Updated PBXT to version 1.0.08 RC2

51 files changed, 3316 insertions, 1365 deletions

diff --git a/storage/pbxt/ChangeLog b/storage/pbxt/ChangeLog
index 958bcd81cd1..5e3d594f103 100644
--- a/storage/pbxt/ChangeLog
+++ b/storage/pbxt/ChangeLog
@@ -1,7 +1,43 @@
 PBXT Release Notes
 ==================
 
-------- 1.0.08 RC - Not yet released
+------- 1.0.08 RC2 - 2009-06-30
+
+RN251: A Windows-specific test update, also removed false assertion that failed on Windows.
+
+RN250: Fixed a bug that caused recovery to fail when the transaction log ID exceeded 255. The problem was a checksum failed in the log record.
+
+RN249: Fixed bug #313176: Test case timeout. This happened because record cache pages where not propoerly freed and as soon as cache filled up the performacne degraded.
+
+RN248: PBXT now compiles and runs with MySQL 5.1.35. All tests pass.
+
+RN247: Fixed bug #369086: Incosistent/Incorrect Truncate behavior
+
+RN246: Fixed bug #378222: Drop sakila causes error: Cannot delete or update a parent row: a foreign key constraint fails
+
+RN245: Fixed bug #379315: Inconsistent behavior of DELETE IGNORE and FK constraint.
+
+RN244: Fixed a recovery problem: during the recovery of "record modified" action the table was updated before the old index entries were removed; then the xres_remove_index_entries was supplied the new record which lead to incorrect index update.
+
+RN243: Fixed a bug that caused a recovery failure if partitioned pbxt tables where present. This happended because the recovery used a MySQL function to open tables and the PBXT handler was not yet registered
+
+RN242: Fixed a bug that caused a deadlock if pbxt initialization failed. This happened because pbxt ceanup was done from pbxt_init() with PLUGIN_lock being held by MySQL which lead to a deadlock in the freeer thread
+
+RN241: Fixed a heap corruption bug (writing to a freed memory location). It happened only when memory mapped files were used leading to heap inconsistency and program crash or termination by heap checker. Likely to happen right after or during DROP TABLE but possible in other cases too.
+
+RN240: Load the record cache on read when no using memory mapped files.
+
+RN239: Added PBXT variable pbxt_max_threads. This is the maximum number of threads that can be created PBXT. By default this value is set to 0 which means the number of threads is derived from the MySQL variable max_connections. The value used is max_connections+7. Under Drizzle the default value is 500.
+
+RN238: Added an option to wait for the sweeper to clean up old transactions on a particular connection. This prevents the sweeper from getting too far behind.
+
+RN237: Added an option to lazy delete fixed length index entries. This means the index entries are just marked for deletion, instead of removing the items from the index page. This has the advantage that an exclusive lock is not always required for deletion.
+
+RN236: Fixed bug #349177: a bug in configure.in script.
+
+RN235: Fixed bug 349176: a compiler warning.
+
+RN234: Completed Drizzle integration. All Drizzle tests now run with PBXT.
 
 RN233: Fixed bugs which occur when PBXT is used together with PBMS (BLOB Streaming engine).
 
diff --git a/storage/pbxt/src/cache_xt.cc b/storage/pbxt/src/cache_xt.cc
index 0e15475f185..89b62c02282 100644
--- a/storage/pbxt/src/cache_xt.cc
+++ b/storage/pbxt/src/cache_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #ifndef XT_WIN
 #include <unistd.h>
 #endif
@@ -51,17 +55,22 @@
 #define IDX_CAC_SEGMENT_COUNT		((off_t) 1 << XT_INDEX_CACHE_SEGMENT_SHIFTS)
 #define IDX_CAC_SEGMENT_MASK		(IDX_CAC_SEGMENT_COUNT - 1)
 
-//#define IDX_USE_SPINRWLOCK
-#define IDX_USE_RWMUTEX
+#ifdef XT_NO_ATOMICS
+#define IDX_CAC_USE_PTHREAD_RW
+#else
+//#define IDX_CAC_USE_RWMUTEX
 //#define IDX_CAC_USE_PTHREAD_RW
+//#define IDX_USE_SPINXSLOCK
+#define IDX_CAC_USE_XSMUTEX
+#endif
 
