_OS. This method should now do the right thing with respect to blocking, polling, and timewaits for Win32. * ace/OS.i: Modified the Win32 implementation of all the socket-related wrappers in ACE_OS. The new implementation very cleverly sets errno to the result of WSAGetLastError() if a socket-related call fails. Since all of the WinSock errors are mapped into their equivalent UNIX errno values this enables socket applications to work portably for UNIX and Win32. * ace/Malloc_T.cpp (ACE_Malloc): Improved error reporting for cases where the Memory_Pool::init_acquire() fails (e.g., if the backing store was already created by another user and we don't have permission to access it...). * ace/Task: modified Task.h so that it is no longer necessary to write #include "ace/Synch_T.h" in order to use ACE_MT_SYNCH, etc. * ace/SOCK.cpp (DllMain): Added a clever "Schwartz counter" to make sure that the WinSock library is correctly initialized, even if we're not using ACE as a DLL!!! * tests/Mem_Map/file-reverse: got the file reverse test program working. The ACE_Mem_Map class should now be ported to Win32... * ace/OS.i (open): Changed how the O_CREAT flag was handled to give it UNIX semantics... Sat Feb 24 12:55:27 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/UPIPE_Stream.cpp (recv): Fixed a signed/unsigned mismatch. There should be no more mismatches in ACE... * ace/Name_Proxy.cpp (recv_reply): Fixed a signed/unsigned mismatch. * ace/Message_Block.cpp (copy): Fixed a signed/unsigned mismatch. * ace/INET_Addr.cpp (get_host_name): Changed the type of the hostnamelen parameter from int to size_t to be more "abstract." This will also prevent a warning from the MSVC++ compiler. * apps/Logger/Client_Daemon/Client_Logger.i: Fixed a braino with network byteoder that was causing Client_Logger::send() to fail on INTEL boxes. Thanks to Bryon G. Rigg for finding this. Fri Feb 23 01:59:34 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i: Enhanced mmap() to give the same semantics as UNIX mmap() by using MapViewOfFileEx(). Also greatly improved the UNIX emulation capabilities so that mmap() now has the same API for UNIX and Win32. * ace/Time_Value: Added functions to constructor and conversion from/to Win32 FILETIME. This enables us to reuse those 100ns conversions in ACE_Profile_Timer. * ace/Time_Value: Fixed a problem with these: // True if tv1 < tv2. INLINE int operator < (const ACE_Time_Value &tv1, const ACE_Time_Value &tv2) { ACE_TRACE ("operator <"); return tv2 > tv1; } This should be "tv2 >= tv1". Same problem was fixed with operator <=. * ace/Profile_Timer.i, fixed the following: this->end_time_ = ACE_OS::gettimeofday (); this->last_time_ = this->end_time_; this->last_usage_ = this->end_usage_; ACE_OS::getrusage (RUSAGE_SELF, &this->end_usage_); This will overwrite end_time before it is saved to last_time... * tests/Threads: Added Jesper's test for readers/writers locks. The test adds a number of reader and writer threads. When active, writers modify a shared variable and verify that only one writer and no readers are active. When active, readers check that no writers are active and that the shared data is never changed. The test passes for different ratios of readers/writers, and debug output shows that several readers are allowed simultaneously. Or, in other words, we have an indication that it should work. * ace/OS.i: Added Win32 implementations of getpid(), fork(), lseek(), dup(), cond_timedwait (), getrusage () * ace/OS.i: Modified all the ACE_OS::str* routines so that they no longer do an ACE_OSCALL_RETURN. This can lead to weird bugs... Thanks to Jesper for noticing this. * ace/config-sunos4-sun4.1.4.h: Added a new config file that should work with SunOS 4.x... Also added support for SunOS 4.1.4. Thanks to Mick Adams (eeimas@eei.ericsson.se) for help with this. Thu Feb 22 18:58:36 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * tests/Naming/Client/Name_Service.h (ACE_Service_Object): Modified test application to make use of ACE_Dynamic_Service. * ace/Dynamic_Service.h: Added a new class called Dynamic_Service which provides a general interface to add and retrieve arbitrary things into the ACE service repository. * apps/Name_Server/README: Rearranged files so that Name_Server and Name_Handler are now contained in the server directory under apps and are no longer under /ace. Also, the Dump_Restore directory has been moved under /tests/Naming. Thu Feb 22 01:56:46 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/INET_Addr.cpp (set): Fixed a byteorder bug in ACE_INET_Addr that was causing problems when ACE was used on Win32 running on Intel platforms... * ace/Mem_Map.cpp (map_it): Fixed ACE_Mem_Map so that it will work correctly on Win32. * ace/SOCK_Stream.i: Changed the implementation of the Win32 versions of send() and recv() so that they will never use the read() and write() calls. This works around limitations with Win32, which can't do a ReadFile() or WriteFile() *synchronously* with a SOCKET. * Changed all occurrences of read (0, ...) to read (ACE_STDIN, ...) and all occurrences of write (1, ...) to write (ACE_STDOUT, ...) in order to be compatible across UNIX and Win32. * tests/ASX/Event_Server: Fixed up all the socket I/O calls so