Merge with 4.0.13

include/thr_alarm.h

@@ -100,7 +100,7 @@ typedef struct st_alarm {
 #define thr_alarm_init(A) (*(A))=0
 #define thr_alarm_in_use(A) (*(A)!= 0)
 void init_thr_alarm(uint max_alarm);
-bool thr_alarm(thr_alarm_t *alarmed, uint sec, ALARM *buff);
+my_bool thr_alarm(thr_alarm_t *alarmed, uint sec, ALARM *buff);
 void thr_alarm_kill(pthread_t thread_id);
 void thr_end_alarm(thr_alarm_t *alarmed);
 void end_thr_alarm(my_bool free_structures);

innobase/com/com0shm.c

@@ -103,7 +103,8 @@ struct com_shm_endpoint_struct{
 				the area currently may contain a datagram;
 				NOTE: automatic event */
 	os_event_t	empty; 	/* this is in the signaled state if the area
-				currently may be empty; NOTE: automatic event */
+				currently may be empty; NOTE: automatic
+				event */
 	ip_mutex_hdl_t*	ip_mutex; /* handle to the interprocess mutex
 				protecting the shared memory */
 	UT_LIST_NODE_T(com_shm_endpoint_t) list; /* If the endpoint struct
@@ -793,16 +794,18 @@ com_shm_create_or_open(
 
 	ut_strcpy(buf + len, (char*)"_IBSHM_EV_NE"),
 
-	event_ne = os_event_create_auto(buf);
+	event_ne = os_event_create(buf);
 
 	ut_ad(event_ne);
 
 	ut_strcpy(buf + len, (char*)"_IBSHM_EV_EM"),
 
-	event_em = os_event_create_auto(buf);
+	event_em = os_event_create(buf);
 
 	ut_ad(event_em);
 
+	ut_a(0); /* event_ne and event_em should be auto events! */
+
 	com_shm_endpoint_set_shm(ep, shm);
 	com_shm_endpoint_set_map(ep, map);
 

innobase/include/os0sync.h

@@ -13,13 +13,26 @@ Created 9/6/1995 Heikki Tuuri
 #include "ut0lst.h"
 
 #ifdef __WIN__
+
 #define os_fast_mutex_t CRITICAL_SECTION
-typedef HANDLE          os_event_t;
+
+typedef HANDLE          os_native_event_t;
+
+typedef struct os_event_struct os_event_struct_t;
+typedef os_event_struct_t*     os_event_t;
+
+struct os_event_struct {
+	os_native_event_t		  handle;
+					/* Windows event */
+	UT_LIST_NODE_T(os_event_struct_t) os_event_list;
+					/* list of all created events */
+};
 #else
 typedef pthread_mutex_t	os_fast_mutex_t;
 
 typedef struct os_event_struct os_event_struct_t;
 typedef os_event_struct_t*     os_event_t;
+
 struct os_event_struct {
 	os_fast_mutex_t	os_mutex;	/* this mutex protects the next
 					fields */
@@ -39,16 +52,16 @@ typedef os_mutex_str_t*		os_mutex_t;
 
 #define OS_SYNC_TIME_EXCEEDED	1
 
-/* Mutex protecting the thread count and event and OS 'slow' mutex lists */
+/* Mutex protecting counts and the event and OS 'slow' mutex lists */
 extern os_mutex_t	os_sync_mutex;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
 extern ulint		os_thread_count;
 
-/* The following are approximate counters for debugging in Unix */
 extern ulint		os_event_count;
 extern ulint		os_mutex_count;
+extern ulint		os_fast_mutex_count;
 
 /*************************************************************
 Initializes global event and OS 'slow' mutex lists. */
@@ -57,15 +70,14 @@ void
 os_sync_init(void);
 /*==============*/
 /*************************************************************
-Frees created events (not in Windows) and OS 'slow' mutexes. */
+Frees created events and OS 'slow' mutexes. */
 
 void
 os_sync_free(void);
 /*==============*/
-/*************************************************************
-Creates an event semaphore, i.e., a semaphore which may
-just have two states: signaled and nonsignaled.
-The created event is manual reset: it must be reset
+/************************************************************* 
+Creates an event semaphore, i.e., a semaphore which may just have two states:
+signaled and nonsignaled. The created event is manual reset: it must be reset
 explicitly by calling sync_os_reset_event. */
 
 os_event_t
@@ -74,10 +86,10 @@ os_event_create(
 			/* out: the event handle */
 	char*	name);	/* in: the name of the event, if NULL
 			the event is created without a name */
+#ifdef __WIN__
 /*************************************************************
-Creates an auto-reset event semaphore, i.e., an event
-which is automatically reset when a single thread is
-released. */
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
 
 os_event_t
 os_event_create_auto(
@@ -85,6 +97,7 @@ os_event_create_auto(
 			/* out: the event handle */
 	char*	name);	/* in: the name of the event, if NULL
 			the event is created without a name */
+#endif
 /**************************************************************
 Sets an event semaphore to the signaled state: lets waiting threads
 proceed. */
@@ -120,7 +133,7 @@ os_event_wait(
 	os_event_t	event);	/* in: event to wait */
 /**************************************************************
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. */
+a timeout is exceeded. In Unix the timeout is always infinite. */
 
 ulint
 os_event_wait_time(
@@ -131,8 +144,9 @@ os_event_wait_time(
 	os_event_t	event,	/* in: event to wait */
 	ulint		time);	/* in: timeout in microseconds, or
 				OS_SYNC_INFINITE_TIME */
+#ifdef __WIN__
 /**************************************************************
-Waits for any event in an event array. Returns if even a single
+Waits for any event in an OS native event array. Returns if even a single
 one is signaled or becomes signaled. */
 
 ulint
@@ -140,14 +154,15 @@ os_event_wait_multiple(
 /*===================*/
 					/* out: index of the event
 					which was signaled */
-	ulint		n,		/* in: number of events in the
+	ulint		        n,	/* in: number of events in the
 					array */
-	os_event_t* 	event_array);	/* in: pointer to an array of event
+	os_native_event_t* 	native_event_array);
+					/* in: pointer to an array of event
 					handles */
+#endif
 /*************************************************************
-Creates an operating system mutex semaphore.
-Because these are slow, the mutex semaphore of the database
-itself (sync_mutex_t) should be used where possible. */
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
 
 os_mutex_t
 os_mutex_create(

innobase/include/os0sync.ic

@@ -44,4 +44,3 @@ os_fast_mutex_trylock(
 #endif
 #endif
 }
-

innobase/include/os0thread.h

@@ -65,7 +65,9 @@ os_thread_pf(
 /********************************************************************
 Creates a new thread of execution. The execution starts from
 the function given. The start function takes a void* parameter
-and returns a ulint. */
+and returns a ulint.
+NOTE: We count the number of threads in os_thread_exit(). A created
+thread should always use that to exit and not use return() to exit. */
 
 os_thread_t
 os_thread_create(

innobase/os/os0file.c

@@ -80,6 +80,8 @@ struct os_aio_slot_struct{
 					which pending aio operation was
 					completed */
 #ifdef WIN_ASYNC_IO
+        os_event_t	event;		/* event object we need in the
+					OVERLAPPED struct */
 	OVERLAPPED	control;	/* Windows control block for the
 					aio request */
 #elif defined(POSIX_ASYNC_IO)
@@ -107,11 +109,14 @@ struct os_aio_array_struct{
 	ulint		n_reserved;/* Number of reserved slots in the
 				  aio array outside the ibuf segment */
 	os_aio_slot_t* 	slots;	  /* Pointer to the slots in the array */
-	os_event_t*	events;	  /* Pointer to an array of event handles
-				  where we copied the handles from slots,
-				  in the same order. This can be used in
-				  WaitForMultipleObjects; used only in
+#ifdef __WIN__
+	os_native_event_t* native_events;	 
+				  /* Pointer to an array of OS native event
+				  handles where we copied the handles from
+				  slots, in the same order. This can be used
+				  in WaitForMultipleObjects; used only in
 				  Windows */
+#endif
 };
 
