branches/zip: Merge revisions 1009:1165 from trunk.

dict/dict0dict.c

@@ -794,16 +794,20 @@ dict_init(void)
 }
 
 /**************************************************************************
-Returns a table object. NOTE! This is a high-level function to be used
+Returns a table object and optionally increment its MySQL open handle count.
-mainly from outside the 'dict' directory. Inside this directory
+NOTE! This is a high-level function to be used mainly from outside the
-dict_table_get_low is usually the appropriate function. */
+'dict' directory. Inside this directory dict_table_get_low is usually the
+appropriate function. */
 
 dict_table_t*
 dict_table_get(
 /*===========*/
 					/* out: table, NULL if
 					does not exist */
-	const char*	table_name)	/* in: table name */
+	const char*	table_name,	/* in: table name */
+	ibool		inc_mysql_count)
+     					/* in: whether to increment the open
+					handle count on the table */
 {
 	dict_table_t*	table;
 
@@ -811,42 +815,17 @@ dict_table_get(
 
 	table = dict_table_get_low(table_name);
 
-	mutex_exit(&(dict_sys->mutex));
+	if (inc_mysql_count && table) {
-
-	if (table != NULL) {
-		if (!table->stat_initialized) {
-			dict_update_statistics(table);
-		}
-	}
-
-	return(table);
-}
-
-/**************************************************************************
-Returns a table object and increments MySQL open handle count on the table. */
-
-dict_table_t*
-dict_table_get_and_increment_handle_count(
-/*======================================*/
-					/* out: table, NULL if
-					does not exist */
-	const char*	table_name)	/* in: table name */
-{
-	dict_table_t*	table;
-
-	mutex_enter(&(dict_sys->mutex));
-
-	table = dict_table_get_low(table_name);
-
-	if (table != NULL) {
-
 		table->n_mysql_handles_opened++;
 	}
 
 	mutex_exit(&(dict_sys->mutex));
 
 	if (table != NULL) {
-		if (!table->stat_initialized && !table->ibd_file_missing) {
+		if (!table->stat_initialized) {
+			/* If table->ibd_file_missing == TRUE, this will
+			print an error message and return without doing
+			anything. */
 			dict_update_statistics(table);
 		}
 	}

handler/ha_innodb.cc

@@ -40,9 +40,6 @@ have disables the InnoDB inlining in this file. */
 #include <myisampack.h>
 #include <mysys_err.h>
 #include <my_sys.h>
-
-#define MAX_ULONG_BIT ((ulong) 1 << (sizeof(ulong)*8-1))
-
 #include "ha_innodb.h"
 
 pthread_mutex_t innobase_share_mutex,	/* to protect innobase_open_files */
@@ -1257,18 +1254,6 @@ trx_is_interrupted(
 	return(trx && trx->mysql_thd && ((THD*) trx->mysql_thd)->killed);
 }
 
-/**************************************************************************
-Obtain a pointer to the MySQL THD object, as in current_thd().	This
-definition must match the one in sql/ha_innodb.cc! */
-extern "C"
-void*
-innobase_current_thd(void)
-/*======================*/
-			/* out: MySQL THD object */
-{
-	return(current_thd);
-}
-
 /*********************************************************************
 Call this when you have opened a new table handle in HANDLER, before you
 call index_read_idx() etc. Actually, we can let the cursor stay open even
@@ -2328,7 +2313,7 @@ ha_innobase::open(
 
 	/* Get pointer to a table object in InnoDB dictionary cache */
 
-	ib_table = dict_table_get_and_increment_handle_count(norm_name);
+	ib_table = dict_table_get(norm_name, TRUE);
 
 	if (NULL == ib_table) {
 		ut_print_timestamp(stderr);
@@ -2490,7 +2475,7 @@ get_field_offset(
 /******************************************************************
 Checks if a field in a record is SQL NULL. Uses the record format
 information in table to track the null bit in record. */
-inline
+static inline
 uint
 field_in_record_is_null(
 /*====================*/
@@ -4915,7 +4900,7 @@ ha_innobase::create(
 
 	log_buffer_flush_to_disk();
 
-	innobase_table = dict_table_get(norm_name);
+	innobase_table = dict_table_get(norm_name, FALSE);
 
 	DBUG_ASSERT(innobase_table != 0);
 
@@ -5526,16 +5511,10 @@ ha_innobase::info(
 		prebuilt->trx->op_info = (char*)
 					  "returning various info to MySQL";
 
