Merge

23 files changed, 528 insertions, 243 deletions

diff --git a/storage/innobase/btr/btr0btr.c b/storage/innobase/btr/btr0btr.c
index 2d84586216a..c27fb73ff8d 100644
--- a/storage/innobase/btr/btr0btr.c
+++ b/storage/innobase/btr/btr0btr.c
@@ -143,7 +143,7 @@ btr_root_get(
 	root_page_no = dict_tree_get_page(tree);
 
 	root = btr_page_get(space, root_page_no, RW_X_LATCH, mtr);
-	ut_a(!!page_is_comp(root) ==
+	ut_a((ibool)!!page_is_comp(root) ==
 			UT_LIST_GET_FIRST(tree->tree_indexes)->table->comp);
 	
 	return(root);
@@ -2014,7 +2014,7 @@ btr_compress(
 	page = btr_cur_get_page(cursor);
 	tree = btr_cur_get_tree(cursor);
 	comp = page_is_comp(page);
-	ut_a(!!comp == cursor->index->table->comp);
+	ut_a((ibool)!!comp == cursor->index->table->comp);
 
 	ut_ad(mtr_memo_contains(mtr, dict_tree_get_lock(tree),
 							MTR_MEMO_X_LOCK));
@@ -2508,7 +2508,7 @@ btr_index_rec_validate(
 	        return(TRUE);
 	}
 
-	if (UNIV_UNLIKELY(!!page_is_comp(page) != index->table->comp)) {
+	if (UNIV_UNLIKELY((ibool)!!page_is_comp(page) != index->table->comp)) {
 		btr_index_rec_validate_report(page, rec, index);
 		fprintf(stderr, "InnoDB: compact flag=%lu, should be %lu\n",
 			(ulong) !!page_is_comp(page),
diff --git a/storage/innobase/btr/btr0cur.c b/storage/innobase/btr/btr0cur.c
index 4ae27f007d6..f81cce5b8e9 100644
--- a/storage/innobase/btr/btr0cur.c
+++ b/storage/innobase/btr/btr0cur.c
@@ -316,7 +316,9 @@ btr_cur_search_to_nth_level(
 	if (btr_search_latch.writer == RW_LOCK_NOT_LOCKED
 		&& latch_mode <= BTR_MODIFY_LEAF && info->last_hash_succ
 		&& !estimate
+#ifdef PAGE_CUR_LE_OR_EXTENDS
 		&& mode != PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
 		&& srv_use_adaptive_hash_indexes
 	        && btr_search_guess_on_hash(index, info, tuple, mode,
 						latch_mode, cursor,
@@ -390,9 +392,12 @@ btr_cur_search_to_nth_level(
 		page_mode = PAGE_CUR_LE;
 		break;
 	default:
-		ut_ad(mode == PAGE_CUR_L
-			|| mode == PAGE_CUR_LE
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+		ut_ad(mode == PAGE_CUR_L || mode == PAGE_CUR_LE
 			|| mode == PAGE_CUR_LE_OR_EXTENDS);
+#else /* PAGE_CUR_LE_OR_EXTENDS */
+		ut_ad(mode == PAGE_CUR_L || mode == PAGE_CUR_LE);
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
 		page_mode = mode;
 		break;
 	}
@@ -507,7 +512,7 @@ retry_page_get:
 				/* x-latch the page */
 				page = btr_page_get(space,
 						page_no, RW_X_LATCH, mtr);
-				ut_a(!!page_is_comp(page)
+				ut_a((ibool)!!page_is_comp(page)
 						== index->table->comp);
 			}
 
@@ -1385,7 +1390,7 @@ btr_cur_parse_update_in_place(
 		goto func_exit;
 	}
 
-	ut_a(!!page_is_comp(page) == index->table->comp);
+	ut_a((ibool)!!page_is_comp(page) == index->table->comp);
 	rec = page + rec_offset;
 	
 	/* We do not need to reserve btr_search_latch, as the page is only
diff --git a/storage/innobase/configure.in b/storage/innobase/configure.in
index baf11272ab9..c56bd8274c4 100644
--- a/storage/innobase/configure.in
+++ b/storage/innobase/configure.in
@@ -117,6 +117,13 @@ case "$target" in
 	 CFLAGS="$CFLAGS -DUNIV_MUST_NOT_INLINE";;
 esac
 
+# must go in pair with AR as set by MYSQL_CHECK_AR
+if test -z "$ARFLAGS"
+then
+  ARFLAGS="cru"
+fi
+AC_SUBST(ARFLAGS)
+
 AC_OUTPUT(Makefile os/Makefile ut/Makefile btr/Makefile dnl
 		buf/Makefile data/Makefile dnl
 		dict/Makefile dyn/Makefile dnl
diff --git a/storage/innobase/fil/fil0fil.c b/storage/innobase/fil/fil0fil.c
index 299b55f3d2b..20f522c1a60 100644
--- a/storage/innobase/fil/fil0fil.c
+++ b/storage/innobase/fil/fil0fil.c
@@ -3410,9 +3410,9 @@ fil_extend_space_to_desired_size(
 	fil_space_t*	space;
 	byte*		buf2;
 	byte*		buf;
+	ulint		buf_size;
 	ulint		start_page_no;
 	ulint		file_start_page_no;
-	ulint		n_pages;
 	ulint		offset_high;
 	ulint		offset_low;
 	ibool		success		= TRUE;
@@ -3437,22 +3437,20 @@ fil_extend_space_to_desired_size(
 
 	fil_node_prepare_for_io(node, system, space);
 
-	/* Extend 1 MB at a time */
-
-	buf2 = mem_alloc(1024 * 1024 + UNIV_PAGE_SIZE);
-	buf = ut_align(buf2, UNIV_PAGE_SIZE);
-
-	memset(buf, '\0', 1024 * 1024);
-
 	start_page_no = space->size;
 	file_start_page_no = space->size - node->size;
 
-	while (start_page_no < size_after_extend) {	
-		n_pages = size_after_extend - start_page_no;
+	/* Extend at most 64 pages at a time */
+	buf_size = ut_min(64, size_after_extend - start_page_no)
+				* UNIV_PAGE_SIZE;
+	buf2 = mem_alloc(buf_size + UNIV_PAGE_SIZE);
+	buf = ut_align(buf2, UNIV_PAGE_SIZE);
 
-		if (n_pages > (1024 * 1024) / UNIV_PAGE_SIZE) {
-			n_pages = (1024 * 1024) / UNIV_PAGE_SIZE;
-		}
+	memset(buf, 0, buf_size);
+
+	while (start_page_no < size_after_extend) {
+		ulint	n_pages = ut_min(buf_size / UNIV_PAGE_SIZE,
+				size_after_extend - start_page_no);
 
 		offset_high = (start_page_no - file_start_page_no)
 				/ (4096 * ((1024 * 1024) / UNIV_PAGE_SIZE));
@@ -4034,7 +4032,7 @@ fil_aio_wait(
 	if (os_aio_use_native_aio) {
 		srv_set_io_thread_op_info(segment, "native aio handle");
 #ifdef WIN_ASYNC_IO
-		ret = os_aio_windows_handle(segment, 0, (void**) &fil_node,
+		ret = os_aio_windows_handle(segment, 0, &fil_node,
 					    &message, &type);
 #elif defined(POSIX_ASYNC_IO)
 		ret = os_aio_posix_handle(segment, &fil_node, &message);
diff --git a/storage/innobase/ibuf/ibuf0ibuf.c b/storage/innobase/ibuf/ibuf0ibuf.c
index 712d43f916c..d7fa48b6e66 100644
--- a/storage/innobase/ibuf/ibuf0ibuf.c
+++ b/storage/innobase/ibuf/ibuf0ibuf.c
@@ -2810,7 +2810,7 @@ ibuf_insert_to_index_page(
 	ut_ad(ibuf_inside());
 	ut_ad(dtuple_check_typed(entry));
 