 /* Array of events used in simulated aio */
@@ -296,7 +301,7 @@ os_file_handle_error(
 				operation */
 	os_file_t	file,	/* in: file pointer */
 	char*		name,	/* in: name of a file or NULL */
-	const char*	operation) /* in: type of operation */
+	const char*	operation)/* in: operation */
 {
 	ulint	err;
 
@@ -338,8 +343,8 @@ os_file_handle_error(
 	        if (name) {
 	                fprintf(stderr, "InnoDB: File name %s\n", name);
 	        }
-		fprintf(stderr, "InnoDB: system call %s\n", operation);
-
+	  
+		fprintf(stderr, "InnoDB: System call %s.\n", operation);
 		fprintf(stderr, "InnoDB: Cannot continue operation.\n");
 
 		fflush(stderr);
@@ -422,9 +427,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -465,10 +469,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -576,9 +578,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -625,9 +626,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -1319,19 +1319,22 @@ os_aio_array_create(
 	array->n_segments	= n_segments;
 	array->n_reserved	= 0;
 	array->slots		= ut_malloc(n * sizeof(os_aio_slot_t));
-	array->events		= ut_malloc(n * sizeof(os_event_t));
-	
+#ifdef __WIN__
+	array->native_events	= ut_malloc(n * sizeof(os_native_event_t));
+#endif	
 	for (i = 0; i < n; i++) {
 		slot = os_aio_array_get_nth_slot(array, i);
 
 		slot->pos = i;
 		slot->reserved = FALSE;
 #ifdef WIN_ASYNC_IO
+		slot->event = os_event_create(NULL);
+
 		over = &(slot->control);
 
-		over->hEvent = os_event_create(NULL);
+		over->hEvent = slot->event->handle;
 
-		*((array->events) + i) = over->hEvent;
+		*((array->native_events) + i) = over->hEvent;
 #endif
 	}
 	
@@ -1429,7 +1432,7 @@ os_aio_array_wake_win_aio_at_shutdown(
 
 	for (i = 0; i < array->n_slots; i++) {
 
-	        os_event_set(*(array->events + i));
+	        os_event_set((array->slots + i)->event);
 	}
 }
 #endif
@@ -1689,7 +1692,7 @@ loop:
 	control = &(slot->control);
 	control->Offset = (DWORD)offset;
 	control->OffsetHigh = (DWORD)offset_high;
-	os_event_reset(control->hEvent);
+	os_event_reset(slot->event);
 
 #elif defined(POSIX_ASYNC_IO)
 
@@ -1747,7 +1750,7 @@ os_aio_array_free_slot(
 	}
 
 #ifdef WIN_ASYNC_IO		
-	os_event_reset(slot->control.hEvent);
+	os_event_reset(slot->event);
 #endif
 	os_mutex_exit(array->mutex);
 }
@@ -1872,7 +1875,7 @@ os_aio(
 				offset where to read or write */
 	ulint		offset_high, /* in: most significant 32 bits of
 				offset */
-	ulint		n,	/* in: number of bytes to read or write */	
+	ulint		n,	/* in: number of bytes to read or write */
 	void*		message1,/* in: messages for the aio handler (these
 				can be used to identify a completed aio
 				operation); if mode is OS_AIO_SYNC, these
@@ -1916,7 +1919,8 @@ os_aio(
 		wait in the Windows case. */
 
 		if (type == OS_FILE_READ) {
-			return(os_file_read(file, buf, offset, offset_high, n));
+			return(os_file_read(file, buf, offset,
+							offset_high, n));
 		}
 
 		ut_a(type == OS_FILE_WRITE);
@@ -1994,8 +1998,7 @@ try_again:
 #ifdef WIN_ASYNC_IO
 	if (os_aio_use_native_aio) {
 		if ((ret && len == n)
-			|| (!ret && GetLastError() == ERROR_IO_PENDING)) {	
-
+		    || (!ret && GetLastError() == ERROR_IO_PENDING)) {
 			/* aio was queued successfully! */
 		
 	    		if (mode == OS_AIO_SYNC) {
@@ -2025,8 +2028,8 @@ try_again:
 
 	os_aio_array_free_slot(array, slot);
 
-	retry = os_file_handle_error(file, name, "aio");
-
+	retry = os_file_handle_error(file, name,
+			type == OS_FILE_READ ? "aio read" : "aio write");
 	if (retry) {
 
 		goto try_again;
@@ -2091,15 +2094,15 @@ os_aio_windows_handle(
 	n = array->n_slots / array->n_segments;
 
 	if (array == os_aio_sync_array) {
-		srv_io_thread_op_info[orig_seg] = "wait Windows aio for 1 page";
-
-		ut_ad(pos < array->n_slots); 
-		os_event_wait(array->events[pos]);
+		srv_io_thread_op_info[orig_seg] =
+						"wait Windows aio for 1 page";
+		os_event_wait(os_aio_array_get_nth_slot(array, pos)->event);
 		i = pos;
 	} else {
 		srv_io_thread_op_info[orig_seg] =
 						"wait Windows aio";
-		i = os_event_wait_multiple(n, (array->events) + segment * n);
+		i = os_event_wait_multiple(n,
+				(array->native_events) + segment * n);
 	}
 
 	os_mutex_enter(array->mutex);
@@ -2124,7 +2127,7 @@ os_aio_windows_handle(
 		         ut_a(TRUE == os_file_flush(slot->file));
 		}
 	} else {
-		os_file_handle_error(slot->file, slot->name, "aio");
+		os_file_handle_error(slot->file, slot->name, "Windows aio");
 		
 		ret_val = FALSE;
 	}		  

innobase/os/os0sync.c

@@ -32,24 +32,23 @@ struct os_mutex_struct{
 				/* list of all 'slow' OS mutexes created */
 };
 
-/* Mutex protecting the thread count and the lists of OS mutexes
-and events */
+/* Mutex protecting counts and the lists of OS mutexes and events */
 os_mutex_t	os_sync_mutex;
 ibool		os_sync_mutex_inited	= FALSE;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
-ulint		os_thread_count		= 0;
+ulint	os_thread_count		= 0;
 
-/* The list of all events created (not in Windows) */
+/* The list of all events created */
 UT_LIST_BASE_NODE_T(os_event_struct_t)	os_event_list;
 
 /* The list of all OS 'slow' mutexes */
 UT_LIST_BASE_NODE_T(os_mutex_str_t)	os_mutex_list;
 
-/* The following are approximate counters for debugging in Unix */
-ulint		os_event_count		= 0;
-ulint		os_mutex_count		= 0;
+ulint	os_event_count		= 0;
+ulint	os_mutex_count		= 0;
+ulint	os_fast_mutex_count	= 0;
 
 
 /*************************************************************
@@ -68,7 +67,7 @@ os_sync_init(void)
 }
 
 /*************************************************************
-Frees created events (not in Windows) and OS 'slow' mutexes. */
+Frees created events and OS 'slow' mutexes. */
 
 void
 os_sync_free(void)
@@ -89,6 +88,12 @@ os_sync_free(void)
 	mutex = UT_LIST_GET_FIRST(os_mutex_list);
 
 	while (mutex) {
+	      if (mutex == os_sync_mutex) {
+		      /* Set the flag to FALSE so that we do not try to
+		      reserve os_sync_mutex any more in remaining freeing
+		      operations in shutdown */
+		      os_sync_mutex_inited = FALSE;
+	      }
 
 	      os_mutex_free(mutex);
 
@@ -97,10 +102,9 @@ os_sync_free(void)
 }
 
 /*************************************************************
-Creates an event semaphore, i.e., a semaphore which may
-just have two states: signaled and nonsignaled.
-The created event is manual reset: it must be reset
-explicitly by calling sync_os_reset_event. */
+Creates an event semaphore, i.e., a semaphore which may just have two
+states: signaled and nonsignaled. The created event is manual reset: it
+must be reset explicitly by calling sync_os_reset_event. */
 
 os_event_t
 os_event_create(
@@ -112,20 +116,18 @@ os_event_create(
 #ifdef __WIN__
         os_event_t event;
 
-	event = CreateEvent(NULL,	/* No security attributes */
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL,/* No security attributes */
 			TRUE,		/* Manual reset */
 			FALSE,		/* Initial state nonsignaled */
 			name);
-	if (!event) {
+	if (!event->handle) {
 	        fprintf(stderr,
 "InnoDB: Could not create a Windows event semaphore; Windows error %lu\n",
 		  (ulint)GetLastError());
 	}
-
-	ut_a(event);
-
-	return(event);
-#else
+#else /* Unix */
 	os_event_t	event;
 