that they use the WinSock send()/recv() methods rather than ReadFile and WriteFile... * ace/OS.i: Updated open() to correctly emulated UNIX open() semantics on Win32. * ace/OS.cpp (thr_exit): Added support to the NT port so that thr_exit() doesn't endup doing multiple deletions of the same object. Thanks to Karlheinz for this fix. * ace/OS.h: Moved the location of ACE_MAXCLIENTIDLEN so that MAXHOSTNAMELEN would be correctly visible for NT. * ace/Reactor.cpp: Added a temporary fix for the fact that writev() and readv() don't work correctly yet on NT. The solution is to replace write() with two writes() and readv() with two reads() for the Reactor's notification scheme. A better solution should be forthcoming. * ace/Reactor.cpp: Made sure to initialize the timer_queue_ to 0 in the constructors so that the Reactor's destructor can avoid crashing horribly if the constructor fails. Thanks to Karlheinz for pointing out the need for this. * ace/SOCK.cpp: Added support for WSAStartup for using WinSock within the ACE_SOCK class. Wed Feb 21 21:05:55 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/ACE.cpp (bind_port): Made a quick fix to work around the fact that NT doesn't set errno... * ace/OS.i (mmap): Fixed a bug in mmap for the NT port. It should return MAP_FAILED on error rather than 0 to be equivalent to the UNIX version. Thanks to Irfan for spotting this. Mon Feb 19 00:31:42 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Typed_SV_Message.i (ACE_Typed_SV_Message): Modified the constructor calls so that length_ and max_size_ are correctly set when created. Thanks to Alexandre Karev for reporting this. * ace/SV_Message_Queue.i (remove): fixed bug in ACE_SV_Message_Queue::remove method where `internal_id_' member was modified *BEFORE* it was used the last time in ACE_SV_Message_Queue::control. Thanks to Alexandre Karev for reporting this. * ace/Typed_SV_Message.i (length): Fixed a minor typo that caused a bigger bug ;-). Thanks to Alexandre Karev for reporting this. * ace/OS.h: Added new macros that allow programs to portably access stdin, stdout, and stderr HANDLEs on Win32 and UNIX as follows: #if defined (ACE_WIN32) #define ACE_STDIN GetStdHandle (STD_INPUT_HANDLE) #define ACE_STDOUT GetStdHandle (STD_OUTPUT_HANDLE) #define ACE_STDERR GetStdHandle (STD_ERROR_HANDLE) #else #define ACE_STDIN 0 #define ACE_STDOUT 1 #define ACE_STDERR 2 #endif /* ACE_WIN32 */ * ace/INET_Addr.i: modified the return value of get_port_number so that it returns the value in host byte order. This has actually been a latent bug in ACE for years, but it only surfaced recently after porting to NT on the Intel instruction set (which is "little-endian"). * tests/ASX/Event_Server: Merged in the latest changes to the Event_Server tests, which enable it to compile on Windows NT. Now, I just need to get it working on NT... * Updated all of ACE to use the new ACE_ASSERT rather than assert. * ace/Assert.h: Added a new file and a new macro called ACE_ASSERT. This uses the ACE_Log_Msg Singleton to provide a more consistent means to do assertions. Thanks to Alexandre Karev for suggesting this. * ace/Log_Msg.cpp (log): added for method ACE_Log_Msg::log lines to skip the `sprintf' part for %N - file name and %l - line number format specifiers. Thanks to Alexandre Karev for reporting this. * Remote_Name_Space: Removed an initalizer to ACE_WString that was driving the HP/UX compiler nuts. Sun Feb 18 18:11:22 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Synch.h (ACE_Null_Condition_Mutex): Added a missing body to the wait() method. Thanks to Byron Riggs for noticing this. Sat Feb 17 19:10:06 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * tests/Naming/README (client): Added description for new features, including support for list_values and list_types. Also added a small description about test programs using DLL. * ace/Name_Request_Reply.cpp (decode): Fixed a small bug in ACE_Name_Request::decode. type_ was not null terminated causing some garbage to be returned. Thu Feb 15 14:57:06 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Remote_Name_Space.cpp: Changed names of some parameters to comply with ACE syntax. Wed Feb 14 13:36:31 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * ace/Local_Name_Space.cpp (resolve): Type support has been added to Name_Server. A new class called ACE_NS_Internal was created to keep value and type information. Fri Feb 9 17:12:00 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * "Officially" released ACE 4.0 * ace/OS.i (sema_post): Fixed a bug in the NT port. #elif defined (ACE_HAS_WTHREADS) return ::ReleaseSemaphore (*s, 1, 0); #endif /* ACE_HAS_STHREADS */ This returns 1 for success, and disturbs cond_broadcast... This was replaced by: #elif defined (ACE_HAS_WTHREADS) return ::ReleaseSemaphore (*s, 1, 0) ? 