-	if (UNIV_UNLIKELY(index->table->comp != !!page_is_comp(page))) {
+	if (UNIV_UNLIKELY(index->table->comp != (ibool)!!page_is_comp(page))) {
 		fputs(
 "InnoDB: Trying to insert a record from the insert buffer to an index page\n"
 "InnoDB: but the 'compact' flag does not match!\n", stderr);
diff --git a/storage/innobase/include/os0file.h b/storage/innobase/include/os0file.h
index e75281dd93c..362e3552411 100644
--- a/storage/innobase/include/os0file.h
+++ b/storage/innobase/include/os0file.h
@@ -373,7 +373,7 @@ os_file_get_size_as_iblonglong(
 				/* out: size in bytes, -1 if error */
 	os_file_t	file);	/* in: handle to a file */
 /***************************************************************************
-Sets a file size. This function can be used to extend or truncate a file. */
+Write the specified number of zeros to a newly created file. */
 
 ibool
 os_file_set_size(
diff --git a/storage/innobase/include/page0cur.h b/storage/innobase/include/page0cur.h
index e89e740e775..b03302b0e77 100644
--- a/storage/innobase/include/page0cur.h
+++ b/storage/innobase/include/page0cur.h
@@ -26,11 +26,13 @@ Created 10/4/1994 Heikki Tuuri
 #define	PAGE_CUR_GE	2
 #define	PAGE_CUR_L	3
 #define	PAGE_CUR_LE	4
-#define PAGE_CUR_LE_OR_EXTENDS 5 /* This is a search mode used in
+/*#define PAGE_CUR_LE_OR_EXTENDS 5*/ /* This is a search mode used in
 				 "column LIKE 'abc%' ORDER BY column DESC";
 				 we have to find strings which are <= 'abc' or
 				 which extend it */
-#define	PAGE_CUR_DBG	6
+#ifdef UNIV_SEARCH_DEBUG
+# define PAGE_CUR_DBG	6	/* As PAGE_CUR_LE, but skips search shortcut */
+#endif /* UNIV_SEARCH_DEBUG */
 
 #ifdef PAGE_CUR_ADAPT
 # ifdef UNIV_SEARCH_PERF_STAT
diff --git a/storage/innobase/include/row0mysql.h b/storage/innobase/include/row0mysql.h
index 4bb9fa63cd1..4e6ff73b0f8 100644
--- a/storage/innobase/include/row0mysql.h
+++ b/storage/innobase/include/row0mysql.h
@@ -243,17 +243,27 @@ row_update_for_mysql(
 					the MySQL format */
 	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
 					handle */
-
 /*************************************************************************
-Does an unlock of a row for MySQL. */
+This can only be used when srv_locks_unsafe_for_binlog is TRUE. Before
+calling this function we must use trx_reset_new_rec_lock_info() and
+trx_register_new_rec_lock() to store the information which new record locks
+really were set. This function removes a newly set lock under prebuilt->pcur,
+and also under prebuilt->clust_pcur. Currently, this is only used and tested
+in the case of an UPDATE or a DELETE statement, where the row lock is of the
+LOCK_X type.
+Thus, this implements a 'mini-rollback' that releases the latest record
+locks we set. */
 
 int
 row_unlock_for_mysql(
 /*=================*/
 					/* out: error code or DB_SUCCESS */
-	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
 					handle */
-
+	ibool		has_latches_on_recs);/* TRUE if called so that we have
+					the latches on the records under pcur
+					and clust_pcur, and we do not need to
+					reposition the cursors. */
 /*************************************************************************
 Creates an query graph node of 'update' type to be used in the MySQL
 interface. */
diff --git a/storage/innobase/include/srv0srv.h b/storage/innobase/include/srv0srv.h
index 6e4241965c1..116ae7b6438 100644
--- a/storage/innobase/include/srv0srv.h
+++ b/storage/innobase/include/srv0srv.h
@@ -182,6 +182,7 @@ extern mutex_t*	kernel_mutex_temp;/* mutex protecting the server, trx structs,
 #define kernel_mutex (*kernel_mutex_temp)
 
 #define SRV_MAX_N_IO_THREADS	100
+#define SRV_CONCURRENCY_THRESHOLD	20
 				
 /* Array of English strings describing the current state of an
 i/o handler thread */
diff --git a/storage/innobase/include/trx0trx.h b/storage/innobase/include/trx0trx.h
index 8df50d6703d..146730d46f8 100644
--- a/storage/innobase/include/trx0trx.h
+++ b/storage/innobase/include/trx0trx.h
@@ -16,10 +16,39 @@ Created 3/26/1996 Heikki Tuuri
 #include "que0types.h"
 #include "mem0mem.h"
 #include "read0types.h"
+#include "dict0types.h"
 #include "trx0xa.h"
 
 extern ulint	trx_n_mysql_transactions;
 
+/*****************************************************************
+Resets the new record lock info in a transaction struct. */
+UNIV_INLINE
+void
+trx_reset_new_rec_lock_info(
+/*========================*/
+	trx_t*	trx);	/* in: transaction struct */
+/*****************************************************************
+Registers that we have set a new record lock on an index. We only have space
+to store 2 indexes! If this is called to store more than 2 indexes after
+trx_reset_new_rec_lock_info(), then this function does nothing. */
+UNIV_INLINE
+void
+trx_register_new_rec_lock(
+/*======================*/
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index);	/* in: trx sets a new record lock on this
+				index */
+/*****************************************************************
+Checks if trx has set a new record lock on an index. */
+UNIV_INLINE
+ibool
+trx_new_rec_locks_contain(
+/*======================*/
+				/* out: TRUE if trx has set a new record lock
+				on index */
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index);	/* in: index */
 /************************************************************************
 Releases the search latch if trx has reserved it. */
 
@@ -495,8 +524,18 @@ struct trx_struct{
 	lock_t*		auto_inc_lock;	/* possible auto-inc lock reserved by
 					the transaction; note that it is also
 					in the lock list trx_locks */
-	ibool		trx_create_lock;/* this is TRUE if we have created a
-					new lock for a record accessed */
+	dict_index_t*	new_rec_locks[2];/* these are normally NULL; if
+					srv_locks_unsafe_for_binlog is TRUE,
+					in a cursor search, if we set a new
+					record lock on an index, this is set
+					to point to the index; this is
+					used in releasing the locks under the
+					cursors if we are performing an UPDATE
+					and we determine after retrieving
+					the row that it does not need to be
+					locked; thus, these can be used to
+					implement a 'mini-rollback' that
+					releases the latest record locks */
 	UT_LIST_NODE_T(trx_t)
 			trx_list;	/* list of transactions */
 	UT_LIST_NODE_T(trx_t)
diff --git a/storage/innobase/include/trx0trx.ic b/storage/innobase/include/trx0trx.ic
index 78e5acda148..54cf2ff331f 100644
--- a/storage/innobase/include/trx0trx.ic
+++ b/storage/innobase/include/trx0trx.ic
@@ -39,4 +39,60 @@ trx_start_if_not_started_low(
 	}
 }
 
+/*****************************************************************
+Resets the new record lock info in a transaction struct. */
+UNIV_INLINE
+void
+trx_reset_new_rec_lock_info(
+/*========================*/
+	trx_t*	trx)	/* in: transaction struct */
+{
+	trx->new_rec_locks[0] = NULL;
+	trx->new_rec_locks[1] = NULL;
+}
+
+/*****************************************************************
+Registers that we have set a new record lock on an index. We only have space
+to store 2 indexes! If this is called to store more than 2 indexes after
+trx_reset_new_rec_lock_info(), then this function does nothing. */
+UNIV_INLINE
+void
+trx_register_new_rec_lock(
+/*======================*/
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index)	/* in: trx sets a new record lock on this
+				index */
+{
+	if (trx->new_rec_locks[0] == NULL) {
+		trx->new_rec_locks[0] = index;
+
+		return;
+	}
+
+	if (trx->new_rec_locks[0] == index) {
 