 	UT_NOT_USED(name);
@@ -141,7 +143,9 @@ os_event_create(
 	ut_a(0 == pthread_cond_init(&(event->cond_var), NULL));
 #endif
 	event->is_set = FALSE;
+#endif /* __WIN__ */
 
+        /* Put to the list of events */
 	os_mutex_enter(os_sync_mutex);
 
 	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
@@ -151,13 +155,12 @@ os_event_create(
 	os_mutex_exit(os_sync_mutex);
 
 	return(event);
-#endif
 }
 
+#ifdef __WIN__
 /*************************************************************
-Creates an auto-reset event semaphore, i.e., an event
-which is automatically reset when a single thread is
-released. */
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
 
 os_event_t
 os_event_create_auto(
@@ -166,26 +169,33 @@ os_event_create_auto(
 	char*	name)	/* in: the name of the event, if NULL
 			the event is created without a name */
 {
-#ifdef __WIN__
         os_event_t event;
 
-	event = CreateEvent(NULL,	/* No security attributes */
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL,/* No security attributes */
 			FALSE,		/* Auto-reset */
 			FALSE,		/* Initial state nonsignaled */
 			name);
-	ut_a(event);
 
-	return(event);
-#else
-	/* Does nothing in Posix because we do not need this with MySQL  */
+	if (!event->handle) {
+	        fprintf(stderr,
+"InnoDB: Could not create a Windows auto event semaphore; Windows error %lu\n",
+		  (ulint)GetLastError());
+	}
 
-	UT_NOT_USED(name);
+        /* Put to the list of events */
+	os_mutex_enter(os_sync_mutex);
 
-	ut_a(0);
+	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
 
-	return(NULL);
-#endif
+	os_event_count++;
+
+	os_mutex_exit(os_sync_mutex);
+
+	return(event);
 }
+#endif
 
 /**************************************************************
 Sets an event semaphore to the signaled state: lets waiting threads
@@ -198,7 +208,7 @@ os_event_set(
 {
 #ifdef __WIN__	
 	ut_a(event);
-	ut_a(SetEvent(event));
+	ut_a(SetEvent(event->handle));
 #else
 	ut_a(event);
 
@@ -227,7 +237,7 @@ os_event_reset(
 #ifdef __WIN__
 	ut_a(event);
 
-	ut_a(ResetEvent(event));
+	ut_a(ResetEvent(event->handle));
 #else
 	ut_a(event);
 
@@ -255,12 +265,14 @@ os_event_free(
 #ifdef __WIN__
 	ut_a(event);
 
-	ut_a(CloseHandle(event));
+	ut_a(CloseHandle(event->handle));
 #else
 	ut_a(event);
 
 	os_fast_mutex_free(&(event->os_mutex));
 	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+        /* Remove from the list of events */
 
 	os_mutex_enter(os_sync_mutex);
 
@@ -271,7 +283,6 @@ os_event_free(
 	os_mutex_exit(os_sync_mutex);
 
 	ut_free(event);
-#endif
 }
 
 /**************************************************************
@@ -291,7 +302,7 @@ os_event_wait(
 	ut_a(event);
 
 	/* Specify an infinite time limit for waiting */
-	err = WaitForSingleObject(event, INFINITE);
+	err = WaitForSingleObject(event->handle, INFINITE);
 
 	ut_a(err == WAIT_OBJECT_0);
 
@@ -324,7 +335,7 @@ loop:
 
 /**************************************************************
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. */
+a timeout is exceeded. In Unix the timeout is always infinite. */
 
 ulint
 os_event_wait_time(
@@ -341,9 +352,9 @@ os_event_wait_time(
 	ut_a(event);
 
 	if (time != OS_SYNC_INFINITE_TIME) {
-		err = WaitForSingleObject(event, time / 1000);
+		err = WaitForSingleObject(event->handle, time / 1000);
 	} else {
-		err = WaitForSingleObject(event, INFINITE);
+		err = WaitForSingleObject(event->handle, INFINITE);
 	}
 	
 	if (err == WAIT_OBJECT_0) {
@@ -367,8 +378,9 @@ os_event_wait_time(
 #endif
 }
 
+#ifdef __WIN__
 /**************************************************************
-Waits for any event in an event array. Returns if even a single
+Waits for any event in an OS native event array. Returns if even a single
 one is signaled or becomes signaled. */
 
 ulint
@@ -376,18 +388,18 @@ os_event_wait_multiple(
 /*===================*/
 					/* out: index of the event
 					which was signaled */
-	ulint		n,		/* in: number of events in the
+	ulint			n,	/* in: number of events in the
 					array */
-	os_event_t*	event_array)	/* in: pointer to an array of event
+	os_native_event_t*	native_event_array)
+					/* in: pointer to an array of event
 					handles */
 {
-#ifdef __WIN__
 	DWORD	index;
 
-	ut_a(event_array);
+	ut_a(native_event_array);
 	ut_a(n > 0);
 
-	index = WaitForMultipleObjects(n, event_array,
+	index = WaitForMultipleObjects(n, native_event_array,
 					FALSE,	   /* Wait for any 1 event */
 					INFINITE); /* Infinite wait time
 						   limit */
@@ -399,21 +411,12 @@ os_event_wait_multiple(
 	}
 
 	return(index - WAIT_OBJECT_0);
-#else
-	ut_a(n == 0);
-	
-	/* In Posix we can only wait for a single event */
-
-	os_event_wait(*event_array);
-
-	return(0);
-#endif
 }
+#endif
 
 /*************************************************************
-Creates an operating system mutex semaphore.
-Because these are slow, the mutex semaphore of the database
-itself (sync_mutex_t) should be used where possible. */
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
 
 os_mutex_t
 os_mutex_create(
@@ -430,50 +433,35 @@ os_mutex_create(
 			FALSE,		/* Initial state: no owner */
 			name);
 	ut_a(mutex);
-
-	mutex_str = ut_malloc(sizeof(os_mutex_str_t));
-
-	mutex_str->handle = mutex;
-	mutex_str->count = 0;
-
-	if (os_sync_mutex_inited) {
-	        os_mutex_enter(os_sync_mutex);
-	}
-
-	UT_LIST_ADD_FIRST(os_mutex_list, os_mutex_list, mutex_str);
-
-	if (os_sync_mutex_inited) {
-		os_mutex_exit(os_sync_mutex);
-	}
-
-	return(mutex_str);
 #else
-	os_fast_mutex_t*	os_mutex;
+	os_fast_mutex_t*	mutex;
 	os_mutex_t		mutex_str;
 
 	UT_NOT_USED(name);
 	
-	os_mutex = ut_malloc(sizeof(os_fast_mutex_t));
-
-	os_fast_mutex_init(os_mutex);
+	mutex = ut_malloc(sizeof(os_fast_mutex_t));
 
+	os_fast_mutex_init(mutex);
+#endif	
 	mutex_str = ut_malloc(sizeof(os_mutex_str_t));
 
-	mutex_str->handle = os_mutex;
+	mutex_str->handle = mutex;
 	mutex_str->count = 0;
 
 	if (os_sync_mutex_inited) {
+		/* When creating os_sync_mutex itself we cannot reserve it */
 	        os_mutex_enter(os_sync_mutex);
 	}
 
 	UT_LIST_ADD_FIRST(os_mutex_list, os_mutex_list, mutex_str);
 
+	os_mutex_count++;
+
 	if (os_sync_mutex_inited) {
 		os_mutex_exit(os_sync_mutex);
 	}
 
 	return(mutex_str);
-#endif	
 }
 
 /**************************************************************
@@ -513,21 +501,14 @@ os_mutex_exit(
 /*==========*/
 	os_mutex_t	mutex)	/* in: mutex to release */
 {
-#ifdef __WIN__
 	ut_a(mutex);
 
 	ut_a(mutex->count == 1);
 