-		if (ib_table->space != 0) {
+		my_snprintf(path, sizeof(path), "%s/%s%s",
-			my_snprintf(path, sizeof(path), "%s/%s%s",
-				mysql_data_home, ib_table->name, ".ibd");
-			unpack_filename(path,path);
-		} else {
-			my_snprintf(path, sizeof(path), "%s/%s%s",
 				mysql_data_home, ib_table->name, reg_ext);
 
-			unpack_filename(path,path);
+		unpack_filename(path,path);
-		}
 
 		/* Note that we do not know the access time of the table,
 		nor the CHECK TABLE time, nor the UPDATE or INSERT time. */
@@ -7344,7 +7323,6 @@ innobase_get_at_most_n_mbchars(
 }
 }
 
-extern "C" {
 /**********************************************************************
 This function returns true if
 
@@ -7354,33 +7332,34 @@ is either REPLACE or LOAD DATA INFILE REPLACE.
 2) SQL-query in the current thread
 is INSERT ON DUPLICATE KEY UPDATE.
 
-NOTE that /mysql/innobase/row/row0ins.c must contain the
+NOTE that storage/innobase/row/row0ins.c must contain the
 prototype for this function ! */
-
+extern "C"
 ibool
 innobase_query_is_update(void)
 /*==========================*/
 {
-	THD*	thd;
+	THD*	thd = current_thd;
 
-	thd = (THD *)innobase_current_thd();
+	if (!thd) {
+		/* InnoDB's internal threads may run InnoDB stored procedures
+		that call this function. Then current_thd is not defined
+		(it is probably NULL). */
 
-	if (thd->lex->sql_command == SQLCOM_REPLACE ||
+		return(FALSE);
-		thd->lex->sql_command == SQLCOM_REPLACE_SELECT ||
-		(thd->lex->sql_command == SQLCOM_LOAD &&
-			thd->lex->duplicates == DUP_REPLACE)) {
-
-		return(1);
 	}
 
-	if (thd->lex->sql_command == SQLCOM_INSERT &&
+	switch (thd->lex->sql_command) {
-		thd->lex->duplicates  == DUP_UPDATE) {
+	case SQLCOM_REPLACE:
-
+	case SQLCOM_REPLACE_SELECT:
-		return(1);
+		return(TRUE);
+	case SQLCOM_LOAD:
+		return(thd->lex->duplicates == DUP_REPLACE);
+	case SQLCOM_INSERT:
+		return(thd->lex->duplicates == DUP_UPDATE);
+	default:
+		return(FALSE);
 	}
-
-	return(0);
-}
 }
 
 /***********************************************************************

include/btr0sea.h

@@ -183,7 +183,7 @@ struct btr_search_struct{
 				number of full fields */
 	ulint	n_bytes;	/* recommended prefix: number of bytes in
 				an incomplete field;
-				cf. BTR_PAGE_MAX_REC_SIZE */
+				see also BTR_PAGE_MAX_REC_SIZE */
 	ibool	left_side;	/* TRUE or FALSE, depending on whether
 				the leftmost record of several records with
 				the same prefix should be indexed in the

include/data0type.ic

@@ -65,8 +65,8 @@ dtype_get_mblen(
 		innobase_get_cset_width(dtype_get_charset_coll(prtype),
 					mbminlen, mbmaxlen);
 		ut_ad(*mbminlen <= *mbmaxlen);
-		ut_ad(*mbminlen <= 2); /* cf. the bit-field in dtype_t */
+		ut_ad(*mbminlen <= 2); /* mbminlen in dtype_t is 0..3 */
-		ut_ad(*mbmaxlen < 1 << 3); /* cf. the bit-field in dtype_t */
+		ut_ad(*mbmaxlen < 1 << 3); /* mbmaxlen in dtype_t is 0..7 */
 #else /* !UNIV_HOTBACKUP */
 		ut_a(mtype <= DATA_BINARY);
 		*mbminlen = *mbmaxlen = 1;

include/dict0dict.h

@@ -333,26 +333,20 @@ dict_foreign_parse_drop_constraints(
 	const char***	constraints_to_drop);	/* out: id's of the
 						constraints to drop */
 /**************************************************************************
-Returns a table object. NOTE! This is a high-level function to be used
+Returns a table object and optionally increment its MySQL open handle count.
-mainly from outside the 'dict' directory. Inside this directory
+NOTE! This is a high-level function to be used mainly from outside the
-dict_table_get_low is usually the appropriate function. */
+'dict' directory. Inside this directory dict_table_get_low is usually the
+appropriate function. */
 