+		return;
+	}
+
+	if (trx->new_rec_locks[1] != NULL) {
+
+		return;
+	}
+
+	trx->new_rec_locks[1] = index;
+}
+
+/*****************************************************************
+Checks if trx has set a new record lock on an index. */
+UNIV_INLINE
+ibool
+trx_new_rec_locks_contain(
+/*======================*/
+				/* out: TRUE if trx has set a new record lock
+				on index */
+	trx_t*		trx,	/* in: transaction struct */
+	dict_index_t*	index)	/* in: index */
+{
+	return(trx->new_rec_locks[0] == index
+	       || trx->new_rec_locks[1] == index);
+}
diff --git a/storage/innobase/lock/lock0lock.c b/storage/innobase/lock/lock0lock.c
index 48db7ced0cb..280c4871ee9 100644
--- a/storage/innobase/lock/lock0lock.c
+++ b/storage/innobase/lock/lock0lock.c
@@ -956,7 +956,7 @@ lock_rec_has_to_wait(
 		cause waits */
 
 		if ((lock_is_on_supremum || (type_mode & LOCK_GAP))
-			&& !(type_mode & LOCK_INSERT_INTENTION)) {
+		    && !(type_mode & LOCK_INSERT_INTENTION)) {
 
 			/* Gap type locks without LOCK_INSERT_INTENTION flag
 			do not need to wait for anything. This is because 
@@ -1765,10 +1765,7 @@ lock_rec_create(
 	lock_rec_set_nth_bit(lock, heap_no);
 
 	HASH_INSERT(lock_t, hash, lock_sys->rec_hash,
-					lock_rec_fold(space, page_no), lock); 
-	/* Note that we have create a new lock */
-	trx->trx_create_lock = TRUE;
-
+					lock_rec_fold(space, page_no), lock);
 	if (type_mode & LOCK_WAIT) {
 
 		lock_set_lock_and_trx_wait(lock, trx);
@@ -1945,15 +1942,6 @@ lock_rec_add_to_queue(
 
 	if (similar_lock && !somebody_waits && !(type_mode & LOCK_WAIT)) {
 
-		/* If the nth bit of a record lock is already set then we
-		do not set a new lock bit, otherwice we set */
-
-		if (lock_rec_get_nth_bit(similar_lock, heap_no)) {
-			trx->trx_create_lock = FALSE;
-		} else {
-			trx->trx_create_lock = TRUE;
-		}
-
 		lock_rec_set_nth_bit(similar_lock, heap_no);
 
 		return(similar_lock);
@@ -2005,11 +1993,14 @@ lock_rec_lock_fast(
 	lock = lock_rec_get_first_on_page(rec);
 
 	trx = thr_get_trx(thr);
-	trx->trx_create_lock = FALSE;
 
 	if (lock == NULL) {
 		if (!impl) {
 			lock_rec_create(mode, rec, index, trx);
+			
+			if (srv_locks_unsafe_for_binlog) {
+				trx_register_new_rec_lock(trx, index);
+			}
 		}
 		
 		return(TRUE);
@@ -2021,23 +2012,22 @@ lock_rec_lock_fast(
 	}
 
 	if (lock->trx != trx
-				|| lock->type_mode != (mode | LOCK_REC)
-				|| lock_rec_get_n_bits(lock) <= heap_no) {
+	    || lock->type_mode != (mode | LOCK_REC)
+	    || lock_rec_get_n_bits(lock) <= heap_no) {
+
 	    	return(FALSE);
 	}
 
 	if (!impl) {
+		/* If the nth bit of the record lock is already set then we
+		do not set a new lock bit, otherwise we do set */
 
-		/* If the nth bit of a record lock is already set then we
-		do not set a new lock bit, otherwice we set */
-
-		if (lock_rec_get_nth_bit(lock, heap_no)) {
-			trx->trx_create_lock = FALSE;
-		} else {
-			trx->trx_create_lock = TRUE;
+		if (!lock_rec_get_nth_bit(lock, heap_no)) {
+			lock_rec_set_nth_bit(lock, heap_no);
+			if (srv_locks_unsafe_for_binlog) {
+				trx_register_new_rec_lock(trx, index);
+			}
 		}
-
-		lock_rec_set_nth_bit(lock, heap_no);
 	}
 
 	return(TRUE);
@@ -2093,12 +2083,19 @@ lock_rec_lock_slow(
 		enough already granted on the record, we have to wait. */
     				
 		err = lock_rec_enqueue_waiting(mode, rec, index, thr);
+
+		if (srv_locks_unsafe_for_binlog) {
+			trx_register_new_rec_lock(trx, index);
+		}
 	} else {
 		if (!impl) {
 			/* Set the requested lock on the record */
 
 			lock_rec_add_to_queue(LOCK_REC | mode, rec, index,
 									trx);
+			if (srv_locks_unsafe_for_binlog) {
+				trx_register_new_rec_lock(trx, index);
+			}
 		}
 
 		err = DB_SUCCESS;
@@ -2436,8 +2433,15 @@ lock_rec_inherit_to_gap(
 	
 	lock = lock_rec_get_first(rec);
 
+	/* If srv_locks_unsafe_for_binlog is TRUE, we do not want locks set
+	by an UPDATE or a DELETE to be inherited as gap type locks. But we
+	DO want S-locks set by a consistency constraint to be inherited also
+	then. */
+
 	while (lock != NULL) {
-		if (!lock_rec_get_insert_intention(lock)) {
+		if (!lock_rec_get_insert_intention(lock)
+		    && !(srv_locks_unsafe_for_binlog
+			 && lock_get_mode(lock) == LOCK_X)) {
 			
 			lock_rec_add_to_queue(LOCK_REC | lock_get_mode(lock)
 						| LOCK_GAP,
@@ -3069,7 +3073,7 @@ lock_update_insert(
 	lock_rec_inherit_to_gap_if_gap_lock(rec, page_rec_get_next(rec));
 
 	lock_mutex_exit_kernel();
-}	
+}
 
 /*****************************************************************
 Updates the lock table when a record is removed. */
diff --git a/storage/innobase/log/log0recv.c b/storage/innobase/log/log0recv.c
index 8d9780bfbda..42e854398ba 100644
--- a/storage/innobase/log/log0recv.c
+++ b/storage/innobase/log/log0recv.c
@@ -768,7 +768,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_REC_INSERT: case MLOG_COMP_REC_INSERT:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 				type == MLOG_COMP_REC_INSERT, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = page_cur_parse_insert_rec(FALSE, ptr, end_ptr,
 							index, page, mtr);
 		}
@@ -776,7 +777,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_REC_CLUST_DELETE_MARK: case MLOG_COMP_REC_CLUST_DELETE_MARK:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 			type == MLOG_COMP_REC_CLUST_DELETE_MARK, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = btr_cur_parse_del_mark_set_clust_rec(ptr,
 						end_ptr, index, page);
 		}
@@ -796,7 +798,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_REC_UPDATE_IN_PLACE: case MLOG_COMP_REC_UPDATE_IN_PLACE:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 			type == MLOG_COMP_REC_UPDATE_IN_PLACE, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = btr_cur_parse_update_in_place(ptr, end_ptr,
 							page, index);
 		}
@@ -806,7 +809,8 @@ recv_parse_or_apply_log_rec_body(
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 			type == MLOG_COMP_LIST_END_DELETE
 			|| type == MLOG_COMP_LIST_START_DELETE, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = page_parse_delete_rec_list(type, ptr, end_ptr,
 							index, page, mtr);
 		}
@@ -814,7 +818,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_LIST_END_COPY_CREATED: case MLOG_COMP_LIST_END_COPY_CREATED:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 			type == MLOG_COMP_LIST_END_COPY_CREATED, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = page_parse_copy_rec_list_to_created_page(ptr,
 						end_ptr, index, page, mtr);
 		}
@@ -822,7 +827,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_PAGE_REORGANIZE: case MLOG_COMP_PAGE_REORGANIZE:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 				type == MLOG_COMP_PAGE_REORGANIZE, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = btr_parse_page_reorganize(ptr, end_ptr, index,
 								page, mtr);
 		}
@@ -855,7 +861,8 @@ recv_parse_or_apply_log_rec_body(
 	case MLOG_REC_DELETE: case MLOG_COMP_REC_DELETE:
 		if (NULL != (ptr = mlog_parse_index(ptr, end_ptr,
 				type == MLOG_COMP_REC_DELETE, &index))) {
-			ut_a(!page||!!page_is_comp(page)==index->table->comp);
+			ut_a(!page
+			  || (ibool)!!page_is_comp(page)==index->table->comp);
 			ptr = page_cur_parse_delete_rec(ptr, end_ptr,
 							index, page, mtr);
 		}
diff --git a/storage/innobase/os/os0file.c b/storage/innobase/os/os0file.c
index c68f5738798..48dc808e36c 100644
--- a/storage/innobase/os/os0file.c
+++ b/storage/innobase/os/os0file.c
@@ -1653,7 +1653,7 @@ os_file_get_size_as_iblonglong(
 }
 