 	(mutex->count)--;
-
+#ifdef __WIN__
 	ut_a(ReleaseMutex(mutex->handle));
 #else
-	ut_a(mutex);
-
-	ut_a(mutex->count == 1);
-
-	(mutex->count)--;
-
 	os_fast_mutex_unlock(mutex->handle);
 #endif	
 }
@@ -540,25 +521,25 @@ os_mutex_free(
 /*==========*/
 	os_mutex_t	mutex)	/* in: mutex to free */
 {
-#ifdef __WIN__
 	ut_a(mutex);
 
-	os_mutex_enter(os_sync_mutex);
+	if (os_sync_mutex_inited) {
+		os_mutex_enter(os_sync_mutex);
+	}
 
 	UT_LIST_REMOVE(os_mutex_list, os_mutex_list, mutex);
+	
+	os_mutex_count--;
 
-	os_mutex_exit(os_sync_mutex);
+	if (os_sync_mutex_inited) {
+		os_mutex_exit(os_sync_mutex);
+	}
 
+#ifdef __WIN__
 	ut_a(CloseHandle(mutex->handle));
 
 	ut_free(mutex);
 #else
-	os_mutex_enter(os_sync_mutex);
-
-	UT_LIST_REMOVE(os_mutex_list, os_mutex_list, mutex);
-
-	os_mutex_exit(os_sync_mutex);
-
 	os_fast_mutex_free(mutex->handle);
 	ut_free(mutex->handle);
 	ut_free(mutex);
@@ -583,8 +564,19 @@ os_fast_mutex_init(
 #else
 	ut_a(0 == pthread_mutex_init(fast_mutex, MY_MUTEX_INIT_FAST));
 #endif
-	os_mutex_count++;
 #endif
+	if (os_sync_mutex_inited) {
+		/* When creating os_sync_mutex itself (in Unix) we cannot
+		reserve it */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count++;
+
+	if (os_sync_mutex_inited) {
+	        os_mutex_exit(os_sync_mutex);
+	}
 }
 
 /**************************************************************
@@ -631,6 +623,17 @@ os_fast_mutex_free(
 	DeleteCriticalSection((LPCRITICAL_SECTION) fast_mutex);
 #else
 	ut_a(0 == pthread_mutex_destroy(fast_mutex));
-	os_mutex_count--;
 #endif
+	if (os_sync_mutex_inited) {
+		/* When freeing the last mutexes, we have
+		already freed os_sync_mutex */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count--;
+
+	if (os_sync_mutex_inited) {
+		os_mutex_exit(os_sync_mutex);
+	}
 }

innobase/os/os0thread.c

@@ -186,6 +186,10 @@ os_thread_exit(
 	void*	exit_value)	/* in: exit value; in Windows this void*
 				is cast as a DWORD */
 {
+#ifdef UNIV_DEBUG_THREAD_CREATION
+        printf("A thread exits.\n");
+        printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	os_mutex_enter(os_sync_mutex);
 	os_thread_count--;
 	os_mutex_exit(os_sync_mutex);

innobase/srv/srv0srv.c

@@ -868,6 +868,7 @@ srv_release_max_if_no_queries(void)
 	mutex_exit(&kernel_mutex);
 }
 
+#ifdef notdefined
 /***********************************************************************
 Releases one utility thread if no queries are active and
 the high-water mark 2 for the utility is exceeded. */
@@ -902,7 +903,6 @@ srv_release_one_if_no_queries(void)
 	mutex_exit(&kernel_mutex);
 }
 
-#ifdef notdefined
 /***********************************************************************
 Decrements the utility meter by the value given and suspends the calling
 thread, which must be an utility thread of the type given, if necessary. */
@@ -1012,6 +1012,8 @@ srv_communication_init(
 	
 	ut_a(ret == 0);
 }
+
+#ifdef notdefined
 	
 /*************************************************************************
 Implements the recovery utility. */
@@ -1072,6 +1074,7 @@ srv_purge_thread(
 
 	return(0);
 }
+#endif /* notdefined */
 
 /*************************************************************************
 Creates the utility threads. */
@@ -1102,6 +1105,7 @@ srv_create_utility_threads(void)
 	ut_a(thread); */
 }
 
+#ifdef notdefined
 /*************************************************************************
 Implements the communication threads. */
 static
@@ -1151,6 +1155,7 @@ srv_com_thread(
 
 	return(0);
 }
+#endif
 
 /*************************************************************************
 Creates the communication threads. */
@@ -1171,6 +1176,7 @@ srv_create_com_threads(void)
 	}
 }
 
+#ifdef notdefined
 /*************************************************************************
 Implements the worker threads. */
 static
@@ -1215,6 +1221,7 @@ srv_worker_thread(
 
 	return(0);
 }
+#endif
 
 /*************************************************************************
 Creates the worker threads. */
@@ -2490,6 +2497,10 @@ srv_lock_timeout_and_monitor_thread(
 	char*		buf;
 	ulint		i;
 
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Lock timeout thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	UT_NOT_USED(arg);
 	srv_last_monitor_time = time(NULL);
 	last_table_monitor_time = time(NULL);
@@ -2630,6 +2641,10 @@ loop:
 exit_func:
 	srv_lock_timeout_and_monitor_active = FALSE;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
 #ifndef __WIN__
         return(NULL);
 #else
@@ -2655,6 +2670,10 @@ srv_error_monitor_thread(
 	ulint	cnt	= 0;
 
 	UT_NOT_USED(arg);
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Error monitor thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 loop:
 	srv_error_monitor_active = TRUE;
 
@@ -2691,6 +2710,9 @@ loop:
 
 	srv_error_monitor_active = FALSE;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
 	os_thread_exit(NULL);
 
 #ifndef __WIN__
@@ -2771,6 +2793,10 @@ srv_master_thread(
 	
 	UT_NOT_USED(arg);
 
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Master thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	srv_main_thread_process_no = os_proc_get_number();
 	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
 	
@@ -3006,6 +3032,15 @@ background_loop:
 
 	n_tables_to_drop = row_drop_tables_for_mysql_in_background();
 
+	if (n_tables_to_drop > 0) {
+	        /* Do not monopolize the CPU even if there are tables waiting
+		in the background drop queue. (It is essentially a bug if
+		MySQL tries to drop a table while there are still open handles
+		to it and we had to put it to the background drop queue.) */
+
+		os_thread_sleep(100000);
+	}
+ 
 	srv_main_thread_op_info = (char*)"purging";
 
 	if (srv_fast_shutdown && srv_shutdown_state > 0) {
@@ -3144,6 +3179,13 @@ suspend_thread:
 
 	goto loop;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit.
+	The thread actually never comes here because it is exited in an
+	os_event_wait(). */
+	
+	os_thread_exit(NULL);
+
 #ifndef __WIN__
         return(NULL);
 #else

innobase/srv/srv0start.c

@@ -414,8 +414,10 @@ io_handler_thread(
 	
 	segment = *((ulint*)arg);
 
-/*	printf("Io handler thread %lu starts\n", segment); */
-
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Io handler thread %lu starts\n", segment);
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	for (i = 0;; i++) {
 		fil_aio_wait(segment);
 
@@ -424,6 +426,13 @@ io_handler_thread(
 		mutex_exit(&ios_mutex);
 	}
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit.
+	The thread actually never comes here because it is exited in an
+	os_event_wait(). */
+
+	os_thread_exit(NULL);
+
 #ifndef __WIN__
 	return(NULL);
 #else
@@ -1585,21 +1594,32 @@ innobase_shutdown_for_mysql(void)
 		      os_thread_count);
 	}
 
-	/* 3. Free all InnoDB's own mutexes */
+	/* 3. Free all InnoDB's own mutexes and the os_fast_mutexes inside
+	them */
 
 	sync_close();
 
-	/* 4. Free all OS synchronization primitives (in Windows currently
-	events are not freed) */
+	/* 4. Free the os_conc_mutex and all os_events and os_mutexes */
 
 	srv_free();
 	os_sync_free();
 
-	/* 5. Free all allocated memory (and the os_fast_mutex created in
+	/* 5. Free all allocated memory and the os_fast_mutex created in
 	ut0mem.c */
 
         ut_free_all_mem();
 