-#ifdef IDX_CAC_USE_FASTWRLOCK
-#define IDX_CAC_LOCK_TYPE				XTFastRWLockRec
-#define IDX_CAC_INIT_LOCK(s, i)			xt_fastrwlock_init(s, &(i)->cs_lock)
-#define IDX_CAC_FREE_LOCK(s, i)			xt_fastrwlock_free(s, &(i)->cs_lock)	
-#define IDX_CAC_READ_LOCK(i, o)			xt_fastrwlock_slock(&(i)->cs_lock, (o))
-#define IDX_CAC_WRITE_LOCK(i, o)		xt_fastrwlock_xlock(&(i)->cs_lock, (o))
-#define IDX_CAC_UNLOCK(i, o)			xt_fastrwlock_unlock(&(i)->cs_lock, (o))
+#ifdef IDX_CAC_USE_XSMUTEX
+#define IDX_CAC_LOCK_TYPE				XTXSMutexRec
+#define IDX_CAC_INIT_LOCK(s, i)			xt_xsmutex_init_with_autoname(s, &(i)->cs_lock)
+#define IDX_CAC_FREE_LOCK(s, i)			xt_xsmutex_free(s, &(i)->cs_lock)	
+#define IDX_CAC_READ_LOCK(i, o)			xt_xsmutex_slock(&(i)->cs_lock, (o)->t_id)
+#define IDX_CAC_WRITE_LOCK(i, o)		xt_xsmutex_xlock(&(i)->cs_lock, (o)->t_id)
+#define IDX_CAC_UNLOCK(i, o)			xt_xsmutex_unlock(&(i)->cs_lock, (o)->t_id)
 #elif defined(IDX_CAC_USE_PTHREAD_RW)
 #define IDX_CAC_LOCK_TYPE				xt_rwlock_type
 #define IDX_CAC_INIT_LOCK(s, i)			xt_init_rwlock(s, &(i)->cs_lock)
@@ -69,13 +78,20 @@
 #define IDX_CAC_READ_LOCK(i, o)			xt_slock_rwlock_ns(&(i)->cs_lock)
 #define IDX_CAC_WRITE_LOCK(i, o)		xt_xlock_rwlock_ns(&(i)->cs_lock)
 #define IDX_CAC_UNLOCK(i, o)			xt_unlock_rwlock_ns(&(i)->cs_lock)
-#elif defined(IDX_USE_RWMUTEX)
+#elif defined(IDX_CAC_USE_RWMUTEX)
 #define IDX_CAC_LOCK_TYPE				XTRWMutexRec
 #define IDX_CAC_INIT_LOCK(s, i)			xt_rwmutex_init_with_autoname(s, &(i)->cs_lock)
 #define IDX_CAC_FREE_LOCK(s, i)			xt_rwmutex_free(s, &(i)->cs_lock)	
 #define IDX_CAC_READ_LOCK(i, o)			xt_rwmutex_slock(&(i)->cs_lock, (o)->t_id)
 #define IDX_CAC_WRITE_LOCK(i, o)		xt_rwmutex_xlock(&(i)->cs_lock, (o)->t_id)
 #define IDX_CAC_UNLOCK(i, o)			xt_rwmutex_unlock(&(i)->cs_lock, (o)->t_id)
+#elif defined(IDX_CAC_USE_SPINXSLOCK)
+#define IDX_CAC_LOCK_TYPE				XTSpinXSLockRec
+#define IDX_CAC_INIT_LOCK(s, i)			xt_spinxslock_init_with_autoname(s, &(i)->cs_lock)
+#define IDX_CAC_FREE_LOCK(s, i)			xt_spinxslock_free(s, &(i)->cs_lock)	
+#define IDX_CAC_READ_LOCK(i, s)			xt_spinxslock_slock(&(i)->cs_lock, (s)->t_id)
+#define IDX_CAC_WRITE_LOCK(i, s)		xt_spinxslock_xlock(&(i)->cs_lock, (s)->t_id)
+#define IDX_CAC_UNLOCK(i, s)			xt_spinxslock_unlock(&(i)->cs_lock, (s)->t_id)
 #endif
 