 /***************************************************************************
-Sets a file size. This function can be used to extend or truncate a file. */
+Write the specified number of zeros to a newly created file. */
 
 ibool
 os_file_set_size(
@@ -1666,44 +1666,46 @@ os_file_set_size(
 				size */
 	ulint		size_high)/* in: most significant 32 bits of size */
 {
-	ib_longlong	offset;
-	ib_longlong	low;
-	ulint   	n_bytes;
+	ib_longlong	current_size;
+	ib_longlong	desired_size;
 	ibool		ret;
 	byte*   	buf;
 	byte*   	buf2;
+	ulint		buf_size;
 
 	ut_a(size == (size & 0xFFFFFFFF));
 
-	/* We use a very big 8 MB buffer in writing because Linux may be
-	extremely slow in fsync on 1 MB writes */
+	current_size = 0;
+	desired_size = (ib_longlong)size + (((ib_longlong)size_high) << 32);
 
-	buf2 = ut_malloc(UNIV_PAGE_SIZE * 513);
+	/* Write up to 1 megabyte at a time. */
+	buf_size = ut_min(64, (ulint) (desired_size / UNIV_PAGE_SIZE))
+							* UNIV_PAGE_SIZE;
+	buf2 = ut_malloc(buf_size + UNIV_PAGE_SIZE);
 
 	/* Align the buffer for possible raw i/o */
 	buf = ut_align(buf2, UNIV_PAGE_SIZE);
 
 	/* Write buffer full of zeros */
-	memset(buf, 0, UNIV_PAGE_SIZE * 512);
+	memset(buf, 0, buf_size);
 
-	offset = 0;
-	low = (ib_longlong)size + (((ib_longlong)size_high) << 32);
-
-	if (low >= (ib_longlong)(100 * 1024 * 1024)) {
+	if (desired_size >= (ib_longlong)(100 * 1024 * 1024)) {
 				
 		fprintf(stderr, "InnoDB: Progress in MB:");
 	}
 
-	while (offset < low) {
-	        if (low - offset < UNIV_PAGE_SIZE * 512) {
-	        	n_bytes = (ulint)(low - offset);
-	        } else {
-	        	n_bytes = UNIV_PAGE_SIZE * 512;
-	        }
-	  
+	while (current_size < desired_size) {
+		ulint	n_bytes;
+
+		if (desired_size - current_size < (ib_longlong) buf_size) {
+			n_bytes = (ulint) (desired_size - current_size);
+		} else {
+			n_bytes = buf_size;
+		}
+
 	        ret = os_file_write(name, file, buf,
-				(ulint)(offset & 0xFFFFFFFF),
-				    (ulint)(offset >> 32),
+				(ulint)(current_size & 0xFFFFFFFF),
+				    (ulint)(current_size >> 32),
 				n_bytes);
 	        if (!ret) {
 			ut_free(buf2);
@@ -1711,18 +1713,18 @@ os_file_set_size(
 	        }
 				
 		/* Print about progress for each 100 MB written */
-		if ((ib_longlong) (offset + n_bytes) / (ib_longlong)(100 * 1024 * 1024)
-		    != offset / (ib_longlong)(100 * 1024 * 1024)) {
+		if ((current_size + n_bytes) / (ib_longlong)(100 * 1024 * 1024)
+		    != current_size / (ib_longlong)(100 * 1024 * 1024)) {
 
 		        fprintf(stderr, " %lu00",
-				(ulong) ((offset + n_bytes)
+				(ulong) ((current_size + n_bytes)
 					/ (ib_longlong)(100 * 1024 * 1024)));
 		}
 		
-	        offset += n_bytes;
+	        current_size += n_bytes;
 	}
 
-	if (low >= (ib_longlong)(100 * 1024 * 1024)) {
+	if (desired_size >= (ib_longlong)(100 * 1024 * 1024)) {
 				
 		fprintf(stderr, "\n");
 	}
@@ -3296,7 +3298,7 @@ os_aio(
 	ibool		retval;
 	BOOL		ret		= TRUE;
 	DWORD		len		= (DWORD) n;
-	void*		dummy_mess1;
+	struct fil_node_struct * dummy_mess1;
 	void*		dummy_mess2;
 	ulint		dummy_type;
 #endif
diff --git a/storage/innobase/page/page0cur.c b/storage/innobase/page/page0cur.c
index df6d898d4ac..d0b89e81787 100644
--- a/storage/innobase/page/page0cur.c
+++ b/storage/innobase/page/page0cur.c
@@ -47,7 +47,6 @@ page_cur_try_search_shortcut(
 				not yet completely matched */
 	page_cur_t*	cursor) /* out: page cursor */ 
 {
-	int	cmp;
 	rec_t*	rec;
 	rec_t*	next_rec;
 	ulint	low_match;
@@ -79,9 +78,8 @@ page_cur_try_search_shortcut(
 	up_match = low_match;
 	up_bytes = low_bytes;
 
-	cmp = page_cmp_dtuple_rec_with_match(tuple, rec, offsets, &low_match,
-								&low_bytes);
-	if (cmp == -1) {
+	if (page_cmp_dtuple_rec_with_match(tuple, rec, offsets,
+						&low_match, &low_bytes) < 0) {
 		goto exit_func;
 	}
 
@@ -89,9 +87,8 @@ page_cur_try_search_shortcut(
 	offsets = rec_get_offsets(next_rec, index, offsets,
 				dtuple_get_n_fields(tuple), &heap);
 
-	cmp = page_cmp_dtuple_rec_with_match(tuple, next_rec, offsets,
-						&up_match, &up_bytes);
-	if (cmp != -1) {
+	if (page_cmp_dtuple_rec_with_match(tuple, next_rec, offsets,
+						&up_match, &up_bytes) >= 0) {
 		goto exit_func;
 	}
 
@@ -115,7 +112,7 @@ page_cur_try_search_shortcut(
 	ut_a(*ilow_matched_fields == low_match);
 	ut_a(*ilow_matched_bytes == low_bytes);
 #endif
-	if (next_rec != page_get_supremum_rec(page)) {
+	if (!page_rec_is_supremum(next_rec)) {
 
 		*iup_matched_fields = up_match;
 		*iup_matched_bytes = up_bytes;
@@ -137,6 +134,7 @@ exit_func:
 
 #endif
 
+#ifdef PAGE_CUR_LE_OR_EXTENDS
 /********************************************************************
 Checks if the nth field in a record is a character type field which extends
 the nth field in tuple, i.e., the field is longer or equal in length and has
@@ -185,6 +183,7 @@ page_cur_rec_field_extends(
 
         return(FALSE);
 }
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
 
 /********************************************************************
 Searches the right position for a page cursor. */
@@ -239,10 +238,17 @@ page_cur_search_with_match(
 	      && ilow_matched_fields && ilow_matched_bytes && cursor);
 	ut_ad(dtuple_validate(tuple));
 	ut_ad(dtuple_check_typed(tuple));
+#ifdef UNIV_DEBUG
+# ifdef PAGE_CUR_DBG
+	if (mode != PAGE_CUR_DBG)
+# endif /* PAGE_CUR_DBG */
+# ifdef PAGE_CUR_LE_OR_EXTENDS
+		if (mode != PAGE_CUR_LE_OR_EXTENDS)
+# endif /* PAGE_CUR_LE_OR_EXTENDS */
 	ut_ad((mode == PAGE_CUR_L) || (mode == PAGE_CUR_LE)
-	      || (mode == PAGE_CUR_G) || (mode == PAGE_CUR_GE)
-	      || (mode == PAGE_CUR_LE_OR_EXTENDS) || (mode == PAGE_CUR_DBG));
-	      
+	      || (mode == PAGE_CUR_G) || (mode == PAGE_CUR_GE));
+#endif /* UNIV_DEBUG */
+
 	page_check_dir(page);
 