+	if (os_thread_count != 0
+	    || os_event_count != 0
+	    || os_mutex_count != 0
+	    || os_fast_mutex_count != 0) {
+	        fprintf(stderr,
+"InnoDB: Warning: some resources were not cleaned up in shutdown:\n"
+"InnoDB: threads %lu, events %lu, os_mutexes %lu, os_fast_mutexes %lu\n",
+		      os_thread_count, os_event_count, os_mutex_count,
+		      os_fast_mutex_count);
+	}
+
 	if (srv_print_verbose_log) {
 	        ut_print_timestamp(stderr);
 	        fprintf(stderr, "  InnoDB: Shutdown completed\n");

mysql-test/r/func_str.result

@@ -64,7 +64,7 @@ concat_ws(NULL,'a')	concat_ws(',',NULL,'')
 NULL	
 select concat_ws(',','',NULL,'a');
 concat_ws(',','',NULL,'a')
-a
+,a
 SELECT CONCAT('"',CONCAT_WS('";"',repeat('a',60),repeat('b',60),repeat('c',60),repeat('d',100)), '"');
 CONCAT('"',CONCAT_WS('";"',repeat('a',60),repeat('b',60),repeat('c',60),repeat('d',100)), '"')
 "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";"bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb";"cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc";"dddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddd"

mysql-test/r/rpl_loaddata.result

@@ -22,3 +22,9 @@ day	id	category	name
 drop table t1;
 drop table t2;
 drop table t3;
+create table t1(a int, b int, unique(b));
+insert into t1 values(1,10);
+load data infile '../../std_data/rpl_loaddata.dat' into table t1;
+show status like 'slave_running';
+Variable_name	Value
+Slave_running	OFF

mysql-test/t/rpl_loaddata.test

@@ -4,6 +4,9 @@
 #
 # check replication of load data for temporary tables with additional parameters
 #
+# check if duplicate entries trigger an error (they should unless IGNORE or
+# REPLACE was used on the master) (bug 571).
+
 source include/master-slave.inc;
 
 create table t1(a int not null auto_increment, b int, primary key(a) );
@@ -27,7 +30,21 @@ connection master;
 drop table t1;
 drop table t2;
 drop table t3;
+create table t1(a int, b int, unique(b));
 
 save_master_pos;
 connection slave;
 sync_with_master;
+insert into t1 values(1,10);
+
+connection master;
+load data infile '../../std_data/rpl_loaddata.dat' into table t1;
+
+save_master_pos;
+connection slave;
+# don't sync_with_master because the slave SQL thread should be stopped because
+# of the error so MASTER_POS_WAIT() will not return; just sleep and hope the
+# slave SQL thread will have had time to stop.
+
+sleep 1;
+show status like 'slave_running';

mysys/thr_alarm.c

@@ -122,12 +122,24 @@ void init_thr_alarm(uint max_alarms)
 
 /*
   Request alarm after sec seconds.
-  A pointer is returned with points to a non-zero int when the alarm has been
-  given. This can't be called from the alarm-handling thread.
-  Returns 0 if no more alarms are allowed (aborted by process)
+
+  SYNOPSIS
+    thr_alarm()
+    alrm		Pointer to alarm detection
+    alarm_data		Structure to store in alarm queue
+
+  NOTES
+    This function can't be called from the alarm-handling thread.
+
+  RETURN VALUES
+    0 ok
+    1 If no more alarms are allowed (aborted by process)
+
+    Stores in first argument a pointer to a non-zero int which is set to 0
+    when the alarm has been given
 */
 
-bool thr_alarm(thr_alarm_t *alrm, uint sec, ALARM *alarm_data)
+my_bool thr_alarm(thr_alarm_t *alrm, uint sec, ALARM *alarm_data)
 {
   ulong now;
   sigset_t old_mask;
@@ -209,7 +221,7 @@ void thr_end_alarm(thr_alarm_t *alarmed)
   ALARM *alarm_data;
   sigset_t old_mask;
   uint i;
-  bool found=0;
+  my_bool found=0;
   DBUG_ENTER("thr_end_alarm");
 
   pthread_sigmask(SIG_BLOCK,&full_signal_set,&old_mask);
@@ -230,10 +242,9 @@ void thr_end_alarm(thr_alarm_t *alarmed)
   DBUG_ASSERT(!*alarmed || found);
   if (!found)
   {
-#ifdef MAIN
-    printf("Warning: Didn't find alarm %lx in queue of %d alarms\n",
-	   (long) *alarmed, alarm_queue.elements);
-#endif
+    if (*alarmed)
+      fprintf(stderr,"Warning: Didn't find alarm %lx in queue of %d alarms\n",
+	      (long) *alarmed, alarm_queue.elements);
     DBUG_PRINT("warning",("Didn't find alarm %lx in queue\n",
 			  (long) *alarmed));
   }

sql/ha_innodb.cc

@@ -131,9 +131,11 @@ static void innobase_print_error(const char* db_errpfx, char* buffer);
 
 /**********************************************************************
 Releases possible search latch and InnoDB thread FIFO ticket. These should
-be released at each SQL statement end. It does no harm to release these
-also in the middle of an SQL statement. */
+be released at each SQL statement end, and also when mysqld passes the
+control to the client. It does no harm to release these also in the middle
+of an SQL statement. */
 static
+inline
 void
 innobase_release_stat_resources(
 /*============================*/
@@ -914,6 +916,11 @@ innobase_commit_low(
 	trx_t*	trx)	/* in: transaction handle */
 {
 #ifdef HAVE_REPLICATION
+	if (trx->conc_state == TRX_NOT_STARTED) {
+
+	        return;
+	}
+
         /* TODO: Guilhem should check if master_log_name, pending
         etc. are right if the master log gets rotated! Possible bug here.
 	Comment by Heikki March 4, 2003. */
@@ -929,11 +936,13 @@ innobase_commit_low(
                                           ));
         }
 #endif /* HAVE_REPLICATION */
+
         trx_commit_for_mysql(trx);
 }
 
 /*********************************************************************
-Commits a transaction in an InnoDB database. */
+Commits a transaction in an InnoDB database or marks an SQL statement
+ended. */
 
 int
 innobase_commit(
@@ -951,29 +960,45 @@ innobase_commit(
   	DBUG_ENTER("innobase_commit");
   	DBUG_PRINT("trans", ("ending transaction"));
 
-	trx = check_trx_exists(thd);
+	/* The flag thd->transaction.all.innodb_active_trans is set to 1
+	in ::external_lock and ::start_stmt, and it is only set to 0 in
+	a commit or a rollback. If it is 0 we know there cannot be resources
+	to be freed and we can return immediately. */
 
-        if (trx->auto_inc_lock) {
-		  	
-		/* If we had reserved the auto-inc lock for
-		some table in this SQL statement, we release it now */
-		  	
-		srv_conc_enter_innodb(trx);
-		row_unlock_table_autoinc_for_mysql(trx);
-		srv_conc_exit_innodb(trx);
+	if (thd->transaction.all.innodb_active_trans == 0) {
+
+	        DBUG_RETURN(0);
 	}
 
-	if (trx_handle != (void*)&innodb_dummy_stmt_trx_handle) {
+	trx = check_trx_exists(thd);
+
+	if (trx_handle != (void*)&innodb_dummy_stmt_trx_handle
+	    || (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))) {
+	        
 		innobase_commit_low(trx);
-		thd->transaction.all.innodb_active_trans=0;
+
+		thd->transaction.all.innodb_active_trans = 0;
+	} else {
+		if (trx->auto_inc_lock) {
+			/* If we had reserved the auto-inc lock for some
+			table in this SQL statement we release it now */
+		  	
+			srv_conc_enter_innodb(trx);
+			row_unlock_table_autoinc_for_mysql(trx);
+			srv_conc_exit_innodb(trx);
+		}
+		/* Store the current undo_no of the transaction so that we
+		know where to roll back if we have to roll back the next
+		SQL statement */
+
+		trx_mark_sql_stat_end(trx);
 	}
 
-	/* Release possible statement level resources */
+	/* Release a possible FIFO ticket and search latch */
 	innobase_release_stat_resources(trx);
-	trx_mark_sql_stat_end(trx);
 
-	/* Tell InnoDB server that there might be work for
-	utility threads: */
+	/* Tell the InnoDB server that there might be work for utility
+	threads: */
 
 	srv_active_wake_master_thread();
 