 #define ID_HANDLE_USE_SPINLOCK
@@ -308,7 +324,8 @@ xtPublic XTIndHandlePtr xt_ind_get_handle(XTOpenTablePtr ot, XTIndexPtr ind, XTI
 
 	hs = &ind_cac_globals.cg_handle_slot[iref->ir_block->cb_address % XT_HANDLE_SLOTS];
 
-	ASSERT_NS(iref->ir_ulock == XT_UNLOCK_READ);
+	ASSERT_NS(iref->ir_xlock == FALSE);
+	ASSERT_NS(iref->ir_updated == FALSE);
 	ID_HANDLE_LOCK(&hs->hs_handles_lock);
 #ifdef CHECK_HANDLE_STRUCTS
 	ic_check_handle_structs();
@@ -337,10 +354,10 @@ xtPublic XTIndHandlePtr xt_ind_get_handle(XTOpenTablePtr ot, XTIndexPtr ind, XTI
 	 * at least an Slock on the index.
 	 * So this excludes anyone who is reading 
 	 * cb_handle_count in the index.
-	 * (all cache block writers, and a freeer).
+	 * (all cache block writers, and the freeer).
 	 *
 	 * The increment is safe because I have the list
-	 * lock, which is required by anyone else
+	 * lock (hs_handles_lock), which is required by anyone else
 	 * who increments or decrements this value.
 	 */
 	iref->ir_block->cb_handle_count++;
@@ -396,8 +413,11 @@ xtPublic void xt_ind_release_handle(XTIndHandlePtr handle, xtBool have_lock, XTT
 		xblock = seg->cs_hash_table[hash_idx];
 		while (xblock) {
 			if (block == xblock) {
-				/* Found the block... */
-				xt_atomicrwlock_xlock(&block->cb_lock, thread->t_id);
+				/* Found the block... 
+				 * {HANDLE-COUNT-SLOCK}
+				 * 04.05.2009, changed to slock.
+				 */
+				XT_IPAGE_READ_LOCK(&block->cb_lock);
 				goto block_found;
 			}
 			xblock = xblock->cb_next;
@@ -431,7 +451,18 @@ xtPublic void xt_ind_release_handle(XTIndHandlePtr handle, xtBool have_lock, XTT
 		/* {HANDLE-COUNT-USAGE}
 		 * This is safe here because I have excluded
 		 * all readers by taking an Xlock on the
-		 * cache block.
+		 * cache block (CHANGED - see below).
+		 *
+		 * {HANDLE-COUNT-SLOCK}
+		 * 04.05.2009, changed to slock.
+		 * Should be OK, because:
+		 * A have a lock on the list lock (hs_handles_lock),
+		 * which prevents concurrent updates to cb_handle_count.
+		 *
+		 * I have also have a read lock on the cache block
+		 * but not a lock on the index. As a result, we cannot
+		 * excluded all index writers (and readers of 
+		 * cb_handle_count.
 		 */
 		block->cb_handle_count--;
 	}
@@ -466,7 +497,7 @@ xtPublic void xt_ind_release_handle(XTIndHandlePtr handle, xtBool have_lock, XTT
 	ID_HANDLE_UNLOCK(&hs->hs_handles_lock);
 
 	if (block)
-		xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+		XT_IPAGE_UNLOCK(&block->cb_lock, FALSE);
 }
 
 /* Call this function before a referenced cache block is modified!
@@ -482,17 +513,28 @@ xtPublic xtBool xt_ind_copy_on_write(XTIndReferencePtr iref)
 
 	hs = &ind_cac_globals.cg_handle_slot[iref->ir_block->cb_address % XT_HANDLE_SLOTS];
 
+	ID_HANDLE_LOCK(&hs->hs_handles_lock);
+
 	/* {HANDLE-COUNT-USAGE}
 	 * This is only called by updaters of this index block, or
 	 * the free which holds an Xlock on the index block.
-	 *
 	 * These are all mutually exclusive for the index block.
+	 *
+	 * {HANDLE-COUNT-SLOCK}
+	 * Do this check again, after we have the list lock (hs_handles_lock).
+	 * There is a small chance that the count has changed, since we last
+	 * checked because xt_ind_release_handle() only holds
+	 * an slock on the index page.
+	 *
+	 * An updater can sometimes have a XLOCK on the index and an slock
+	 * on the cache block. In this case xt_ind_release_handle()
+	 * could have run through.
 	 */
-	ASSERT_NS(iref->ir_block->cb_handle_count);
-	if (!iref->ir_block->cb_handle_count)
+	if (!iref->ir_block->cb_handle_count) {
+		ID_HANDLE_UNLOCK(&hs->hs_handles_lock);
 		return OK;
+	}
 