 dict_table_t*
 dict_table_get(
 /*===========*/
 					/* out: table, NULL if
 					does not exist */
-	const char*	table_name);	/* in: table name */
+	const char*	table_name,	/* in: table name */
-/**************************************************************************
+	ibool		inc_mysql_count);
-Returns a table object and increments MySQL open handle count on the table.
+					/* in: whether to increment the open
-*/
+					handle count on the table */
-
-dict_table_t*
-dict_table_get_and_increment_handle_count(
-/*======================================*/
-					/* out: table, NULL if
-					does not exist */
-	const char*	table_name);	/* in: table name */
 /**************************************************************************
 Returns a table object based on table id. */
 

include/sync0sync.h

@@ -516,12 +516,6 @@ to 20 microseconds. */
 
 #define	SYNC_SPIN_ROUNDS	srv_n_spin_wait_rounds
 
-#define SYNC_INFINITE_TIME	((ulint)(-1))
-
-/* Means that a timeout elapsed when waiting */
-
-#define SYNC_TIME_EXCEEDED	(ulint)1
-
 /* The number of system calls made in this module. Intended for performance
 monitoring. */
 

include/univ.i

@@ -77,25 +77,37 @@ memory is read outside the allocated blocks. */
 /* Make a non-inline debug version */
 
 #if 0
-#define UNIV_DEBUG_VALGRIND
+#define UNIV_DEBUG_VALGRIND			/* Enable extra
-#define UNIV_DEBUG_PRINT
+						Valgrind instrumentation */
-#define UNIV_BUF_DEBUG
+#define UNIV_DEBUG_PRINT			/* Enable the compilation of
-#define UNIV_DEBUG
+						some debug print functions */
-#define UNIV_DEBUG_FILE_ACCESSES
+#define UNIV_BUF_DEBUG				/* Enable buffer pool
-#define UNIV_LIST_DEBUG
+						debugging without UNIV_DEBUG */
-#define UNIV_MEM_DEBUG
+#define UNIV_DEBUG				/* Enable ut_ad() assertions */
-#define UNIV_IBUF_DEBUG
+#define UNIV_DEBUG_FILE_ACCESSES		/* Debug .ibd file access
-#define UNIV_SYNC_DEBUG
+						(field file_page_was_freed
-#define UNIV_SEARCH_DEBUG
+						in buf_page_t) */
-#define UNIV_SYNC_PERF_STAT
+#define UNIV_LIST_DEBUG				/* debug UT_LIST_ macros */
-#define UNIV_SEARCH_PERF_STAT
+#define UNIV_MEM_DEBUG				/* detect memory leaks etc */
-#define UNIV_SRV_PRINT_LATCH_WAITS
+#define UNIV_IBUF_DEBUG				/* debug the insert buffer;
-#define UNIV_BTR_PRINT
+this limits the database to IBUF_COUNT_N_SPACES and IBUF_COUNT_N_PAGES,
+and the insert buffer must be empty when the database is started */
+#define UNIV_SYNC_DEBUG				/* debug mutex and latch
+operations (very slow); also UNIV_DEBUG must be defined */
+#define UNIV_SEARCH_DEBUG			/* debug B-tree comparisons */
+#define UNIV_SYNC_PERF_STAT			/* operation counts for
+						rw-locks and mutexes */
+#define UNIV_SEARCH_PERF_STAT			/* statistics for the
+						adaptive hash index */
+#define UNIV_SRV_PRINT_LATCH_WAITS		/* enable diagnostic output
+						in sync0sync.c */
+#define UNIV_BTR_PRINT				/* enable functions for
+						printing B-trees */
 #define UNIV_ZIP_DEBUG
 #endif
 
-#define UNIV_BTR_DEBUG
+#define UNIV_BTR_DEBUG				/* check B-tree links */
-#define UNIV_LIGHT_MEM_DEBUG
+#define UNIV_LIGHT_MEM_DEBUG			/* light memory debugging */
 