@@ -1044,7 +1069,7 @@ innobase_commit_complete(
 }
 
 /*********************************************************************
-Rolls back a transaction in an InnoDB database. */
+Rolls back a transaction or the latest SQL statement in an InnoDB database. */
 
 int
 innobase_rollback(
@@ -1085,11 +1110,9 @@ innobase_rollback(
 
 	srv_conc_exit_innodb(trx);
 
-	/* Release possible statement level resources */
+	/* Release a possible FIFO ticket and search latch */
 	innobase_release_stat_resources(trx);
 
-	trx_mark_sql_stat_end(trx);
-
 	DBUG_RETURN(convert_error_code_to_mysql(error, NULL));
 }
 
@@ -3032,6 +3055,8 @@ create_index(
 	KEY*		key;
 	KEY_PART_INFO*	key_part;
 	ulint		ind_type;
+	ulint		col_type;
+	ulint		prefix_len;
   	ulint		i;
 
   	DBUG_ENTER("create_index");
@@ -3059,6 +3084,27 @@ create_index(
 	for (i = 0; i < n_fields; i++) {
 		key_part = key->key_part + i;
 
+		if (key_part->length != key_part->field->pack_length()) {
+		        prefix_len = key_part->length;
+
+			col_type = get_innobase_type_from_mysql_type(
+							key_part->field);
+			if (col_type == DATA_INT
+			    || col_type == DATA_FLOAT
+			    || col_type == DATA_DOUBLE
+			    || col_type == DATA_DECIMAL) {
+			        fprintf(stderr,
+"InnoDB: error: MySQL is trying to create a column prefix index field\n"
+"InnoDB: on an inappropriate data type %lu. Table name %s, column name %s.\n",
+				  col_type, table_name, 
+				  key_part->field->field_name);
+			        
+			        prefix_len = 0;
+			}
+		} else {
+		        prefix_len = 0;
+		}	   
+
 		/* We assume all fields should be sorted in ascending
 		order, hence the '0': */
 		dict_mem_index_add_field(index,
@@ -3600,8 +3646,7 @@ ha_innobase::records_in_range(
 
 /*************************************************************************
 Gives an UPPER BOUND to the number of rows in a table. This is used in
-filesort.cc and its better if the upper bound hold.
-*/
+filesort.cc. */
 
 ha_rows
 ha_innobase::estimate_number_of_rows(void)
@@ -3636,11 +3681,11 @@ ha_innobase::estimate_number_of_rows(void)
 
 	/* Calculate a minimum length for a clustered index record and from
 	that an upper bound for the number of rows. Since we only calculate
-	new statistics in row0mysql.c when a tablehas grown
-        by a threshold factor, we must add a safety factor 2 in front
-	of the formula below. */
+	new statistics in row0mysql.c when a table has grown by a threshold
+	factor, we must add a safety factor 2 in front of the formula below. */
 
-	estimate = 2 * local_data_file_length / dict_index_calc_min_rec_len(index);
+	estimate = 2 * local_data_file_length /
+					 dict_index_calc_min_rec_len(index);
 
 	prebuilt->trx->op_info = (char*)"";
 
@@ -3667,27 +3712,36 @@ ha_innobase::scan_time()
 	return((double) (prebuilt->table->stat_clustered_index_size));
 }
 
-/*
-  Calculate the time it takes to read a set of ranges through and index
-  This enables us to optimise reads for clustered indexes.
-*/
+/**********************************************************************
+Calculate the time it takes to read a set of ranges through an index
+This enables us to optimise reads for clustered indexes. */
 
-double ha_innobase::read_time(uint index, uint ranges, ha_rows rows)
+double
+ha_innobase::read_time(
+/*===================*/
+			/* out: estimated time measured in disk seeks */
+	uint    index,	/* in: key number */
+	uint	ranges,	/* in: how many ranges */
+	ha_rows rows)	/* in: estimated number of rows in the ranges */
 {
-  ha_rows total_rows;
-  double time_for_scan;
-  if (index != table->primary_key)
-    return handler::read_time(index, ranges, rows); // Not clustered
-  if (rows <= 2)
-    return (double) rows;
-  /*
-    Assume that the read is proportional to scan time for all rows + one
-    seek per range.
-  */
-  time_for_scan= scan_time();
-  if ((total_rows= estimate_number_of_rows()) < rows)
-    return time_for_scan;
-  return (ranges + (double) rows / (double) total_rows * time_for_scan);
+	ha_rows total_rows;
+	double  time_for_scan;
+  
+	if (index != table->primary_key)
+	  return handler::read_time(index, ranges, rows); // Not clustered
+
+	if (rows <= 2)
+	  return (double) rows;
+
+	/* Assume that the read time is proportional to the scan time for all
+	rows + at most one seek per range. */
+
+	time_for_scan= scan_time();
+
+	if ((total_rows= estimate_number_of_rows()) < rows)
+	  return time_for_scan;
+
+	return (ranges + (double) rows / (double) total_rows * time_for_scan);
 }
 
 /*************************************************************************
@@ -4040,10 +4094,10 @@ ha_innobase::reset(void)
 }
 
 /**********************************************************************
-Inside LOCK TABLES MySQL will not call external_lock() between SQL
-statements. It will call this function at the start of each SQL statement.
-Note also a spacial case: if a temporary table is created inside LOCK
-TABLES, MySQL has not called external_lock() at all on that table. */
+MySQL calls this function at the start of each SQL statement. Inside LOCK
+TABLES the ::external_lock method does not work to mark SQL statement
+borders. Note also a special case: if a temporary table is created inside
+LOCK TABLES, MySQL has not called external_lock() at all on that table. */
 
 int
 ha_innobase::start_stmt(
@@ -4058,8 +4112,14 @@ ha_innobase::start_stmt(
 
 	trx = prebuilt->trx;
 
+	/* Here we release the search latch and the InnoDB thread FIFO ticket
+	if they were reserved. They should have been released already at the
+	end of the previous statement, but because inside LOCK TABLES the
+	lock count method does not work to mark the end of a SELECT statement,
+	that may not be the case. We MUST release the search latch before an
+	INSERT, for example. */
+
 	innobase_release_stat_resources(trx);
-	trx_mark_sql_stat_end(trx);
 
 	if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
 	    						&& trx->read_view) {
@@ -4082,7 +4142,8 @@ ha_innobase::start_stmt(
 	  
 	        prebuilt->select_lock_type = LOCK_X;
 	}
-
+	
+	/* Set the MySQL flag to mark that there is an active transaction */
 	thd->transaction.all.innodb_active_trans = 1;
 
 	return(0);
@@ -4146,17 +4207,20 @@ ha_innobase::external_lock(
 	}
 
 	if (lock_type != F_UNLCK) {
-		if (trx->n_mysql_tables_in_use == 0) {
-			trx_mark_sql_stat_end(trx);
-		}
+		/* MySQL is setting a new table lock */
 
+		/* Set the MySQL flag to mark that there is an active
+		transaction */
 		thd->transaction.all.innodb_active_trans = 1;
+
 		trx->n_mysql_tables_in_use++;
 		prebuilt->mysql_has_locked = TRUE;
 
-		trx->isolation_level = innobase_map_isolation_level(
+		if (trx->n_mysql_tables_in_use == 1) {
+		        trx->isolation_level = innobase_map_isolation_level(
 						(enum_tx_isolation)
 						thd->variables.tx_isolation);
+		}
 
 		if (trx->isolation_level == TRX_ISO_SERIALIZABLE
 		    && prebuilt->select_lock_type == LOCK_NONE) {
@@ -4172,37 +4236,44 @@ ha_innobase::external_lock(
 
 		  	trx->mysql_n_tables_locked++;
 		}
-	} else {
-		trx->n_mysql_tables_in_use--;
-		prebuilt->mysql_has_locked = FALSE;
-		auto_inc_counter_for_this_stat = 0;
 
-		if (trx->n_mysql_tables_in_use == 0) {
+		DBUG_RETURN(error);
+	}
 
-		  	trx->mysql_n_tables_locked = 0;
+	/* MySQL is releasing a table lock */
 
-			prebuilt->used_in_HANDLER = FALSE;
+	trx->n_mysql_tables_in_use--;
+	prebuilt->mysql_has_locked = FALSE;
+	auto_inc_counter_for_this_stat = 0;
 