0 : -1; #endif /* ACE_HAS_STHREADS */ Thanks to Jesper S. M|ller (stophph@diku.dk) for this. * ace/OS.i (cond_init): Fixed a bug in the NT condition synch code, as the initial count of the semaphore was 1. This gives an inconsistent condition: The first waiter went straight through... Change in ACE_OS::cond_init: if (ACE_OS::sema_init (&cv->sema_, 0, type, name, arg) != 0) return -1; Thanks to Jesper S. M|ller (stophph@diku.dk) for this. * ace/Singleton.cpp: Fixed a stupid bug that was caused by failure to #include "ace/Synch_T.h"... Thanks to Neil Cohen and Byron Riggs for giving me insights on what the problem was. ACE now compiles on HP/UX. Fri Feb 9 11:07:04 1996 Prashant Jain (pjain@merengue.cs.wustl.edu) * tests/Naming/server/Server_Test.cpp (init): Modified server tests to dynamically link Naming Services. Both client and server tests now make use of the ACE dynamic linking mechanisms. * ace/Local_Name_Space.cpp: Added list_values() to Name Server. The functionality is very similar to list_names (returns list of values that match a pattern). * tests/Naming/client/Client_Test.cpp (bind): Replaced all uses of cerr and cout with ACE_ERROR, ACE_ERROR_RETURN, and ACE_DEBUG. Split the file Client_Test.cpp into two files -- Client_Test.cpp and Name_Service.cpp (similarly .h file) each containing the corresponding class. Modified svc.conf to work with the changes. Thu Feb 8 02:05:26 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Log_Msg.cpp (log): Enhanced the Log_Msg class so that it supports "nested" calls via %r. If SILENT mode enabled, nested outputs will enclosed in {..} brackets in thread-specific storage. Thanks to Alexandre Karev for suggesting this. * ace/Synch: Added a new class called ACE_Null_Condition_Mutex. This is necessary to fix bugs with stupid compilers... Thanks to Zusman Mark for reporting this. * tests/Naming/client: made a first pass implementation of a client application that dynamically links in a Naming_Context based on information provided by the svc.conf file. * ace/Name_Options: Fixed yet more problems with dynamic memory management. The old version didn't make a copy of the strings it was passed and this caused major problems. * ace: Fixed a whole slew of problems with the ACE Name Server stuff that arose from inconsistent use of const char * vs. char *... Wed Feb 7 00:58:45 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/CORBA_Handler.cpp (ACE_MT_CORBA_Handler): Added new enhancements to ACE_MT_CORBA_Handler to make it work correctly with Windows NT. Thanks to Irfan for making these changes. * ace/Connector.cpp (fini): Make another workaround for bugs with MSVC++ 2.0... * ace/SPIPE_Connector.cpp (connect): Added a call to ACE_CLR_BITS (flags, O_CREAT) to make darn sure that the O_CREAT flag is not set. Thanks to Chris Cleeland for suggesting this. Tue Feb 6 01:40:29 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.h: added a new prototype to OS.h that should help with portability to nasty platforms with broken rand_r() definitions. Thanks to Aurelio Nocerino for suggesting this. * Local_Tokens: Changed all uses of enumerals MAXTOKENNAMELEN and MAXCLIENTIDLEN to symbol #defines ACE_MAXTOKENNAMELEN and ACE_MAXCLIENTIDLEN. This works around bugs with HP/UX. Thanks to Neil Cohen for reporting this problem. * ChangeLog: split the ChangeLog files into 4 separate files -- one for '96, '95, '94', and '93. Hard to believe I've been working on this stuff for so long! * ace/OS: Fixed the use of ::memset to replace sizeof buffer with sizeof (ACE_SERVENT_DATA) and sizeof (ACE_HOSTENT_DATA). This fixes bugs that surfaced on AIX. Thanks to Chris Lahey for finding these. * ace/SPIPE_Addr.cpp (ACE_SPIPE_Addr): Fixed a bug in SPIPE_Addr::SPIPE_Addr(const SPIPE_Addr&). What happens is that when the size is calculated it does not take into account the required zero byte at the end of the rendezvous_ member. Thanks to Chris Cleeland for reporting this. * ace/Local_Tokens: Had to move all ACE_Token_Proxy methods from Local_Tokens.i to Local_Tokens.cpp and remove INLINE - otherwise gcc reported them as undefined methods during linkage stage. Thanks to Zusman Mark for reporting this. * ace: Changed ACE_Name_Request_Reply::LAST to ACE_Name_Request_Reply::MAX_ENUM to avoid problems with name conflicts on OSF/1. Thanks to Eshel Liran for suggesting this. * ace/Synch_T: Fix all definitions of ACE_Atomic_Ops so that they compile correctly on platforms that lack threads! Thanks to Alexandre Karev for noticing this. * include/makeinclude: Added a new config file from Bryon G. Rigg , which should allow ACE to build on Linux. * ace/Remote_Tokens.cpp: Moved ACE_TSS_Connection to Remote_Tokens.h to avoid problems with AIX. Thanks to the ever-vigilant Chris Lahey for reporting this. Mon Feb 5 23:34:42 1996 Tim H. Harrison (harrison@lambada.cs.wustl.edu) * Added CORBA_Ref.h which contains the new class ACE_CORBA_Ref. This class provides a nice wrapper for making the use of Orbix object references more transparent. This is done by automating the release and duplicate calls. Mon Feb 5 15:43:17 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: added Prashant's latest changes to the Name_Handler*, Naming_Context*, and Remote_Name_Space*. These changes make it possible to retrieve lists of values based on a pattern, which can be a regular expression. * Moved the apps/Name_Server/Client-Server tests into tests/Naming and most of the apps/Token_Server/* tests into tests/Tokens to conform to Karlheinz's testing style. * ace: Modified a number of files in ACE to make it compile correctly with MSVC++ 2.0. Sun Feb 4 23:58:43 1996 Douglas C. Schmidt (schmidt@mambo.cs.wustl.edu) * ace/Reactor: Generalized ACE_Null_Callback to work correctly with Windows NT (which lacks pipes). The new design uses the ACE socket wrappers for NT. It requires very few changes to the code to handle NT now! Also changed the name to ACE_Notification_Handler, which is more descriptive. Sun Feb 4 14:47:50 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Remote_Tokens.cpp: Reworked ACE_Singleton to avoid portability problems on compilers like GCC. Thu Feb 1 12:47:46 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Remote_Tokens.cpp: Generalized the parameters passed into ACE_Singleton so they would compile on HP/UX and other platforms that lack threads. Thanks to Neil Cohen for noticing the problem. Wed Jan 31 22:49:13 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Modified a number of minor things in ACE to get it to compile successfully with Windows NT MSVC++ 4.0. Updated the INSTALL file to explain this process better... Tue Jan 30 01:12:07 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/CORBA_Handler.cpp: changed things so that no diagonistic messages are printed. * Regenerated all the ACE manual pages to bring them up to date with the recent changes. * ace/Malloc.h: Modified the ACE_Malloc header file so that it works correctly when ACE_MALLOC_STATS is enabled. Thanks to Alexandre Karev for reporting this. * ace/OS.cpp (mutex_lock_cleanup): removed the leading & before p_lock. Thanks to Chris Lahey for noticing this. * ace: Added the new ACE_Singleton class. This class uses the Adapter pattern to turn ordinary classes into Singletons optimized with the Double-Check pattern. * Added Tim's latest Token Server changes (which include support for local and remote mutexes and readers/writer locks) and Prashant's latest Name Server changes (which allows the contents of a Name Server to be dumped and restored to/from an ASCII file). Mon Jan 29 02:22:23 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Handle_Set: Changed MAX_SIZE and WORD_SIZE to MAXSIZE and WORDSIZE to avoid name collisions with HP/UX. Thanks to Byron Rigg for suggesting this. * ace/Time_Value.h: Added a new config symbol ACE_HAS_SVR4_TIME to differentiate between UNIX platforms that support "POSIX_TIME" (i.e., timespec_t) and those that support "SVR4_TIME" (i.e., timestruc_t). This is necessary to work around HP/UX... Thanks to Byron Rigg for suggesting this. * ace/config-hpux-10.x.h: config-hpux.h has now been renamed to config-hpux-10.x.h and config-hpux-9.x.h. Thanks to Byron Rigg for suggesting this. * ace/Synch_T: Added a new method called ts_get() and made both the operator-> and operator TYPE * call this. I hope this will fix a bug with HP/UX reported by Neil Cohen... * ace/OS.i (dlclose): Added Win32 support for dlclose(). Thanks to Todd L. Montgomery for pointing the way here... * ace: Split off the old class ACE_Dynamically_Allocated, renamed it ACE_Dynamic, put it in a separate file called Dynamic.*. This is necessary to solve nasty multiple definition problems with compilers that require the source of template. * ace/Synch_T: moved template TYPE * ACE_TSS::make_TSS_TYPE (void) const { return new TYPE; } out from condition compilation block #if (defined (ACE_HAS_THREADS) && defined(ACE_HAS_THREAD_SPECIFIC_STORAGE)) in file Synch_T.cpp. This class member called in many other places regardless ACE_HAS_THREADS and ACE_HAS_THREAD_SPECIFIC_STORAGE defined or not... * ace/Name_Handler.cpp: Changed the use of the "inherited" typedef to fully expand to ACE_Svc_Handler. This is necessary to workaround bugs with AIX. Thanks to Chris Lahey for reporting this. * tests and apps: Fixed an odd problem that only surfaced recently with GNU G++. Apparently, it doesn't like function-style casts of the form ACE_SignalHandler (foo), but will accept (ACE_SignalHandler) foo. Since both approaches are equally good for what I'm doing, I've updated all the code in ACE to be more portable. Sun Jan 28 12:24:58 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * tests/Shared_Malloc/test_malloc.cpp (spawn): when forming argv[] for execv, moved argv[ ... ] = Options::instance ()->debug () ? "-d" : ""; to the end of the vector just before terminator. This will allow to pass rest of arguments to exec if -d switch not specified. Thanks to Alexandre Karev for reporting this. * tests/Shared_Malloc/Malloc.cpp: To avoid conflict with ACE_DEFAULT_SEM_KEY in OS.h changed SEMA_KEY value in tests/Shared_Malloc/Malloc.cpp to ACE_DEFAULT_SEM_KEY + 1. Thanks to Alexandre Karev for reporting this. * ace/SV_Semaphore_Complex.i: Changed all arguments in calls like this->acquire( n+2 ) and this->tryacquire( n+2 ) to this->acquire( n ) and this->tryacquire( n ) . Since the semaphore number is incremented by 2 in SV_Semaphore_Complex::acquire and SV_Semaphore_Complex::tryacquire when SV_Semaphore_Simple::(acquire/tryacquire) is called. Thanks to Alexandre Karev for pointing this out. Sat Jan 27 16:14:27 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Read_Buffer: Added a new component that efficiently reads in an artibrarily large buffer from a file up to an including a termination character. Also performs search/replace on single occurrences a character in the buffer using the priniciples of