-	ID_HANDLE_LOCK(&hs->hs_handles_lock);
 #ifdef CHECK_HANDLE_STRUCTS
 	ic_check_handle_structs();
 #endif
@@ -609,7 +651,7 @@ xtPublic void xt_ind_init(XTThreadPtr self, size_t cache_size)
 #endif
 
 		for (u_int i=0; i<ind_cac_globals.cg_block_count; i++) {
-			xt_atomicrwlock_init_with_autoname(self, &block->cb_lock);
+			XT_IPAGE_INIT_LOCK(self, &block->cb_lock);
 			block->cb_state = IDX_CAC_BLOCK_FREE;
 			block->cb_next = ind_cac_globals.cg_free_list;
 #ifdef XT_USE_DIRECT_IO_ON_INDEX
@@ -836,10 +878,10 @@ static xtBool ind_free_block(XTOpenTablePtr ot, XTIndBlockPtr block)
 	while (xblock) {
 		if (block == xblock) {
 			/* Found the block... */
-			xt_atomicrwlock_xlock(&block->cb_lock, ot->ot_thread->t_id);
+			XT_IPAGE_WRITE_LOCK(&block->cb_lock, ot->ot_thread->t_id);
 			if (block->cb_state != IDX_CAC_BLOCK_CLEAN) {
 				/* This block cannot be freeed: */
-				xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+				XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 				IDX_CAC_UNLOCK(seg, ot->ot_thread);
 #ifdef DEBUG_CHECK_IND_CACHE
 				xt_ind_check_cache(NULL);
@@ -878,11 +920,12 @@ static xtBool ind_free_block(XTOpenTablePtr ot, XTIndBlockPtr block)
 	if (block->cb_handle_count) {
 		XTIndReferenceRec	iref;
 		
-		iref.ir_ulock = XT_UNLOCK_WRITE;
+		iref.ir_xlock = TRUE;
+		iref.ir_updated = FALSE;
 		iref.ir_block = block;
 		iref.ir_branch = (XTIdxBranchDPtr) block->cb_data;
 		if (!xt_ind_copy_on_write(&iref)) {
-			xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+			XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 			return FALSE;
 		}
 	}
@@ -918,7 +961,7 @@ static xtBool ind_free_block(XTOpenTablePtr ot, XTIndBlockPtr block)
 	IDX_TRACE("%d- f%x\n", (int) XT_NODE_ID(address), (int) XT_GET_DISK_2(block->cb_data));
 
 	/* Unlock BEFORE the block is reused! */
-	xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+	XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 
 	xt_unlock_mutex_ns(&ind_cac_globals.cg_lock);
 
@@ -1001,7 +1044,7 @@ static u_int ind_cac_free_lru_blocks(XTOpenTablePtr ot, u_int blocks_required, X
  * Fetch the block. Note, if we are about to write the block
  * then there is no need to read it from disk!
  */
-static XTIndBlockPtr ind_cac_fetch(XTOpenTablePtr ot, xtIndexNodeID address, DcSegmentPtr *ret_seg, xtBool read_data)
+static XTIndBlockPtr ind_cac_fetch(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID address, DcSegmentPtr *ret_seg, xtBool read_data)
 {
 	register XTOpenFilePtr	file = ot->ot_ind_file;
 	register XTIndBlockPtr	block, new_block;
@@ -1110,6 +1153,7 @@ static XTIndBlockPtr ind_cac_fetch(XTOpenTablePtr ot, xtIndexNodeID address, DcS
 	new_block->cb_state = IDX_CAC_BLOCK_CLEAN;
 	new_block->cb_handle_count = 0;
 	new_block->cp_flush_seq = 0;
+	new_block->cp_del_count = 0;
 	new_block->cb_dirty_next = NULL;
 	new_block->cb_dirty_prev = NULL;
 