-			/* Here we release the search latch and InnoDB
-			thread FIFO ticket if they were reserved. */
+	/* If the MySQL lock count drops to zero we know that the current SQL
+	statement has ended */
 
-			innobase_release_stat_resources(trx);
+	if (trx->n_mysql_tables_in_use == 0) {
 
+	        trx->mysql_n_tables_locked = 0;
+		prebuilt->used_in_HANDLER = FALSE;
+			
+		if (!(thd->options
+				 & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))) {
+			if (thd->transaction.all.innodb_active_trans != 0) {
+		    	        innobase_commit(thd, trx);
+			}
+		} else {
 			if (trx->isolation_level <= TRX_ISO_READ_COMMITTED
 	    						&& trx->read_view) {
 
-	    			/* At low transaction isolation levels we let
+				/* At low transaction isolation levels we let
 				each consistent read set its own snapshot */
 
-	    			read_view_close_for_mysql(trx);
+				read_view_close_for_mysql(trx);
 			}
-			
-		  	if (!(thd->options
-				 & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))) {
-
-		    		innobase_commit(thd, trx);
-		  	}
 		}
+		        
+		/* Here we release the search latch and the InnoDB thread FIFO
+		ticket if they were reserved. */
+
+		innobase_release_stat_resources(trx);
 	}
 
 	DBUG_RETURN(error);
@@ -4513,4 +4584,3 @@ ha_innobase::get_auto_increment()
 }
 
 #endif /* HAVE_INNOBASE_DB */
-

sql/handler.cc

@@ -218,8 +218,12 @@ void ha_close_connection(THD* thd)
 }
 
 /*
-  This is used to commit or rollback a single statement depending
-  on the value of error
+  This is used to commit or rollback a single statement depending on the value
+  of error. Note that if the autocommit is on, then the following call inside
+  InnoDB will commit or rollback the whole transaction (= the statement). The
+  autocommit mechanism built into InnoDB is based on counting locks, but if
+  the user has used LOCK TABLES then that mechanism does not know to do the
+  commit.
 */
 
 int ha_autocommit_or_rollback(THD *thd, int error)

sql/item_strfunc.cc

@@ -515,18 +515,18 @@ String *Item_func_concat_ws::val_str(String *str)
   str->length(0);				// QQ; Should be removed
   res=str;
 
-  // Skip until non-null and non-empty argument is found.
+  // Skip until non-null argument is found.
   // If not, return the empty string
   for (i=0; i < arg_count; i++)
-    if ((res= args[i]->val_str(str)) && res->length())
+    if ((res= args[i]->val_str(str)))
       break;
   if (i ==  arg_count)
     return &empty_string;
 
   for (i++; i < arg_count ; i++)
   {
-    if (!(res2= args[i]->val_str(use_as_buff)) || !res2->length())
-      continue;					// Skip NULL and empty string
+    if (!(res2= args[i]->val_str(use_as_buff)))
+      continue;					// Skip NULL
 
     if (res->length() + sep_str->length() + res2->length() >
 	current_thd->variables.max_allowed_packet)

sql/log_event.cc

@@ -1573,9 +1573,27 @@ int Load_log_event::exec_event(NET* net, struct st_relay_log_info* rli,
     else
     {
       char llbuff[22];
-      enum enum_duplicates handle_dup = DUP_IGNORE;
+      enum enum_duplicates handle_dup;
       if (sql_ex.opt_flags & REPLACE_FLAG)
 	handle_dup= DUP_REPLACE;
+      else if (sql_ex.opt_flags & IGNORE_FLAG)
+        handle_dup= DUP_IGNORE;
+      else
+        /*
+          Note that when replication is running fine, if it was DUP_ERROR on the 
+          master then we could choose DUP_IGNORE here, because if DUP_ERROR
+          suceeded on master, and data is identical on the master and slave,
+          then there should be no uniqueness errors on slave, so DUP_IGNORE is
+          the same as DUP_ERROR. But in the unlikely case of uniqueness errors
+          (because the data on the master and slave happen to be different (user
+          error or bug), we want LOAD DATA to print an error message on the
+          slave to discover the problem.
+
+          If reading from net (a 3.23 master), mysql_load() will change this
+          to DUP_IGNORE.
+        */
+        handle_dup= DUP_ERROR;
+
       sql_exchange ex((char*)fname, sql_ex.opt_flags & DUMPFILE_FLAG);
       String field_term(sql_ex.field_term,sql_ex.field_term_len,log_cs);
       String enclosed(sql_ex.enclosed,sql_ex.enclosed_len,log_cs);
@@ -1637,12 +1655,19 @@ int Load_log_event::exec_event(NET* net, struct st_relay_log_info* rli,
   close_thread_tables(thd);
   if (thd->query_error)
   {
-    int sql_error= thd->net.last_errno;
-    if (!sql_error)
-      sql_error= ER_UNKNOWN_ERROR;
-    slave_print_error(rli,sql_error,
-		      "Error '%s' running LOAD DATA INFILE",
-		      ER_SAFE(sql_error));
+    /* this err/sql_errno code is copy-paste from send_error() */
+    const char *err;
+    int sql_errno;
+    if ((err=thd->net.last_error)[0])
+      sql_errno=thd->net.last_errno;
+    else
+    {
+      sql_errno=ER_UNKNOWN_ERROR;
+      err=ER(sql_errno);       
+    }
+    slave_print_error(rli,sql_errno,
+		      "Error '%s' running load data infile",
+		      err);
     free_root(&thd->mem_root,0);
     return 1;
   }

sql/mysqld.cc

@@ -267,7 +267,7 @@ ulong slave_net_timeout;
 ulong thread_cache_size=0, binlog_cache_size=0, max_binlog_cache_size=0;
 ulong query_cache_size=0;
 ulong com_stat[(uint) SQLCOM_END], com_other;
-ulong bytes_sent, bytes_received;
+ulong bytes_sent, bytes_received, net_big_packet_count;
 ulong refresh_version=1L,flush_version=1L;	/* Increments on each reload */
 ulong query_id, long_query_count, aborted_threads, aborted_connects;
 ulong delayed_insert_timeout, delayed_insert_limit, delayed_queue_size;
@@ -987,14 +987,21 @@ static void set_ports()
 static void set_user(const char *user)
 {
 #if !defined(__WIN__) && !defined(OS2) && !defined(__NETWARE__)
-    struct passwd *ent;
+  struct passwd *ent;
+  uid_t user_id= geteuid();
 
   // don't bother if we aren't superuser
-  if (geteuid())
+  if (user_id)
   {
     if (user)
-      fprintf(stderr,
-	      "Warning: One can only use the --user switch if running as root\n");
+    {
+      /* Don't give a warning, if real user is same as given with --user */
+      struct passwd *user_info= getpwnam(user);
+
+      if (!user_info || user_id != user_info->pw_uid)
+	fprintf(stderr,
+		"Warning: One can only use the --user switch if running as root\n");
+    }
     return;
   }
   else if (!user)

sql/net_serv.cc

@@ -73,11 +73,13 @@ void sql_print_error(const char *format,...);
 #define USE_QUERY_CACHE
 extern uint test_flags;
 extern void query_cache_insert(NET *net, const char *packet, ulong length);
-extern ulong bytes_sent, bytes_received;
+extern ulong bytes_sent, bytes_received, net_big_packet_count;
 extern pthread_mutex_t LOCK_bytes_sent , LOCK_bytes_received;
 #else
 #undef statistic_add
+#undef statistic_increment
 #define statistic_add(A,B,C)
+#define statistic_increment(A,B)
 #endif
 
 #define TEST_BLOCKING		8
@@ -562,7 +564,7 @@ static my_bool net_safe_read(NET *net, char *buff, uint32 length,
     if ((tmp=vio_read(net->vio,(char*) net->buff, length)) <= 0)
     {
       my_bool interrupted = vio_should_retry(net->vio);
-      if (!thr_got_alarm(&alarmed) && interrupted)
+      if (!thr_got_alarm(alarmed) && interrupted)
       {					/* Probably in MIT threads */
 	if (retry_count++ < net->retry_count)
 	  continue;
@@ -596,10 +598,13 @@ static my_bool my_net_skip_rest(NET *net, uint32 remain, thr_alarm_t *alarmed,
   DBUG_ENTER("my_net_skip_rest");
   DBUG_PRINT("enter",("bytes_to_skip: %u", (uint) remain));
 