Integrated Layer Processing. Fri Jan 26 12:01:06 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Thread_Manager: Made a small change to workaround the use of a struct type for thread_t and hthread_t in HP/UX. Thanks to Bryon Rigg for noticing this. Thu Jan 25 19:54:01 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Svc_Handler: Added some minor changes to account for limitations with GNU G++. I think the new implementation will be portable across all C++ platforms. * ace/SV_Semaphore_*: Widened the interface of SV_Semaphore* classes to include a flags parameter. This can be used to pass in SEM_UNDO, which is important in many applications. Thanks to Andrew Gilbert for suggesting this. * ace/Synch_T.cpp: We are forced to "cast away const" inside of of the ACE_TSS const methods in order to make the ACE_Guard work. The right thing to do would be to make the lock "mutable" but that's hopelessly non-portable. Thu Jan 25 14:34:12 1996 Douglas C. Schmidt (schmidt@merengue.cs.wustl.edu) * ace/Log_Msg: Fixed a minor bug in ACE_Log_Msg that kept resetting the restart and iostream value no matter what we'd set it to before. Thanks to Prashant for finding this. * bin: Fixed up the Makefile process for the clone program. Now it builds without having to depend on ACE. Thanks to Brad Brown (bbrown@rdxsunhost.aud.alcatel.com) for suggesting this. * ace/Synch: Modified ACE_Condition*.remove so that cond_destroy() is called in a loop if the condition variable is BUSY. This avoids a condition where a condition is signaled and because of some timing problem, the thread that is to be signaled has called the cond_wait routine after the signal call. Since the condition signal is not queued in any way, deadlock occurs unless this loop is used. Thanks to Chris Lahey for pointing this out. * ace/CORBA_Handler: Made all the methods in ACE_CORBA_Handler private to make sure that users don't inherit from this class! Instead, the ACE_MT_CORBA_HAndler and ACE_ST_CORBA_Handler should be used as Singletons. * ace/CORBA_Handler: Added new support for Orbix on Windows NT. This requires clever use of macros in order to handle inherent differences between generated code. * ace/Svc_Handler: Added a clever idiom that transparently checks if we were allocated dynamically. This information is used by the method to decide if we need to delete ... The idiom is based on a paper by Michael van Rooyen (mrooyen@cellnet.co.uk) that appeared in the April '96 issue of the C++ Report. We've spruced it up to work correctly in multi-threaded programs by using our ACE_TSS class. * ace/config-win32-msvc4.0.h (ACE_LACKS_RECVMSG): Fixed a typo in the MSVC++ config files. Thu Jan 25 02:59:22 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Finally got the frigging library built on SGI. This was a chore due to compiler bugs, but at least it's another datapoint for successful cross-platform building... Wed Jan 24 00:10:29 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Time_Value.cpp (operator-=): Changed operator += and operator -= to use const ACE_Time_Value & rather than ACE_Time_Value &. Thanks to Alex V Maclinvosky for noticing this. In addition, also changed *all* ACE_Time_Value parameters to be const ACE_Time_Value &. This should reduce the number of excess constructors called... * ace/OS: Added Chris Lahey's latest changes that introduce POSIX pthread cancellation cleanup handlers. Also added an herror() function that is comparable to perror(). * ace/OS: (thr_join): Added a new version of thr_join() which works for Windows NT. Windows NT requires a HANDLE, which is fundamentally incompatible with other things... * ace/Thread_Manager: added two new methods that return the ACE_Thread_Descriptor corresponding to either a thr_id or a thr_handle. * ace/Thread_Manager: Made the spawn() interface compatible with the other spawn() interfaces in ACE_OS and ACE_Thread by adding the hthread_t type. * ace/Log_Msg.cpp (log): Fixed things up so that we can now print out the thread id for all versions of threads! * ace/OS: Added a new overloaded version of ACE_OS::thr_self() that returns the "kernel" thread id. This is necessary on systems like NT and AIX, which separate kernel thread ids from user thread ids. Thanks to Chris Lahey for suggesting this. Tue Jan 23 01:17:23 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (sigwait): Changed the parameter of sigwait() from const sigset * to sigset *. Thanks to Neil Cohen for chasing this down on SunOS 5.5.... * ace/Malloc_T.cpp: Fixed a typo in bind() whereby the test if (duplicates == 0 || this->shared_find (name) != 0) should be if (duplicates == 0 && this->shared_find (name) != 0). Thanks to the ever vigilant Alexandre Karev for reporting this. * tests: Fixed all test programs that previously used the "execl" family of system calls to use the "execv" family instead. * ace/OS: Added support for the "execv" family of exec() system calls. Unlike the "execl" family, these calls are easy to support since they don't require variadic arguments. I've added the hooks for the "execl" family, as well, but haven't actually implemented them yet. * ace/Memory_Pool.cpp (ACE_Shared_Memory_Pool): Fixed a bug that was tickled when pool_name == 0. Thanks to Alexandre