@@ -1172,6 +1216,13 @@ static XTIndBlockPtr ind_cac_fetch(XTOpenTablePtr ot, xtIndexNodeID address, DcS
 #endif
 	xt_unlock_mutex_ns(&dcg->cg_lock);
 
+	/* {LAZY-DEL-INDEX-ITEMS}
+	 * Conditionally count the number of deleted entries in the index:
+	 * We do this before other threads can read the block.
+	 */
+	if (ind->mi_lazy_delete && read_data)
+		xt_ind_count_deleted_items(ot->ot_table, ind, block);
+
 	/* Add to the hash table: */
 	block->cb_next = seg->cs_hash_table[hash_idx];
 	seg->cs_hash_table[hash_idx] = block;
@@ -1221,10 +1272,10 @@ xtPublic xtBool xt_ind_write(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID ad
 	XTIndBlockPtr	block;
 	DcSegmentPtr	seg;
 
-	if (!(block = ind_cac_fetch(ot, address, &seg, FALSE)))
+	if (!(block = ind_cac_fetch(ot, ind, address, &seg, FALSE)))
 		return FAILED;
 
-	xt_atomicrwlock_xlock(&block->cb_lock, ot->ot_thread->t_id);
+	XT_IPAGE_WRITE_LOCK(&block->cb_lock, ot->ot_thread->t_id);
 	ASSERT_NS(block->cb_state == IDX_CAC_BLOCK_CLEAN || block->cb_state == IDX_CAC_BLOCK_DIRTY);
 	memcpy(block->cb_data, data, size);
 	block->cp_flush_seq = ot->ot_table->tab_ind_flush_seq;
@@ -1239,7 +1290,7 @@ xtPublic xtBool xt_ind_write(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID ad
 		xt_spinlock_unlock(&ind->mi_dirty_lock);
 		block->cb_state = IDX_CAC_BLOCK_DIRTY;
 	}
-	xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+	XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 	IDX_CAC_UNLOCK(seg, ot->ot_thread);
 #ifdef XT_TRACK_INDEX_UPDATES
 	ot->ot_ind_changed++;
@@ -1259,10 +1310,10 @@ xtPublic xtBool xt_ind_write_cache(XTOpenTablePtr ot, xtIndexNodeID address, siz
 		return FAILED;
 
 	if (block) {
-		xt_atomicrwlock_xlock(&block->cb_lock, ot->ot_thread->t_id);
+		XT_IPAGE_WRITE_LOCK(&block->cb_lock, ot->ot_thread->t_id);
 		ASSERT_NS(block->cb_state == IDX_CAC_BLOCK_CLEAN || block->cb_state == IDX_CAC_BLOCK_DIRTY);
 		memcpy(block->cb_data, data, size);
-		xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+		XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 		IDX_CAC_UNLOCK(seg, ot->ot_thread);
 	}
 
@@ -1277,7 +1328,7 @@ xtPublic xtBool xt_ind_clean(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID ad
 	if (!ind_cac_get(ot, address, &seg, &block))
 		return FAILED;
 	if (block) {
-		xt_atomicrwlock_xlock(&block->cb_lock, ot->ot_thread->t_id);
+		XT_IPAGE_WRITE_LOCK(&block->cb_lock, ot->ot_thread->t_id);
 		ASSERT_NS(block->cb_state == IDX_CAC_BLOCK_CLEAN || block->cb_state == IDX_CAC_BLOCK_DIRTY);
 
 		if (block->cb_state == IDX_CAC_BLOCK_DIRTY) {
@@ -1293,7 +1344,7 @@ xtPublic xtBool xt_ind_clean(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID ad
 			xt_spinlock_unlock(&ind->mi_dirty_lock);
 			block->cb_state = IDX_CAC_BLOCK_CLEAN;
 		}
-		xt_atomicrwlock_unlock(&block->cb_lock, TRUE);
+		XT_IPAGE_UNLOCK(&block->cb_lock, TRUE);
 
 		IDX_CAC_UNLOCK(seg, ot->ot_thread);
 	}
@@ -1301,29 +1352,33 @@ xtPublic xtBool xt_ind_clean(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID ad
 	return OK;
 }
 