-  if (!thr_alarm_in_use(&alarmed))
+  /* The following is good for debugging */
+  statistic_increment(net_big_packet_count,&LOCK_bytes_received);
+
+  if (!thr_alarm_in_use(alarmed))
   {
     my_bool old_mode;
-    if (!thr_alarm(alarmed,net->read_timeout, alarm_buff) ||
+    if (thr_alarm(alarmed,net->read_timeout, alarm_buff) ||
 	vio_blocking(net->vio, TRUE, &old_mode) < 0)
       DBUG_RETURN(1);				/* Can't setup, abort */
   }

sql/slave.cc

@@ -287,7 +287,7 @@ int init_relay_log_pos(RELAY_LOG_INFO* rli,const char* log,
       goto err;
     rli->cur_log = &rli->cache_buf;
   }
-  if (pos > BIN_LOG_HEADER_SIZE)
+  if (pos >= BIN_LOG_HEADER_SIZE)
     my_b_seek(rli->cur_log,(off_t)pos);
 
 err:
@@ -2154,8 +2154,13 @@ static int exec_relay_log_event(THD* thd, RELAY_LOG_INFO* rli)
   else
   {
     sql_print_error("\
-Could not parse log event entry, check the master for binlog corruption\n\
-This may also be a network problem, or just a bug in the master or slave code.\
+Could not parse relay log event entry. The possible reasons are: the master's \
+binary log is corrupted (you can check this by running 'mysqlbinlog' on the \
+binary log), the slave's relay log is corrupted (you can check this by running \
+'mysqlbinlog' on the relay log), a network problem, or a bug in the master's \
+or slave's MySQL code. If you want to check the master's binary log or slave's \
+relay log, you will be able to know their names by issuing 'SHOW SLAVE STATUS' \
+on this slave.\
 ");
     return 1;
   }

sql/sql_parse.cc

@@ -891,7 +891,11 @@ pthread_handler_decl(handle_one_connection,arg)
       send_error(thd,net->last_errno,NullS);
       statistic_increment(aborted_threads,&LOCK_status);
     }
-
+    else if (thd->killed)
+    {
+      statistic_increment(aborted_threads,&LOCK_status);
+    }
+    
 end_thread:
     close_connection(thd, 0, 1);
     end_thread(thd,1);
@@ -1068,7 +1072,10 @@ bool do_command(THD *thd)
 		       vio_description(net->vio)));
     /* Check if we can continue without closing the connection */
     if (net->error != 3)
+    {
+      statistic_increment(aborted_threads,&LOCK_status);
       DBUG_RETURN(TRUE);			// We have to close it.
+    }
     send_error(thd,net->last_errno,NullS);
     net->error= 0;
     DBUG_RETURN(FALSE);

sql/sql_repl.cc

@@ -997,7 +997,7 @@ int show_binlog_events(THD* thd)
   {
     LEX_MASTER_INFO *lex_mi = &thd->lex.mi;
     ha_rows event_count, limit_start, limit_end;
-    my_off_t pos = lex_mi->pos;
+    my_off_t pos = max(BIN_LOG_HEADER_SIZE, lex_mi->pos); // user-friendly
     char search_file_name[FN_REFLEN], *name;
     const char *log_file_name = lex_mi->log_file_name;
     pthread_mutex_t *log_lock = mysql_bin_log.get_log_lock();
@@ -1025,12 +1025,6 @@ int show_binlog_events(THD* thd)
     if ((file=open_binlog(&log, linfo.log_file_name, &errmsg)) < 0)
       goto err;
 
-    if (pos < 4)
-    {
-      errmsg = "Invalid log position";
-      goto err;
-    }
-
     pthread_mutex_lock(log_lock);
     my_b_seek(&log, pos);
 

sql/sql_yacc.yy

@@ -804,6 +804,18 @@ master_def:
        MASTER_LOG_POS_SYM EQ ulonglong_num
        {
 	 Lex->mi.pos = $3;
+         /* 
+            If the user specified a value < BIN_LOG_HEADER_SIZE, adjust it
+            instead of causing subsequent errors. 
+            We need to do it in this file, because only there we know that 
+            MASTER_LOG_POS has been explicitely specified. On the contrary
+            in change_master() (sql_repl.cc) we cannot distinguish between 0
+            (MASTER_LOG_POS explicitely specified as 0) and 0 (unspecified),
+            whereas we want to distinguish (specified 0 means "read the binlog
+            from 0" (4 in fact), unspecified means "don't change the position
+            (keep the preceding value)").
+         */
+         Lex->mi.pos = max(BIN_LOG_HEADER_SIZE, Lex->mi.pos);
        }
        |
        MASTER_CONNECT_RETRY_SYM EQ ULONG_NUM
@@ -819,6 +831,8 @@ master_def:
        RELAY_LOG_POS_SYM EQ ULONG_NUM
        {
 	 Lex->mi.relay_log_pos = $3;
+         /* Adjust if < BIN_LOG_HEADER_SIZE (same comment as Lex->mi.pos) */
+         Lex->mi.relay_log_pos = max(BIN_LOG_HEADER_SIZE, Lex->mi.relay_log_pos);
        }
        ;
 

sql/unireg.h

@@ -141,10 +141,13 @@
  */
 #define MIN_TURBOBM_PATTERN_LEN 3
 
-/* Defines for binary logging */
-
-#define BIN_LOG_HEADER_SIZE	4
+/* 
+   Defines for binary logging.
+   Do not decrease the value of BIN_LOG_HEADER_SIZE.
+   Do not even increase it before checking code.
+*/
 
+#define BIN_LOG_HEADER_SIZE    4 
 #define FLOATING_POINT_BUFFER 331
 
 	/* Include prototypes for unireg */

support-files/mysql.server.sh

@@ -19,7 +19,7 @@
 # Required-Start: $local_fs $network $remote_fs
 # Required-Stop: $local_fs $network $remote_fs
 # Default-Start:  2 3 4 5
-# Default-Stop: 2 3 4 5
+# Default-Stop: 0 1 6
 # Short-Description: start and stop MySQL
 # Description: MySQL is a very fast and reliable SQL database engine.
 ### END INIT INFO

support-files/mysql.spec.sh

@@ -254,6 +254,13 @@ export PATH
 # Build the 4.0 Max binary (includes BDB and UDFs and therefore
 # cannot be linked statically against the patched glibc)
 
+# If we want to compile with RAID using gcc 3, we need to use
+# gcc instead of g++ to avoid linking problems (RAID code is written in C++)
+if gcc -v 2>&1 | grep 'version 3' > /dev/null 2>&1
+then
+	export CXX="gcc"
+fi
+
 BuildMySQL "--enable-shared \
 		--with-openssl \
 		--with-berkeley-db \
@@ -319,6 +326,10 @@ install -m644 $MBD/sql/mysqld.sym $RBR/usr/lib/mysql/mysqld.sym
 install -m644 $MBD/support-files/mysql-log-rotate $RBR/etc/logrotate.d/mysql
 install -m755 $MBD/support-files/mysql.server $RBR/etc/init.d/mysql
 
+# Create a symlink "rcmysql", pointing to the init.script. SuSE users
+# will appreciate that, as all services usually offer this.
+ln -s ../../sbin/init.d/mysql $RPM_BUILD_ROOT/usr/sbin/rcmysql
+
 # Create symbolic compatibility link safe_mysqld -> mysqld_safe
 # (safe_mysqld will be gone in MySQL 4.1)
 ln -sf ./mysqld_safe $RBR/usr/bin/safe_mysqld
@@ -463,6 +474,7 @@ fi
 %attr(755, root, root) /usr/bin/safe_mysqld
 
 %attr(755, root, root) /usr/sbin/mysqld
+%attr(755, root, root) /usr/sbin/rcmysql
 %attr(644, root, root) /usr/lib/mysql/mysqld.sym
 
 %attr(644, root, root) /etc/logrotate.d/mysql