 #ifdef PAGE_CUR_ADAPT
@@ -261,16 +267,18 @@ page_cur_search_with_match(
 	    		return;
 	    	}
 	}
-/*#ifdef UNIV_SEARCH_DEBUG */
+# ifdef PAGE_CUR_DBG
 	if (mode == PAGE_CUR_DBG) {
 		mode = PAGE_CUR_LE;
 	}
-/*#endif */
+# endif
 #endif	
 
 	/* The following flag does not work for non-latin1 char sets because
 	cmp_full_field does not tell how many bytes matched */
+#ifdef PAGE_CUR_LE_OR_EXTENDS
 	ut_a(mode != PAGE_CUR_LE_OR_EXTENDS); 
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
 
 	/* If mode PAGE_CUR_G is specified, we are trying to position the
 	cursor to answer a query of the form "tuple < X", where tuple is
@@ -308,33 +316,36 @@ page_cur_search_with_match(
 		cmp = cmp_dtuple_rec_with_match(tuple, mid_rec, offsets,
 						&cur_matched_fields,
 						&cur_matched_bytes);
-		if (cmp == 1) {
+		if (UNIV_LIKELY(cmp > 0)) {
+low_slot_match:
 			low = mid;
 			low_matched_fields = cur_matched_fields;
 			low_matched_bytes = cur_matched_bytes;
 
-		} else if (cmp == -1) {
+		} else if (UNIV_LIKELY(cmp /* == -1 */)) {
+#ifdef PAGE_CUR_LE_OR_EXTENDS
 			if (mode == PAGE_CUR_LE_OR_EXTENDS
 			    && page_cur_rec_field_extends(tuple, mid_rec,
 						offsets, cur_matched_fields)) {
-				low = mid;
-				low_matched_fields = cur_matched_fields;
-				low_matched_bytes = cur_matched_bytes;
-			} else {
-				up = mid;
-				up_matched_fields = cur_matched_fields;
-				up_matched_bytes = cur_matched_bytes;
-			}
 
-		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
-			   || mode == PAGE_CUR_LE_OR_EXTENDS) {
-			low = mid;
-			low_matched_fields = cur_matched_fields;
-			low_matched_bytes = cur_matched_bytes;
-		} else {
+				goto low_slot_match;
+			}
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+up_slot_match:
 			up = mid;
 			up_matched_fields = cur_matched_fields;
 			up_matched_bytes = cur_matched_bytes;
+
+		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			   || mode == PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+		) {
+
+			goto low_slot_match;
+		} else {
+
+			goto up_slot_match;
 		}
    	}
 
@@ -360,32 +371,35 @@ page_cur_search_with_match(
 		cmp = cmp_dtuple_rec_with_match(tuple, mid_rec, offsets,
 						&cur_matched_fields,
 						&cur_matched_bytes);
-		if (cmp == 1) {
+		if (UNIV_LIKELY(cmp > 0)) {
+low_rec_match:
 			low_rec = mid_rec;
 			low_matched_fields = cur_matched_fields;
 			low_matched_bytes = cur_matched_bytes;
 
-		} else if (cmp == -1) {
+		} else if (UNIV_LIKELY(cmp /* == -1 */)) {
+#ifdef PAGE_CUR_LE_OR_EXTENDS
 			if (mode == PAGE_CUR_LE_OR_EXTENDS
 			    && page_cur_rec_field_extends(tuple, mid_rec,
 						offsets, cur_matched_fields)) {
-				low_rec = mid_rec;
-				low_matched_fields = cur_matched_fields;
-				low_matched_bytes = cur_matched_bytes;
-			} else {
-				up_rec = mid_rec;
-				up_matched_fields = cur_matched_fields;
-				up_matched_bytes = cur_matched_bytes;
+
+				goto low_rec_match;
 			}
-		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
-			   || mode == PAGE_CUR_LE_OR_EXTENDS) {
-			low_rec = mid_rec;
-			low_matched_fields = cur_matched_fields;
-			low_matched_bytes = cur_matched_bytes;
-		} else {
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+up_rec_match:
 			up_rec = mid_rec;
 			up_matched_fields = cur_matched_fields;
 			up_matched_bytes = cur_matched_bytes;
+		} else if (mode == PAGE_CUR_G || mode == PAGE_CUR_LE
+#ifdef PAGE_CUR_LE_OR_EXTENDS
+			   || mode == PAGE_CUR_LE_OR_EXTENDS
+#endif /* PAGE_CUR_LE_OR_EXTENDS */
+		) {
+
+			goto low_rec_match;
+		} else {
+
+			goto up_rec_match;
 		}
    	}
 
diff --git a/storage/innobase/page/page0page.c b/storage/innobase/page/page0page.c
index 1fe7f1d9356..7e09cdf073e 100644
--- a/storage/innobase/page/page0page.c
+++ b/storage/innobase/page/page0page.c
@@ -483,7 +483,7 @@ page_copy_rec_list_end_no_locks(
 		page_cur_move_to_next(&cur1);
 	}
 
-	ut_a(!!page_is_comp(new_page) == index->table->comp);
+	ut_a((ibool)!!page_is_comp(new_page) == index->table->comp);
 	ut_a(page_is_comp(new_page) == page_is_comp(page));
 	ut_a(mach_read_from_2(new_page + UNIV_PAGE_SIZE - 10) == (ulint)
 		(page_is_comp(new_page)
@@ -1347,7 +1347,7 @@ page_print_list(
 	ulint*		offsets		= offsets_;
 	*offsets_ = (sizeof offsets_) / sizeof *offsets_;
 
-	ut_a(!!page_is_comp(page) == index->table->comp);
+	ut_a((ibool)!!page_is_comp(page) == index->table->comp);
 
 	fprintf(stderr,
 		"--------------------------------\n"
@@ -1741,7 +1741,7 @@ page_validate(
 	ulint*		offsets		= NULL;
 	ulint*		old_offsets	= NULL;
 
-	if (!!comp != index->table->comp) {
+	if ((ibool)!!comp != index->table->comp) {
 		fputs("InnoDB: 'compact format' flag mismatch\n", stderr);
 		goto func_exit2;
 	}
diff --git a/storage/innobase/rem/rem0rec.c b/storage/innobase/rem/rem0rec.c
index 580a7bfe509..fbc33aea669 100644
--- a/storage/innobase/rem/rem0rec.c
+++ b/storage/innobase/rem/rem0rec.c
@@ -601,30 +601,38 @@ rec_set_nth_field_extern_bit_new(
 
 	/* read the lengths of fields 0..n */
 	for (i = 0; i < n_fields; i++) {
-		ibool	is_null;
-		ulint	len;
 		field = dict_index_get_nth_field(index, i);
 		type = dict_col_get_type(dict_field_get_col(field));
-		is_null = !(dtype_get_prtype(type) & DATA_NOT_NULL);
-		if (is_null) {
-			/* nullable field => read the null flag */
-			is_null = !!(*nulls & null_mask);
+		if (!(dtype_get_prtype(type) & DATA_NOT_NULL)) {
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				/* NULL fields cannot be external. */
+				ut_ad(i != ith);
+				continue;
+			}
+
 			null_mask <<= 1;
-			if (null_mask == 0x100)
-				nulls--, null_mask = 1;
 		}
-		if (is_null || field->fixed_len) {
-			/* No length (or extern bit) is stored for
-			fields that are NULL or fixed-length. */
+		if (field->fixed_len) {
+			/* fixed-length fields cannot be external
+			(Fixed-length fields longer than
+			DICT_MAX_COL_PREFIX_LEN will be treated as
+			variable-length ones in dict_index_add_col().) */
 			ut_ad(i != ith);
 			continue;
 		}
-		len = *lens--;
+		lens--;
 		if (dtype_get_len(type) > 255
 				|| dtype_get_mtype(type) == DATA_BLOB) {
+			ulint	len = lens[1];
 			if (len & 0x80) { /* 1exxxxxx: 2-byte length */
 				if (i == ith) {
-					if (!val == !(len & 0x20)) {
+					if (!val == !(len & 0x40)) {
 						return; /* no change */
 					}
 					/* toggle the extern bit */
@@ -823,6 +831,7 @@ rec_convert_dtuple_to_rec_new(
 	byte*		lens;
 	ulint		len;
 	ulint		i;
+	ulint		n_node_ptr_field;
 	ulint		fixed_len;
 	ulint		null_mask	= 1;
 	const ulint	n_fields	= dtuple_get_n_fields(dtuple);
@@ -831,16 +840,26 @@ rec_convert_dtuple_to_rec_new(
 	ut_ad(index->table->comp);
 