Karev for reporting this. * ace/OS.h: removed FD_SETSIZE from the ACE_WIN32 stuff since this is defined in winsock.h. * ace/OS.cpp (ACE_Spawn_Args): Added "f" to the parameter list. * ace/CORBA_Handler: Added a new macro called CORBA_T that masks the incompatibilities between the version of Orbix on NT and on Solaris. * ace/OS.h: Added support for compilers (like NT) that don't support "mode masks" (these are used to give symbolic names for file creation modes passed to open() and creat(). * ace/SString.cpp: Added Tim's new copy constructor code for SString.cpp. This doesn't appear to be strictly necessary , but it is more explicit and therefore it useful. * ace/Map_Manager.cpp: Fixed a bug in Map_Manager that was failing to call the placement new operator for every element in the map that was dynamically created by the allocator(). * ace/Synch_T: Added Tim's new ACE_TSS implementation. This version is more flexible than the old one. Mon Jan 22 00:03:24 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Time_Value.h: changed a few minor things to get this compiling again on NT. Thanks to Irfan for helping with this. * ace/Synch_Options: Changed all the "accessor" methods to be const member functions at the request of Irfan. I just know this is going to cause const cancer... ;-) * ace/Acceptor and Connector: Changed all parameters that take ACE_Synch_Options & to take const ACE_Synch_Options & at the request of Irfan. * ace/OS.i (uname): Added rudimentary support for the uname() function for Windows NT. If anyone has suggestions on how to fill in all these fields portably please let me know! * ace/Connector.h: Changed a stray -1 to ACE_INVALID_HANDLE. Thanks to Irfan for noticing this. * ace/Profile_Timer.h: Added the keyword "public" to handle one arm of the conditional compilation that is only exercised on NT... Thanks to Irfan for noticing this. * ace/OS.i (mutex_init): Removed the call to pthread_mutexattr_setkind_np() it isn't in the final POSIX standard. * ace/Reactor.cpp (unbind): Removed the ACE_MAX3 template function from OS.h and inlined its only use in the Reactor. This was causing problems with some C++ compilers. Thanks to Mark Zusman for reporting this. * ace/OS.i: Fixed a number of omitted return results in gettimeofday() and ctime_r() that were masked by the ACE_OSCALL_RETURN macros. Thanks to Mark Zusman for reporting these. * ace/config-linux.h: Added new symbolic constants to handle the fact that Linux seems to lack recvmsg(), sendmsg(), msync(), and madvise(). Thanks to Neil Cohen for noticing this. * include/makeinclude: Fixed all the *-orbix.GNU files to use $(ORBIX_ROOT)/lib rather than $(ORBIX_ROOT)/ace. Thanks to Pramod Kumar Singh for reporting this. * ace/config*.h: Changed all typos of the form ACE_HAS_SVSV_IPC to ACE_HAS_SYSV_IPC. Thanks to Alexandre Karev for reporting this! * ace/OS: Changed the order of parameters passed to thr_create() to be consistent with all the spawn() and spawn_n() usages in ACE. Thanks to Chris Lahey for pointing this out. Sun Jan 21 15:06:15 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Added a new "priority" parameter to the spawn() and spawn_n() methods in ACE_Thread and ACE_Thread_Manager. This allows us to pass the priority on down to thr_create(). Thanks to Chris Lahey for pointing this out. Note that this change required reordering the parameters so that the stack and stack_size parameters now come last for these methods. Since those are not likely to change much this is a good place to put them (i.e., at the end). * ace/Thread.i (spawn_n): Added a new overloaded spawn_n() method that accepts an array of thread_t's to be filled in. This is useful if you want to know the thread IDs of any threads created. * Changed ACE_OS::signal() to ACE_Sig_Action across all of ACE to be portable to pthreads implementations. Thanks to Chris Lahey for pointing this out. * ace/OS: Added a new set of OS C++ wrappers for POSIX pthread cancellation routines. Thanks to Chris Lahey for these. * ace/Thread: Added a new set of ACE C++ wrappers for POSIX pthread cancellation routines. Thanks to Chris Lahey for these. * ace/Map_Manager: Revised the Map_Manager::bind method. This has now been split into bind() and rebind() methods. The bind() method is very simple -- if you try to bind() to something that already exists you fail. The rebind() method allows you to atomically update existing values in a map. It also gives you back the existing values so that you can delete them if necessary. The Local_Name_Space::bind() and rebind() methods have been updated to use this new interface. * ace/Malloc_T.cpp (find): Fixed a typo -- find() should have set its pointer parameter to node->pointer_, rather than node->name_! This bug was revealed during testing of the ACE Name Server. * ace/Local_Name_Space.cpp: Fixed bugs with computation of the size of ACE_NS_Strings -- they needed to count the wchar_t's correctly, along with several other minor fixes. These were revealed during testing of the ACE Name Server. * ace/SString.cpp: Fixed several bugs that prevented the assignment operators from working correctly. These were revealed during testing of the ACE Name Server. Sat Jan 20 08:33:54 