 #ifdef HAVE_purify
 /* The following sets all new allocated memory to zero before use:

include/ut0ut.h

@@ -127,7 +127,8 @@ ulint
 ut_2_power_up(
 /*==========*/
 			/* out: first power of 2 which is >= n */
-	ulint	n);	/* in: number != 0 */
+	ulint	n)	/* in: number != 0 */
+	__attribute__((const));
 /****************************************************************
 Sort function for ulint arrays. */
 

log/log0recv.c

@@ -34,6 +34,7 @@ Created 9/20/1997 Heikki Tuuri
 #include "btr0cur.h"
 #include "dict0boot.h"
 #include "fil0fil.h"
+#include "sync0sync.h"
 
 #ifdef UNIV_HOTBACKUP
 /* This is set to FALSE if the backup was originally taken with the
@@ -191,6 +192,7 @@ recv_sys_empty_hash(void)
 	recv_sys->addr_hash = hash_create(buf_pool_get_curr_size() / 256);
 }
 
+#ifndef UNIV_LOG_DEBUG
 /************************************************************
 Frees the recovery system. */
 static
@@ -210,6 +212,7 @@ recv_sys_free(void)
 
 	mutex_exit(&(recv_sys->mutex));
 }
+#endif /* UNIV_LOG_DEBUG */
 
 /************************************************************
 Truncates possible corrupted or extra records from a log group. */
@@ -2887,6 +2890,15 @@ recv_recovery_from_checkpoint_finish(void)
 #ifndef UNIV_LOG_DEBUG
 	recv_sys_free();
 #endif
+
+#ifdef UNIV_SYNC_DEBUG
+	/* Wait for a while so that created threads have time to suspend
+	themselves before we switch the latching order checks on */
+	os_thread_sleep(1000000);
+
+	/* Switch latching order checks on in sync0sync.c */
+	sync_order_checks_on = TRUE;
+#endif
 	if (srv_force_recovery < SRV_FORCE_NO_TRX_UNDO) {
 		/* Rollback the uncommitted transactions which have no user
 		session */

que/que0que.c

@@ -335,6 +335,8 @@ que_fork_start_command(
 	que_fork_t*	fork)	/* in: a query fork */
 {
 	que_thr_t*	thr;
+	que_thr_t*	suspended_thr = NULL;
+	que_thr_t*	completed_thr = NULL;
 
 	fork->state = QUE_FORK_ACTIVE;
 
@@ -344,14 +346,18 @@ que_fork_start_command(
 	but in a parallelized select, which necessarily is non-scrollable,
 	there may be several to choose from */
 
-	/*---------------------------------------------------------------
+	/* First we try to find a query thread in the QUE_THR_COMMAND_WAIT
-	First we try to find a query thread in the QUE_THR_COMMAND_WAIT state
+	state. Then we try to find a query thread in the QUE_THR_SUSPENDED
-	*/
+	state, finally we try to find a query thread in the QUE_THR_COMPLETED
+	state */
 
 	thr = UT_LIST_GET_FIRST(fork->thrs);
 
-	while (thr != NULL) {
+	/* We make a single pass over the thr list within which we note which
-		if (thr->state == QUE_THR_COMMAND_WAIT) {
+	threads are ready to run. */
+	while (thr) {
+		switch (thr->state) {
+		case QUE_THR_COMMAND_WAIT:
 
 			/* We have to send the initial message to query thread
 			to start it */
@@ -359,49 +365,44 @@ que_fork_start_command(
 			que_thr_init_command(thr);
 
 			return(thr);
-		}
 
-		ut_ad(thr->state != QUE_THR_LOCK_WAIT);
+		case QUE_THR_SUSPENDED:
-
-		thr = UT_LIST_GET_NEXT(thrs, thr);
-	}
-
-	/*----------------------------------------------------------------
-	Then we try to find a query thread in the QUE_THR_SUSPENDED state */
-
-	thr = UT_LIST_GET_FIRST(fork->thrs);
-
-	while (thr != NULL) {
-		if (thr->state == QUE_THR_SUSPENDED) {
 			/* In this case the execution of the thread was
 			suspended: no initial message is needed because
 			execution can continue from where it was left */
+			if (!suspended_thr) {
+				suspended_thr = thr;
+			}
 
-			que_thr_move_to_run_state(thr);
+			break;
+
+		case QUE_THR_COMPLETED:
+			if (!completed_thr) {
+				completed_thr = thr;
+			}
+
+			break;
+
+		case QUE_THR_LOCK_WAIT:
+			ut_error;
 