 	ut_ad(n_fields > 0);
-	switch (status) {
+
+	/* Try to ensure that the memset() between the for() loops
+	completes fast.  The address is not exact, but UNIV_PREFETCH
+	should never generate a memory fault. */
+	UNIV_PREFETCH_RW(rec - REC_N_NEW_EXTRA_BYTES - n_fields);
+	UNIV_PREFETCH_RW(rec);
+
+	switch (UNIV_EXPECT(status, REC_STATUS_ORDINARY)) {
 	case REC_STATUS_ORDINARY:
 		ut_ad(n_fields <= dict_index_get_n_fields(index));
+		n_node_ptr_field = ULINT_UNDEFINED;
 		break;
 	case REC_STATUS_NODE_PTR:
 		ut_ad(n_fields == dict_index_get_n_unique_in_tree(index) + 1);
+		n_node_ptr_field = n_fields - 1;
 		break;
 	case REC_STATUS_INFIMUM:
 	case REC_STATUS_SUPREMUM:
 		ut_ad(n_fields == 1);
+		n_node_ptr_field = ULINT_UNDEFINED;
 		goto init;
 	default:
 		ut_a(0);
@@ -852,15 +871,18 @@ rec_convert_dtuple_to_rec_new(
 	rec += (index->n_nullable + 7) / 8;
 
 	for (i = 0; i < n_fields; i++) {
+		if (UNIV_UNLIKELY(i == n_node_ptr_field)) {
+#ifdef UNIV_DEBUG
+			field = dtuple_get_nth_field(dtuple, i);
+			type = dfield_get_type(field);
+			ut_ad(dtype_get_prtype(type) & DATA_NOT_NULL);
+			ut_ad(dfield_get_len(field) == 4);
+#endif /* UNIV_DEBUG */
+			goto init;
+		}
 		field = dtuple_get_nth_field(dtuple, i);
 		type = dfield_get_type(field);
 		len = dfield_get_len(field);
-		if (status == REC_STATUS_NODE_PTR && i == n_fields - 1) {
-			fixed_len = 4;
-			ut_ad(dtype_get_prtype(type) & DATA_NOT_NULL);
-			ut_ad(len == 4);
-			continue;
-		}
 		fixed_len = dict_index_get_nth_field(index, i)->fixed_len;
 
 		if (!(dtype_get_prtype(type) & DATA_NOT_NULL)) {
@@ -902,27 +924,33 @@ init:
 		type = dfield_get_type(field);
 		len = dfield_get_len(field);
 
-		if (status == REC_STATUS_NODE_PTR && i == n_fields - 1) {
-			fixed_len = 4;
+		if (UNIV_UNLIKELY(i == n_node_ptr_field)) {
 			ut_ad(dtype_get_prtype(type) & DATA_NOT_NULL);
 			ut_ad(len == 4);
-			goto copy;
+			memcpy(end, dfield_get_data(field), len);
+			break;
 		}
 		fixed_len = dict_index_get_nth_field(index, i)->fixed_len;
 
 		if (!(dtype_get_prtype(type) & DATA_NOT_NULL)) {
 			/* nullable field */
 			ut_ad(index->n_nullable > 0);
+
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
+				null_mask = 1;
+			}
+
 			ut_ad(*nulls < null_mask);
+
 			/* set the null flag if necessary */
 			if (len == UNIV_SQL_NULL) {
 				*nulls |= null_mask;
+				null_mask <<= 1;
+				continue;
 			}
+
 			null_mask <<= 1;
-			if (null_mask == 0x100)
-				nulls--, null_mask = 1;
-			if (len == UNIV_SQL_NULL)
-				continue;
 		}
 		/* only nullable fields can be null */
 		ut_ad(len != UNIV_SQL_NULL);
@@ -942,7 +970,7 @@ init:
 				*lens-- = (byte) len;
 			}
 		}
-	copy:
+
 		memcpy(end, dfield_get_data(field), len);
 		end += len;
 	}
@@ -1105,7 +1133,6 @@ rec_copy_prefix_to_buf(
 	dtype_t*	type;
 	ulint		i;
 	ulint		prefix_len;
-	ibool		is_null;
 	ulint		null_mask;
 	ulint		status;
 
@@ -1146,20 +1173,22 @@ rec_copy_prefix_to_buf(
 	for (i = 0; i < n_fields; i++) {
 		field = dict_index_get_nth_field(index, i);
 		type = dict_col_get_type(dict_field_get_col(field));
-		is_null = !(dtype_get_prtype(type) & DATA_NOT_NULL);
-		if (is_null) {
+		if (!(dtype_get_prtype(type) & DATA_NOT_NULL)) {
 			/* nullable field => read the null flag */
-			is_null = !!(*nulls & null_mask);
-			null_mask <<= 1;
-			if (null_mask == 0x100) {
-				--nulls;
-				UNIV_PREFETCH_R(nulls);
+			if (UNIV_UNLIKELY(!(byte) null_mask)) {
+				nulls--;
 				null_mask = 1;
 			}
+
+			if (*nulls & null_mask) {
+				null_mask <<= 1;
+				continue;
+			}
+
+			null_mask <<= 1;
 		}
 
-		if (is_null) {
-		} else if (field->fixed_len) {
+		if (field->fixed_len) {
 			prefix_len += field->fixed_len;
 		} else {
 			ulint	len = *lens--;
diff --git a/storage/innobase/row/row0mysql.c b/storage/innobase/row/row0mysql.c
index eb50b83a4d5..2ac0824b331 100644
--- a/storage/innobase/row/row0mysql.c
+++ b/storage/innobase/row/row0mysql.c
@@ -1429,51 +1429,106 @@ run_again:
 }
 
 /*************************************************************************
-Does an unlock of a row for MySQL. */
+This can only be used when srv_locks_unsafe_for_binlog is TRUE. Before
+calling this function we must use trx_reset_new_rec_lock_info() and
+trx_register_new_rec_lock() to store the information which new record locks
+really were set. This function removes a newly set lock under prebuilt->pcur,
+and also under prebuilt->clust_pcur. Currently, this is only used and tested
+in the case of an UPDATE or a DELETE statement, where the row lock is of the
+LOCK_X type.
+Thus, this implements a 'mini-rollback' that releases the latest record
+locks we set. */
 
 int
 row_unlock_for_mysql(
 /*=================*/
 					/* out: error code or DB_SUCCESS */
-	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL
 					handle */
+	ibool		has_latches_on_recs)/* TRUE if called so that we have
+					the latches on the records under pcur
+					and clust_pcur, and we do not need to
+					reposition the cursors. */
 {
-	rec_t*		rec;
-	btr_pcur_t*	cur		= prebuilt->pcur;
+	dict_index_t*	index;
+	btr_pcur_t*	pcur		= prebuilt->pcur;
+	btr_pcur_t*	clust_pcur	= prebuilt->clust_pcur;
 	trx_t*		trx		= prebuilt->trx;
+	rec_t*		rec;
 	mtr_t           mtr;
 	
 	ut_ad(prebuilt && trx);
 	ut_ad(trx->mysql_thread_id == os_thread_get_curr_id());
-		
+			
+	if (!srv_locks_unsafe_for_binlog) {
+
+		fprintf(stderr,
+"InnoDB: Error: calling row_unlock_for_mysql though\n"
+"InnoDB: srv_locks_unsafe_for_binlog is FALSE.\n");
+
+		return(DB_SUCCESS);
+	}
+
 	trx->op_info = "unlock_row";
-	
-	if (srv_locks_unsafe_for_binlog) {
-		if (trx->trx_create_lock == TRUE) {
 
-			mtr_start(&mtr);
+	index = btr_pcur_get_btr_cur(pcur)->index;
+
+	if (index != NULL && trx_new_rec_locks_contain(trx, index)) {
+
+		mtr_start(&mtr);
 			