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Local_Name_Space.cpp (shared_bind): The args to to the NS_String constructors were not in the correct order: wchar_t *value_rep = (wchar_t *) (ptr); wchar_t *name_rep = (wchar_t *) (ptr + name_len); ACE_NS_String ext_id (name.fast_rep (), name_rep, name_len); ACE_NS_String int_id (value.fast_rep (), value_rep, value_len); They should be: ACE_NS_String ext_id (name_rep, name.fast_rep (), name_len); ACE_NS_String int_id (value_rep, value.fast_rep (), value_len); Thanks to Irfan for noticing this. * ace/OS.h: defined the following in ACE_OS: #define MS_ASYNC 0 #define MS_SYNC 0 #define MS_INVALIDATE 0 under NT so that calls to mmap will port transparently. Thanks to Irfan for suggesting this. Thu Jan 18 16:25:16 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Added a new symbol that indicates if the platform supports the new BSD sin_len field of inet_addr. * ace/OS: added new support for AIX gethostbyaddr_r(), gethostbyname_r(), and getservbyname_r(). This fixes some problems caused by improper choice of buffer size. Also changed the INET_Addr.cpp file to use these new changes. Thanks to Chris Lahey for recommending these changes. Wed Jan 17 01:10:48 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (printf): the ACE_OS::printf() function did not work. Needed to call vprintf() internally, not printf. Thanks to Chris Lahey for pointing this out. * Released version 4.0. * include/makeinclude: added the -lgen library to all the platform_sunos5.*.GNU files. * ace/Memory_Pool.cpp: removed the INLINE macros in Memory_Pool.cpp. Thanks to Alexandre Karev (karev@vxcern.cern.ch) for noticing this! * ace/Malloc_T.cpp: moved the comment "Ce n'est pas une catst" outside of the macro ACE_NEW_RETURN. The prepocessor truncates the macro expansion when it encounters the comment. Thanks to Greg Baker for pointing this out. * ace/Reactor.cpp (close): added "return 0" at the end of the close member function. The function is prototyped to return an int and the compiler expects a return outside of the if clause. Thanks to Greg Baker for pointing this out. * ace/Stack.cpp (free_all_nodes): Removed a spurious ACE_OS that was stuck on the front of ::delete. Thanks to Greg Baker for pointing this out. * ace/OS.i: Added extern "C" to the prototype for syscall in OS.i. Without it, you get unresolved symbols at link time. Thanks to Greg Baker for pointing this out. * ace/Message_Queue.cpp: Changed the return values for enqueue_head(), enqueue_tail(), dequeue_head(), and peek_dequeue_head() to return the number of messages that are still on the queue. This helps some algorithms perform better and avoid blocking unnecessarily. Thanks to Alex V Maclinvosky for suggesting this... * ace/Synch_T.cpp (ACE_Condition): Fixed a bug in ~ACE_Recursive_Lock and ~ACE_Recursive_Mutex that was causing the destructor to be called twice (once implicitly and once explicitly). Thanks to Chris Lahey for pointing this out. Mon Jan 15 12:44:29 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Created a new config symbol called ACE_HAS_ALLOCA_H. This is required to differentiate between AIX and other versions of UNIX. Thanks to Chris Lahey for pointing this out. * ace: moved config-hpux.h to config-hpux-9.x.h in anticipation of the changes required to handle HP/UX 10.0! Sun Jan 14 23:38:23 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.h (ACE_MAX3): Fixed the definition of ACE_MAX3 (it needed a return value of template type T). Thanks to Mark Zusman for noticing this. * ace: Added a new section to all the template files called ACE_TEMPLATES_REQUIRE_SPECIALIZATION to deal with the screwy GNU G++ semantics that are required if you *don't* have the template repository patches. Thanks to Mark Zusman for this. Fri Jan 12 00:47:57 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Time_Value.h: #included to the Time_Value.h class so that it would be the first file included. This fixes a problem with AIX. Thanks to Chris Lahey for reporting this and suggesting the fix. * ace: Fixed a braino with the #ifdefs in SPIPE_Stream and SPIPE_Acceptor. These were checking for ACE_WIN32 rather than ACE_HAS_STREAM_PIPES. Naturally, that failed on versions of UNIX (e.g., Linux) that lack STREAM pipes! Thanks to Neil Cohen for taking time out of fighting blizzards to report this! * ace/Svc_Handler: Added a new method for handle_timeout, which shuts things down by default. Thanks to Irfan for suggesting this. * Fixed a minor problem caused when Tim added the Double-Check pattern to ACE_ODB... Thu Jan 11 01:48:02 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * Integrated AIX fixes from Chris Lahey . These should allow ACE to build correctly using the AIX compiler and it's screwy template mechanisms! * ace: Fixed some minor bugs that caused problems when compiling with G++. Wed Jan 10 00:17:05 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Trace.cpp: Fixed a typo that was causing problems when using ACE_TRACE (we were ending up with recursive tracing in the ACE_Trace class!). Thanks to Detlef for noticing this. * ace/ACE.cpp: Moved methods from ACE.i to ACE.cpp and removed the INLINE macro to deal with order of include problems with GNU G++. Tue Jan 9 19:00:41 