-			return(thr);
 		}
 
 		thr = UT_LIST_GET_NEXT(thrs, thr);
 	}
 
-	/*-----------------------------------------------------------------
+	if (suspended_thr) {
-	Then we try to find a query thread in the QUE_THR_COMPLETED state */
 
-	thr = UT_LIST_GET_FIRST(fork->thrs);
+		thr = suspended_thr;
+		que_thr_move_to_run_state(thr);
 
-	while (thr != NULL) {
+	} else if (completed_thr) {
-		if (thr->state == QUE_THR_COMPLETED) {
-			que_thr_init_command(thr);
 
-			return(thr);
+		thr = completed_thr;
-		}
+		que_thr_init_command(thr);
-
-		thr = UT_LIST_GET_NEXT(thrs, thr);
 	}
 
-	/* Else we return NULL */
+	return(thr);
-	return(NULL);
 }
 
 /**************************************************************************

row/row0ins.c

@@ -1535,7 +1535,8 @@ row_ins_check_foreign_constraints(
 		if (foreign->foreign_index == index) {
 
 			if (foreign->referenced_table == NULL) {
-				dict_table_get(foreign->referenced_table_name);
+				dict_table_get(foreign->referenced_table_name,
+					       FALSE);
 			}
 
 			if (0 == trx->dict_operation_lock_mode) {

row/row0sel.c

@@ -4506,7 +4506,7 @@ row_search_check_if_query_cache_permitted(
 	dict_table_t*	table;
 	ibool		ret	= FALSE;
 
-	table = dict_table_get(norm_name);
+	table = dict_table_get(norm_name, FALSE);
 
 	if (table == NULL) {
 

row/row0upd.c

@@ -203,7 +203,8 @@ row_upd_check_references_constraints(
 				foreign->n_fields))) {
 
 			if (foreign->foreign_table == NULL) {
-				dict_table_get(foreign->foreign_table_name);
+				dict_table_get(foreign->foreign_table_name,
+					       FALSE);
 			}
 
 			if (foreign->foreign_table) {

srv/srv0start.c

@@ -1558,17 +1558,6 @@ innobase_start_or_create_for_mysql(void)
 	srv_was_started = TRUE;
 	srv_is_being_started = FALSE;
 
-#ifdef UNIV_SYNC_DEBUG
-	/* Wait a while so that the created threads have time to suspend
-	themselves before we switch sync debugging on; otherwise a thread may
-	execute mutex_enter() before the checks are on, and mutex_exit() after
-	the checks are on, which will cause an assertion failure in sync
-	debug. */
-
-	os_thread_sleep(3000000);
-	sync_order_checks_on = TRUE;
-#endif
-
 	if (trx_doublewrite == NULL) {
 		/* Create the doublewrite buffer to a new tablespace */
 

trx/trx0roll.c

@@ -131,8 +131,26 @@ trx_rollback_for_mysql(
 
 	trx->op_info = "rollback";
 
-	err = trx_general_rollback_for_mysql(trx, FALSE, NULL);
+	/* If we are doing the XA recovery of prepared transactions, then
+	the transaction object does not have an InnoDB session object, and we
+	set a dummy session that we use for all MySQL transactions. */
 
+	mutex_enter(&kernel_mutex);
+
+	if (trx->sess == NULL) {
+		/* Open a dummy session */
+
+		if (!trx_dummy_sess) {
+			trx_dummy_sess = sess_open();
+		}
+
+		trx->sess = trx_dummy_sess;
+	}
+
+	mutex_exit(&kernel_mutex);
+
+	err = trx_general_rollback_for_mysql(trx, FALSE, NULL);
+	
 	trx->op_info = "";
 
 	return(err);

trx/trx0trx.c

@@ -1595,6 +1595,24 @@ trx_commit_for_mysql(
 
 	trx->op_info = "committing";
 
+	/* If we are doing the XA recovery of prepared transactions, then
+	the transaction object does not have an InnoDB session object, and we
+	set the dummy session that we use for all MySQL transactions. */
+
+	mutex_enter(&kernel_mutex);
+
+	if (trx->sess == NULL) {
+		/* Open a dummy session */
+
+		if (!trx_dummy_sess) {
+			trx_dummy_sess = sess_open();
+		}
+
+		trx->sess = trx_dummy_sess;
+	}
+	
+	mutex_exit(&kernel_mutex);
+
 	trx_start_if_not_started(trx);
 
 	mutex_enter(&kernel_mutex);