-			/* Restore a cursor position and find a record */
-			btr_pcur_restore_position(BTR_SEARCH_LEAF, cur, &mtr);
-			rec = btr_pcur_get_rec(cur);
+		/* Restore the cursor position and find the record */
+		
+		if (!has_latches_on_recs) {
+			btr_pcur_restore_position(BTR_SEARCH_LEAF, pcur, &mtr);
+		}
 
-			if (rec) {
+		rec = btr_pcur_get_rec(pcur);
 
-				lock_rec_reset_and_release_wait(rec);
-			} else {
-				fputs("InnoDB: Error: "
-				      "Record for the lock not found\n",
-				      stderr);
-				mem_analyze_corruption((byte*) trx);
-				ut_error;
-			}
+		mutex_enter(&kernel_mutex);
 
-			trx->trx_create_lock = FALSE;
-			mtr_commit(&mtr);
+		lock_rec_reset_and_release_wait(rec);
+
+		mutex_exit(&kernel_mutex);
+
+		mtr_commit(&mtr);
+
+		/* If the search was done through the clustered index, then
+		we have not used clust_pcur at all, and we must NOT try to
+		reset locks on clust_pcur. The values in clust_pcur may be
+		garbage! */
+
+		if (index->type & DICT_CLUSTERED) {
+			
+			goto func_exit;
 		}
-		
+	}
+
+	index = btr_pcur_get_btr_cur(clust_pcur)->index;
+
+	if (index != NULL && trx_new_rec_locks_contain(trx, index)) {
+
+		mtr_start(&mtr);
+			
+		/* Restore the cursor position and find the record */
+
+		if (!has_latches_on_recs) {
+			btr_pcur_restore_position(BTR_SEARCH_LEAF, clust_pcur,
+									&mtr);
+		}
+
+		rec = btr_pcur_get_rec(clust_pcur);
+
+		mutex_enter(&kernel_mutex);
+
+		lock_rec_reset_and_release_wait(rec);
+
+		mutex_exit(&kernel_mutex);
+
+		mtr_commit(&mtr);
 	}
 			
+func_exit:
 	trx->op_info = "";
 	
 	return(DB_SUCCESS);
diff --git a/storage/innobase/row/row0sel.c b/storage/innobase/row/row0sel.c
index c7a548fe448..15439bed7e7 100644
--- a/storage/innobase/row/row0sel.c
+++ b/storage/innobase/row/row0sel.c
@@ -2784,6 +2784,10 @@ sel_restore_position_for_mysql(
 					process the record the cursor is
 					now positioned on (i.e. we should
 					not go to the next record yet) */
+	ibool*		same_user_rec,	/* out: TRUE if we were able to restore
+					the cursor on a user record with the
+					same ordering prefix in in the
+					B-tree index */
 	ulint		latch_mode,	/* in: latch mode wished in
 					restoration */
 	btr_pcur_t*	pcur,		/* in: cursor whose position
@@ -2800,6 +2804,8 @@ sel_restore_position_for_mysql(
 	
 	success = btr_pcur_restore_position(latch_mode, pcur, mtr);
 
+	*same_user_rec = success;
+
 	if (relative_position == BTR_PCUR_ON) {
 		if (success) {
 			return(FALSE);
@@ -3064,10 +3070,12 @@ row_search_for_mysql(
 	ulint		cnt				= 0;
 #endif /* UNIV_SEARCH_DEBUG */
 	ulint		next_offs;
+	ibool		same_user_rec;
 	mtr_t		mtr;
 	mem_heap_t*	heap				= NULL;
 	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 	ulint*		offsets				= offsets_;
+
 	*offsets_ = (sizeof offsets_) / sizeof *offsets_;
 
 	ut_ad(index && pcur && search_tuple);
@@ -3138,6 +3146,16 @@ row_search_for_mysql(
 		trx->search_latch_timeout = BTR_SEA_TIMEOUT;
 	}
 	
+	/* Reset the new record lock info if we srv_locks_unsafe_for_binlog
+	is set. Then we are able to remove the record locks set here on an
+	individual row. */
+
+	if (srv_locks_unsafe_for_binlog
+	    && prebuilt->select_lock_type != LOCK_NONE) {
+
+		trx_reset_new_rec_lock_info(trx);
+	}
+
 	/*-------------------------------------------------------------*/
 	/* PHASE 1: Try to pop the row from the prefetch cache */
 
@@ -3396,8 +3414,9 @@ shortcut_fails_too_big_rec:
 	clust_index = dict_table_get_first_index(index->table);
 
 	if (UNIV_LIKELY(direction != 0)) {
-		if (!sel_restore_position_for_mysql(BTR_SEARCH_LEAF, pcur,
-							moves_up, &mtr)) {
+		if (!sel_restore_position_for_mysql(&same_user_rec,
+						BTR_SEARCH_LEAF,
+						pcur, moves_up, &mtr)) {
 			goto next_rec;
 		}
 
@@ -3659,7 +3678,7 @@ rec_loop:
 			goto normal_return;
 		}
 	}
-		
+
 	/* We are ready to look at a possible new index entry in the result
 	set: the cursor is now placed on a user record */
 
@@ -3679,6 +3698,7 @@ rec_loop:
 		    || srv_locks_unsafe_for_binlog
 		    || (unique_search && !UNIV_UNLIKELY(rec_get_deleted_flag(
 					rec, page_rec_is_comp(rec))))) {
+
 			goto no_gap_lock;
 		} else {
 			lock_type = LOCK_ORDINARY;
@@ -3701,7 +3721,7 @@ rec_loop:
 		    && dtuple_get_n_fields_cmp(search_tuple)
 		       == dict_index_get_n_unique(index)
 		    && 0 == cmp_dtuple_rec(search_tuple, rec, offsets)) {
-		no_gap_lock:
+no_gap_lock:
 			lock_type = LOCK_REC_NOT_GAP;
 		}
 
@@ -3764,6 +3784,7 @@ rec_loop:
 			/* Get the clustered index record if needed */
 			index_rec = rec;
 			ut_ad(index != clust_index);
+
 			goto requires_clust_rec;
 		}
 	}
@@ -3773,6 +3794,17 @@ rec_loop:
 		/* The record is delete-marked: we can skip it if this is
 		not a consistent read which might see an earlier version
 		of a non-clustered index record */
+
+		if (srv_locks_unsafe_for_binlog
+	    	    && prebuilt->select_lock_type != LOCK_NONE) {
+
+			/* No need to keep a lock on a delete-marked record
+			if we do not want to use next-key locking. */
+
+			row_unlock_for_mysql(prebuilt, TRUE);
+			
+			trx_reset_new_rec_lock_info(trx);
+		}
 		
 		goto next_rec;
 	}
@@ -3783,7 +3815,8 @@ rec_loop:
 	index_rec = rec;
 
 	if (index != clust_index && prebuilt->need_to_access_clustered) {
-	requires_clust_rec:
+
+requires_clust_rec:
 		/* Before and after this "if" block, "offsets" will be
 		related to "rec", which may be in a secondary index "index" or
 		the clustered index ("clust_index").  However, after this
@@ -3816,6 +3849,18 @@ rec_loop:
 
 			/* The record is delete marked: we can skip it */
 
+			if (srv_locks_unsafe_for_binlog
+	    		    && prebuilt->select_lock_type != LOCK_NONE) {
+
+				/* No need to keep a lock on a delete-marked
+				record if we do not want to use next-key
+				locking. */
+
+				row_unlock_for_mysql(prebuilt, TRUE);
+			
+				trx_reset_new_rec_lock_info(trx);
+			}
+
 			goto next_rec;
 		}
 		