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Task.i (flush): Added a check to make sure that msg_queue_ != 0 before trying to close() the queue. This solves a problem if the queue has already been closed. Thanks to Alex V Maclinvosky for reporting this... * tests/ASX/Event_Server/Event_Server: Fixed minor bugs in Supplier_Router.cpp and Consumer_Router.cpp -- I was checking for this->getq (mb) > 0 rather than this->getq (mb) >= 0. Thanks to Alex V Maclinvosky for noticing this... * ace/Synch_T.h: Commented out the ACE_ALLOC_HOOK stuff. This is causing order-of-include problems on HP/UX. Also moved the ACE_Null_Condition::wait() method to Synch_T.i to avoid the same problem. Thanks to Greg Baker for pointing this out. Sun Jan 7 18:57:49 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Makefile: Added the Dump file to ACE. Thanks to Tim for fixing this up to make it ready for prime time. * Thanks to Craig Rodrigues for sending me a new version of psnup that *finally* fixes the nasty problems with "4-up" printing of postscript! This makes it much easier to distribute ACE documentation... Sun Jan 7 18:31:07 1996 Tim H. Harrison (harrison@merengue.cs.wustl.edu) * ace: Used the "Double-Check" pattern to eliminate potential race conditions when using Singletons in multi-threaded programs. Classes affected included ACE_[MS]T_CORBA_Handler, Name_Options, Trace, and ACE_ODB. Fri Jan 5 00:03:29 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i: Added the following to OS.i: #if defined (ACE_LACKS_SYSCALL) int syscall (int, ACE_HANDLE, struct rusage *); #endif /* ACE_LACKS_SYSCALL */ This should fix some problems with HP/UX. Thanks to Walt Akers (akers@cebaf.gov) for pointing out the problem. * ace/Handle_Set: Fixed the Handle_Set conversion operator so that it returns fd_set * rather than ACE_FD_SET_TYPE *. Thanks to Walt Akers (akers@cebaf.gov) for pointing out the problem. * ace/Time_Value.h: I think I've fixed the problem with the order of #includes with respect to POSIX and SVR4 time. The trick was to put this test in ace/Time_Value.h and then to make sure that the ace/config.h file was included at this point! Also changed the symbol from ACE_HAS_POSIX_TIMERS to ACE_HAS_POSIX_TIME to be more accurate. Thu Jan 4 23:16:59 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/OS.i (gettimeofday): Changed the implementation of gettimeofday() to use the ::_ftime call rather than the more complex ::GetSystemTime(), ::SystemTimeToFileTime() duo. Thu Jan 4 9:32:38 1996 Chris Lahey (clahey@ix.netcom.com) * ace/config-aix.h: added #define ACE_TEMPLATES_REQUIRE_SOURCE to make ACE work with AIX C++ compiler. * include/makeinclude/platform_aix.GNU: Added -qtempinc to the CPPFLAGS Removed the PIC= definition completely * ace/Task.h: ACE_Task::svc() should not be declared as a pure virtual, removed = 0. * ace/Map_Manager.cpp: in trybind() member function, the line int_id = ss.int_id_; should be: int_id = search_structure[index].int_id_; Thu Jan 4 01:23:38 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/Map_Manager.cpp: Fixed a braino in Map_Manager -- didn't need the parameter to shared_bind(). Thanks to Tim for finding this! * ace/Log_Msg.cpp (instance): Tim Harrison did a noble service to humanity by fixing a potential race condition in the instance() Singleton of ACE_Log_Msg. Wed Jan 3 00:49:57 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace: Made some minor changes to fix portability problems with SGI. Fortunately, this should also fix some other problems with HP/UX another other non-thread platforms. * ace/ACE.cpp (format_hexdump): Added Todd Montgomery's amazing "format_hexdump" method to class ACE. Tue Jan 2 20:47:57 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/SV_Semaphore*: updated the interface of the _Simple and _Complex System V semaphore wrappers to include the _read() and _write() forms of acquire() and tryacquire(). Note that I've implemented these as calls to acquire() and tryacquire(), so they don't add any extra behavior. This is needed for interface conformance with other forms of ACE synchronization mechanisms. * ace/Time_Value.h: Changed the header to #include in order to work around nasty "order of include" problems. * tests: Created a new directory called Win32. This contains test programs that This directory contains test programs that have been compiled and tested successfully under Windows NT. As we get further along, we'll move this stuff into the general ACE $WRAPPER_ROOT/tests directory (since of course all ACE programs will run without any problems under Win32, right? ;-)). See the ./tests/Win32/README file for instructions on how to build these tests. * ace/Reactor.h: Moved the ACE_Handler_Repository out of the ACE_MT_SAFE section of the code. This was a "typo." Thanks to Walt Akers for reporting this. Mon Jan 1 01:18:12 1996 Douglas C. Schmidt (schmidt@tango.cs.wustl.edu) * ace/ACE.i: Moved the recv_n() and send_n() methods from ACE.cpp into ACE.i and set things so that if __INLINE__ is set then we'll inline those too! * ace: Irfan and I got the first major part of the Win32 to work finally! We made two socket programs (client and server) that talked to each other and to UNIX. Things are really starting to take shape!