Tue Jan 7 13:03:25 1997 Douglas C. Schmidt * ace/Message_Block: Added a new constructor to ACE_Message_Block that takes an ACE_Data_Block * and "owns" it. Thanks to Tim for this idea. * ace/Log_Msg.cpp: Now that we've prevented signal handlers from occurring within critical sections of ACE_Log_Msg::log() we don't need to use the ACE_Recursive_Thread_Mutex anymore. Instead, we just need ACE_Thread_Mutex. * ace/Log_Msg.cpp (log): Added an ACE_Sig_Guard to the block in ACE_Log_Msg::log() that acquires the mutex that serializes output. This prevents nasty problems with recursive to ACE_Log_Msg::log() from within signal handlers. * ace/Service_Config.cpp (end_reactor_event_loop): Added a timeout of ACE_Time_Value::zero to the ACE_Reactor::notify() method when called in the ACE_Service_Config::end_reactor_event_loop(). This prevents the Reactor from blocking indefinitely if there's no longer a thread to receive from the notification pipe. * netsvcs/clients/Naming/Dump_Restore: Removed the vestigal nametest.cpp and nametest.h files. I'm not sure why they were still there, but they shouldn't have been! * ace/Synch.i (remove): this->owner_ should be set to -1 before calling this->release(). Thanks to Per Andersson for suggesting this. * ace/Thread_Manager.cpp (exit): Added a flag called "do_thr_exit" to the ACE_Thread_Control::exit() method. This controls whether we call ACE_OS::thr_exit() after removing the thread from the Thread Manager's internal table. This is necessary to work around pthread bugs on certain platforms (e.g., AIX and Linux). * ace/Synch.h: Removed the ACE_Recursive_Thread_Mutex::set_nesting() method since this isn't being used and is potentially dangerous. Thanks to Per Andersson for suggesting this. * ace/XtReactor: Updated the XtReactor so that it uses the next internal Reactor implementation. Thanks to Mark Rabotnikov for reporting this. * ace/Task.i: Updated the grp_id(int) method to cache the group id in the task and then set it in the Thread_Manager, if there is one. Thanks to Hamutal Yanay for suggesting this. * ace/Log_Record.h: Updated the ACE_Log_Record class to use the new ACE_MAXLOGMSGLEN macro. * ace/OS.h: Added a new #define called ACE_MAXLOGMSGLEN. By default this is 4k. However, you can reset it in the config.h file if you'd like to increase or decrease the size. Thanks to Anthony McConnell for suggesting this. * examples/Threads/tss[12].cpp: Tidied up the tss1.cpp and tss2.cpp examples and improved the comments. Thanks to Luca for suggesting this. * ace/Service_Object.cpp (fini): Added the "virtual" keyword to the new destructor of ACE_Service_Type. Thanks to David for reporting this. * ace/Message_Block.i: Rearranged the order of the ACE_Data_Block and ACE_Message_Block locking_strategy() methods so that they'd be defined as inline before being used. Thanks to David for reporting this. Mon Jan 6 16:46:06 1997 Douglas C. Schmidt * ace/Service_Manager.cpp (fini): Only try to remove the ACE_Service_Manager from the Reactor if it's actually been registered in the first place (i.e., if it's get_handle() method does not return ACE_INVALID_HANDLE). * netsvcs/servers/svc.conf: Changed the name of the net services .so file from libnet_svcs to netsvcs so that it will work correctly with our new "cross-platform" svc.conf ACE::ldfind() strategy. Thanks to Marius Kjeldahl for reporting this. * ace/Reactor.cpp (find): Added a strategic break and a check for handle_in_range(). Thanks to Tim for this help! * netsvcs/clients/Naming/Client/svc.conf: Removed the "lib" part from the executable shared object file names in the svc.conf file so that it will work with the new "cross-platform" svc.conf ACE::ldfind() strategy. Thanks to Marius Kjeldahl for reporting this. * ace/Service_Record.cpp (fini): Added an extra assignment to a temporary const void *obj to avoid problems with MSVC++ and it's cast mechanism. * ace/ReactorEx.h: Added a default parameter to ACE_ReactorEx::notify(). Thanks to Tim for noticing this. Sun Jan 5 10:48:23 1997 Douglas C. Schmidt * tests: Added a new test for ACE_Message_Block, which is called Message_Block_Test.cpp (no surprise there ;-)). This test exercises the reference counting mechanisms. * ace/Message_Block.h: Made the destructors for ACE_Message_Block and ACE_Data_Block virtual, so that these classes can be subclassed and ACE_Message_Blocks can delete these subclasses (either via release() or delete). Thanks to Tilo Christ for suggesting this. * ace/Message_Block.cpp: Modified the implementation of ACE_Message_Block/ACE_Data_Block so that we always create an allocator_strategy_. By default, this strategy will be ACE_Allocator_Adapter >; This greatly simplifies the code. * ace/Message_Block.h: It's now possible to assign a Message_Block's priority when you create it. * ace/Reactor: Sped things up some more by merging the Reactor's any_ready() and fill_in_ready() methods together. This reduces redundant computations. * ace/Reactor: Changed the name of ACE_Reactor::detach() to ACE_Reactor::remove_handler_i() and changed the name of ACE_Reactor::attach() to ACE_Reactor::register_handler_i(). This is more consistent with the other naming schemes. * ace/Reactor.cpp: Reworked the implementation of the Handler Repository's unbind() method in order to simplfy the logic. Also merged in all the Reactor's detach() functionality into the unbind() method in order to localize all the handler removal processing. * ace/Reactor[Ex].cpp: Added more descriptive errno entries for failed operations in the handler repository. * ace/Reactor.cpp: Modified the implementation of the Handler Repository's bind() method so that it reports an error if you try to register an Event_Handler for an invalid HANDLE. * ace/Reactor.cpp: Reimplemented the internals of the Reactor to prepare for the upcoming merge of the Reactor and ReactorEx to use a single code base. * ace/Timer_Queue.cpp: Modified the behavior of the expire() methods so that they return the number of ACE_Event_Handlers whose handle_timeout() method was called. Sat Jan 4 11:47:05 1997 Douglas C. Schmidt * ace/Reactor.h: Removed the #ifdefs for ACE_HAS_POLL. This will be recast as a strategy in subsequent implementations. * ace/Reactor.cpp: Added a new call, ACE_Handler_Repository::handle_in_range(), which checks to make sure that the HANDLE is within the range of 0 .. max_handlep1_. This factors out some redundant code. * ace/Reactor.cpp: Sped up the performance of the Reactor by removing redundant calls to ACE_Handler_Repository::invalid_handle() and ACE_Handler_Repository::find(). * tests/Reactors_Test.cpp: Added a test for the new Reactor::notify() method with timeouts. * examples/Reactor/Misc/notification.cpp (svc): Added a test for the new Reactor::notify() method with timeouts. * ace/Reactor[Ex]: Modified the Reactor and ReactorEx so that they use timed send() operations for their notify() methods. This makes it possible for users to bound the amount of time they're willing to wait to try and notify the Reactor. This is mostly useful to avoid infinite blocking if there is congestion in the Reactor's notification mechanism. * ace/ACE.cpp: Added a new timed send() method that uses the write() system call on UNIX and the socket send() call on Win32. This new method is used in the new ACE_Reactor::notify() method, which now supported timed notifications. * ace/Reactor.cpp (ACE_Handler_Repository): To simplify the code and reduce unnecessary #ifdefs, I've merged the max_handlep1_ and cur_size_ fields (which are now called max_handlep1_). Fri Jan 3 10:47:15 1997 Douglas C. Schmidt * ace: Replaced all uses of "delete mb" with mb->release (); * ace/Stream_Modules.cpp: Replaced the use of explicit bit twiddling with the ACE_BIT* macros. * ace/Message_Block.cpp: Make sure that we use the allocator_strategy_ to create the memory for the reference count since this may need to go into shared memory if that's the memory pool where the Message_Block allocations are coming from. * ace/OS.h: Added two new macros, ACE_ALLOCATOR_RETURN and ACE_ALLOCATOR, which are similar to ACE_NEW_RETURN and ACE_NEW, except that these * ace/Message_Block.cpp (release): Make sure to "delete this" outside the scope of the locking_strategy_. * ace/Service_Object.cpp: Added a destructor to ACE_Service_Type. Thanks to Per.Andersson@hfera.ericsson.se (Per Andersson) for suggesting this. * ace/Service_Object.i: Be smarter about how we reassign the name_ pointer, i.e., delete the old one and make a copy. Thanks to Per.Andersson@hfera.ericsson.se (Per Andersson) for reporting this. * ace/Module.cpp (open): Rearranged the assignments to reader_q->mod_ and writer_q->mod_ so that we don't try to initialize through NULL pointers. Thanks to Per.Andersson@hfera.ericsson.se (Per Andersson) for reporting this. * ace/Service_Record.cpp (ACE_Service_Record): Initialized name_ to NULL so that the following change works correctly now. Thanks to Per.Andersson@hfera.ericsson.se (Per Andersson) for reporting this. * ace/Service_Record.i (name): Make sure to delete [] (char *) this->name_ before allocating a new one. Thanks to Per.Andersson@hfera.ericsson.se (Per Andersson) for reporting this. * ace/Message_Block: Reworked the reference counting implemention so that reference counts are shared correctly amongst their various owners. This requires making a deep copy the "header" portion, but a shallow copy of the "data." * ace/Message_Block.cpp (ACE_Message_Block): Updated all three ACE_Message_Block constructors so that they all call the init() method. This centralizes all the initialization logic in one place. Thu Jan 2 00:42:21 1997 Douglas C. Schmidt * ace/Message_Block.cpp (ACE_Message_Block): Make sure to set the cont_ field to 0 after "releasing" it so that we don't mistakenly think it's still around later on. This problem arose in the ACE_Message_Queue::close() method, which was trying to count the number of bytes being freed. * ace/Message_Queue.cpp (close): Fixed a subtle bug where we weren't actually deleting messages from the ACE_Message_Queue::close() routine. This should work now... * ace/Message_Queue.cpp (close): Replaced the use of "delete mb" with "mb->release()" since the Message_Blocks are now reference counted. * ace/Message_Block: Enhanced the reference counting scheme so that you can increment and decrement the count by an arbitrary amount. This is particular useful when you know you'll be sending the same Message_Block to N consumers. * ace/Singleton: The dump() must be used same as instance() (without supplying an object) so it must be declarated *static*, i.e., static void dump (void); Thanks to Sandro Doro for reporting this. * examples/ASX/Event_Server: Completely rewrote and retested the ACE Event Server example. The new code is *much* easier to understand, has many more comments, is more robust, and compiles much faster since I removed many of the templates. * examples/ASX/Event_Server/Tranceiver/tranceiver.cpp: Fixed the tranceiver so that it shuts down correctly when the Event_Server exits. * examples/Connection/non_blocking/CPP-connector.cpp: Fixed a problem where we were trying to select() on ACE_STDIN on Win32. Naturally, this doesn't work, so we used the ACE::register_stdin_handler() to fix this. Thanks to Samuel_Bercovici for reporting this. * examples/ASX/Event_Server/Event_Server/Options: Changed the Options class to be a Singleton... * ace/Task.h: Added "const" method qualifier to accessor methods like is_reader() and is_writer(). Wed Jan 1 00:10:47 1997 Douglas C. Schmidt * ace/Message_Block: Added a new static release() method that behaves like the non-static method release(), except that it checks if is 0. This is similar to CORBA::release(), which is useful if you want to eliminate lots of checks for NULL pointers before calling release() on them. * Updated all places in ACE where we use ACE_Event_Handler::READ_MASK when we should be using ACE_Event_Handler::ACCEPT_MASK. * examples/Service_Configurator/IPC-tests/server/Handle_Thr_Stream.cpp: Changed the name of the acceptor class from Handle_Thr_Stream to Handle_Thr_Acceptor, which is more accurate. * ace/Reactor: Since we've now got an ACE_Event_Handler::ACCEPT_MASK (intended primarily to accept connections asynchronously using Win32 overlapped I/O) we can now use this with the ACE_Reactor, as well. In particular, any time that we are registering a "passive-mode" socket acceptor we can now specify ACE_Event_Handler::ACCEPT_MASK. This is much more intuitive than using the READ_MASK (which never made any sense...). The ACE_Reactor now treats the ACCEPT_MASK as a READ_MASK internally, so you can still use READ_MASK if you'd like (i.e., this change doesn't break any existing code). * ace/Event_Handler: Changed "get_priority" and "set_priority" to simply "priority(void)" and "priority(int)" to be more consistent with other parts of ACE. * apps/Gateway/Gateway: Updated the Gateway so that the concurrency strategies can now be specified on the command-line (or in the svc.conf file), rather than being determined at compile-time. This is much more flexible. See the ./apps/Gateway/Gateway/README file for details. * apps/Gateway/Gateway: Verified that all the multi-threading strategies still work. * ace/Message_Block.cpp (ACE_Message_Block): Make sure that we "release" the continuation field, rather than delete it, since its reference count may be > 1. * ace/Log_Msg.cpp (log): Change the abort_prog logic a bit so that we always print a message to stderr if we're exiting, regardless... Thanks to David Levine for pointing this out. * apps/Gateway/Gateway: Moved all of the configuration file parsing logic *outside* of the Event_Channel into the Gateway class so that we wouldn't have unnecessary dependencies. * apps/Gateway/Gateway: Redesigned the Gateway so that the Proxy_Handlers (i.e., the Consumer_Proxy and Supplier_Proxy) most of their work to the Event_Channel. This "lightweight proxy" design is an improvement since it is now possible to emulate the COS Event Channel semantics within the Event_Channel "kernel." * Happy new year! Let's start a new ChangeLog to celebrate the new year.