@@ -3908,7 +3953,7 @@ got_row:
 next_rec:
 	/*-------------------------------------------------------------*/
 	/* PHASE 5: Move the cursor to the next index record */
-	
+
 	if (UNIV_UNLIKELY(mtr_has_extra_clust_latch)) {
 		/* We must commit mtr if we are moving to the next
 		non-clustered index record, because we could break the
@@ -3921,8 +3966,9 @@ next_rec:
 		mtr_has_extra_clust_latch = FALSE;
 	
 		mtr_start(&mtr);
-		if (sel_restore_position_for_mysql(BTR_SEARCH_LEAF, pcur,
-							moves_up, &mtr)) {
+		if (sel_restore_position_for_mysql(&same_user_rec,
+						BTR_SEARCH_LEAF,
+						pcur, moves_up, &mtr)) {
 #ifdef UNIV_SEARCH_DEBUG
 			cnt++;
 #endif /* UNIV_SEARCH_DEBUG */
@@ -3973,11 +4019,34 @@ lock_wait_or_error:
 	thr->lock_state = QUE_THR_LOCK_ROW;
 
 	if (row_mysql_handle_errors(&err, trx, thr, NULL)) {
+		/* It was a lock wait, and it ended */
+
 		thr->lock_state = QUE_THR_LOCK_NOLOCK;
 		mtr_start(&mtr);
 
-		sel_restore_position_for_mysql(BTR_SEARCH_LEAF, pcur,
-							moves_up, &mtr);
+		sel_restore_position_for_mysql(&same_user_rec,
+						BTR_SEARCH_LEAF, pcur,
+						moves_up, &mtr);
+		if (srv_locks_unsafe_for_binlog && !same_user_rec) {
+			/* Since we were not able to restore the cursor
+			on the same user record, we cannot use
+			row_unlock_for_mysql() to unlock any records, and
+			we must thus reset the new rec lock info. Since
+			in lock0lock.c we have blocked the inheriting of gap
+			X-locks, we actually do not have any new record locks
+			set in this case.
+
+			Note that if we were able to restore on the 'same'
+			user record, it is still possible that we were actually
+			waiting on a delete-marked record, and meanwhile
+			it was removed by purge and inserted again by some
+			other user. But that is no problem, because in
+			rec_loop we will again try to set a lock, and
+			new_rec_lock_info in trx will be right at the end. */
+
+			trx_reset_new_rec_lock_info(trx);
+		}
+	
 		mode = pcur->search_mode;
 
 		goto rec_loop;
diff --git a/storage/innobase/row/row0upd.c b/storage/innobase/row/row0upd.c
index cf2b8db5d32..514fb6bd577 100644
--- a/storage/innobase/row/row0upd.c
+++ b/storage/innobase/row/row0upd.c
@@ -818,7 +818,7 @@ row_upd_build_difference_binary(
 		extern_bit = upd_ext_vec_contains(ext_vec, n_ext_vec, i);
 		
 		if (UNIV_UNLIKELY(extern_bit ==
-				!rec_offs_nth_extern(offsets, i))
+			    (ibool)!rec_offs_nth_extern(offsets, i))
 		    || !dfield_data_is_binary_equal(dfield, len, data)) {
 
 			upd_field = upd_get_nth_field(update, n_diff);
diff --git a/storage/innobase/srv/srv0srv.c b/storage/innobase/srv/srv0srv.c
index f901425a5f9..837c5be2bb6 100644
--- a/storage/innobase/srv/srv0srv.c
+++ b/storage/innobase/srv/srv0srv.c
@@ -260,7 +260,7 @@ semaphore contention and convoy problems can occur withput this restriction.
 Value 10 should be good if there are less than 4 processors + 4 disks in the
 computer. Bigger computers need bigger values. */
 
-ulong	srv_thread_concurrency	= 8;
+ulong	srv_thread_concurrency	= SRV_CONCURRENCY_THRESHOLD;
 
 os_fast_mutex_t	srv_conc_mutex;		/* this mutex protects srv_conc data
 					structures */
@@ -983,12 +983,6 @@ srv_conc_enter_innodb(
 	srv_conc_slot_t*	slot	  = NULL;
 	ulint			i;
 
-	if (srv_thread_concurrency >= 500) {
-		/* Disable the concurrency check */
-	
-		return;
-	}
-
 	/* If trx has 'free tickets' to enter the engine left, then use one
 	such ticket */
 
@@ -1134,7 +1128,7 @@ srv_conc_force_enter_innodb(
 	trx_t*	trx)	/* in: transaction object associated with the
 			thread */
 {
-	if (srv_thread_concurrency >= 500) {
+	if (srv_thread_concurrency >= SRV_CONCURRENCY_THRESHOLD) {
 	
 		return;
 	}
@@ -1160,7 +1154,7 @@ srv_conc_force_exit_innodb(
 {
 	srv_conc_slot_t*	slot	= NULL;
 
-	if (srv_thread_concurrency >= 500) {
+	if (srv_thread_concurrency >= SRV_CONCURRENCY_THRESHOLD) {
 	
 		return;
 	}
@@ -1212,11 +1206,6 @@ srv_conc_exit_innodb(
 	trx_t*	trx)	/* in: transaction object associated with the
 			thread */
 {
-	if (srv_thread_concurrency >= 500) {
-	
-		return;
-	}
-
 	if (trx->n_tickets_to_enter_innodb > 0) {
 		/* We will pretend the thread is still inside InnoDB though it
 		now leaves the InnoDB engine. In this way we save
diff --git a/storage/innobase/trx/trx0trx.c b/storage/innobase/trx/trx0trx.c
index 9e155ee1de0..10fbf3468c0 100644
--- a/storage/innobase/trx/trx0trx.c
+++ b/storage/innobase/trx/trx0trx.c
@@ -166,6 +166,8 @@ trx_create(
 	memset(&trx->xid, 0, sizeof(trx->xid));
 	trx->xid.formatID = -1;
 
+	trx_reset_new_rec_lock_info(trx);
+
 	return(trx);
 }
 
diff --git a/storage/innobase/trx/trx0undo.c b/storage/innobase/trx/trx0undo.c
index c14e4a1f3ab..7441dd3f152 100644
--- a/storage/innobase/trx/trx0undo.c
+++ b/storage/innobase/trx/trx0undo.c
@@ -559,14 +559,14 @@ trx_undo_write_xid(
 	const XID*	xid,	/* in: X/Open XA Transaction Identification */
 	mtr_t*		mtr)	/* in: mtr */
 {
-	mlog_write_ulint(log_hdr + TRX_UNDO_XA_FORMAT, xid->formatID,
-							MLOG_4BYTES, mtr);
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_FORMAT,
+			(ulint)xid->formatID, MLOG_4BYTES, mtr);
 
-	mlog_write_ulint(log_hdr + TRX_UNDO_XA_TRID_LEN, xid->gtrid_length,
-							MLOG_4BYTES, mtr);
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_TRID_LEN,
+			(ulint)xid->gtrid_length, MLOG_4BYTES, mtr);
 
-	mlog_write_ulint(log_hdr + TRX_UNDO_XA_BQUAL_LEN, xid->bqual_length,
-							MLOG_4BYTES, mtr);
+	mlog_write_ulint(log_hdr + TRX_UNDO_XA_BQUAL_LEN,
+			(ulint)xid->bqual_length, MLOG_4BYTES, mtr);
 
 	mlog_write_string(log_hdr + TRX_UNDO_XA_XID, (const byte*) xid->data,
 							XIDDATASIZE, mtr);
@@ -581,18 +581,14 @@ trx_undo_read_xid(
 	trx_ulogf_t*	log_hdr,/* in: undo log header */
 	XID*		xid)	/* out: X/Open XA Transaction Identification */
 {
-	ulint i;
-
-	xid->formatID = mach_read_from_4(log_hdr + TRX_UNDO_XA_FORMAT);
+	xid->formatID = (long)mach_read_from_4(log_hdr + TRX_UNDO_XA_FORMAT);
 	
-	xid->gtrid_length = mach_read_from_4(log_hdr + TRX_UNDO_XA_TRID_LEN);
-
-	xid->bqual_length = mach_read_from_4(log_hdr + TRX_UNDO_XA_BQUAL_LEN);
+	xid->gtrid_length =
+			(long)mach_read_from_4(log_hdr + TRX_UNDO_XA_TRID_LEN);
+	xid->bqual_length =
+		  (long)mach_read_from_4(log_hdr + TRX_UNDO_XA_BQUAL_LEN);
 
-	for (i = 0; i < XIDDATASIZE; i++) {
-		xid->data[i] = (char)mach_read_from_1(log_hdr + 
-							TRX_UNDO_XA_XID + i);
-	}
+	memcpy(xid->data, log_hdr + TRX_UNDO_XA_XID, XIDDATASIZE);
 }
 
 /*******************************************************************