-xtPublic xtBool xt_ind_read_bytes(XTOpenTablePtr ot, xtIndexNodeID address, size_t size, xtWord1 *data)
+xtPublic xtBool xt_ind_read_bytes(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID address, size_t size, xtWord1 *data)
 {
 	XTIndBlockPtr	block;
 	DcSegmentPtr	seg;
 
-	if (!(block = ind_cac_fetch(ot, address, &seg, TRUE)))
+	if (!(block = ind_cac_fetch(ot, ind, address, &seg, TRUE)))
 		return FAILED;
 
-	xt_atomicrwlock_slock(&block->cb_lock);
+	XT_IPAGE_READ_LOCK(&block->cb_lock);
 	memcpy(data, block->cb_data, size);
-	xt_atomicrwlock_unlock(&block->cb_lock, FALSE);
+	XT_IPAGE_UNLOCK(&block->cb_lock, FALSE);
 	IDX_CAC_UNLOCK(seg, ot->ot_thread);
 	return OK;
 }
 
-xtPublic xtBool xt_ind_fetch(XTOpenTablePtr ot, xtIndexNodeID address, XTPageLockType ltype, XTIndReferencePtr iref)
+xtPublic xtBool xt_ind_fetch(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID address, XTPageLockType ltype, XTIndReferencePtr iref)
 {
 	register XTIndBlockPtr	block;
 	DcSegmentPtr			seg;
 	xtWord2					branch_size;
+	xtBool					xlock = FALSE;
 
-	ASSERT_NS(iref->ir_ulock == XT_UNLOCK_NONE);
-	if (!(block = ind_cac_fetch(ot, address, &seg, TRUE)))
+#ifdef DEBUG
+	ASSERT_NS(iref->ir_xlock == 2);
+	ASSERT_NS(iref->ir_xlock == 2);
+#endif
+	if (!(block = ind_cac_fetch(ot, ind, address, &seg, TRUE)))
 		return NULL;
 
 	branch_size = XT_GET_DISK_2(((XTIdxBranchDPtr) block->cb_data)->tb_size_2);
@@ -1333,21 +1388,50 @@ xtPublic xtBool xt_ind_fetch(XTOpenTablePtr ot, xtIndexNodeID address, XTPageLoc
 		return FAILED;
 	}
 
-	if (ltype == XT_XLOCK_LEAF) {
-		if (XT_IS_NODE(branch_size))
-			ltype = XT_LOCK_READ;
-		else
-			ltype = XT_LOCK_WRITE;
+	switch (ltype) {
+		case XT_LOCK_READ:
+			break;
+		case XT_LOCK_WRITE:
+			xlock = TRUE;
+			break;
+		case XT_XLOCK_LEAF:
+			if (!XT_IS_NODE(branch_size))
+				xlock = TRUE;
+			break;
+		case XT_XLOCK_DEL_LEAF:
+			if (!XT_IS_NODE(branch_size)) {
+				if (ot->ot_table->tab_dic.dic_no_lazy_delete)
+					xlock = TRUE;
+				else {
+					/*
+					 * {LAZY-DEL-INDEX-ITEMS}
+					 *
+					 * We are fetch a page for delete purpose.
+					 * we decide here if we plan to do a lazy delete,
+					 * Or if we plan to compact the node.
+					 *
+					 * A lazy delete just requires a shared lock.
+					 *
+					 */
+					if (ind->mi_lazy_delete) {
+						/* If the number of deleted items is greater than
+						 * half of the number of times that can fit in the
+						 * page, then we will compact the node.
+						 */
+						if (!xt_idx_lazy_delete_on_leaf(ind, block, XT_GET_INDEX_BLOCK_LEN(branch_size)))
+							xlock = TRUE;
+					}
+					else
+						xlock = TRUE;
+				}
+			}
+			break;
 	}
 
-	if (ltype == XT_LOCK_WRITE) {
-		xt_atomicrwlock_xlock(&block->cb_lock, ot->ot_thread->t_id);
-		iref->ir_ulock = XT_UNLOCK_WRITE;
-	}
-	else {
-		xt_atomicrwlock_slock(&block->cb_lock);
-		iref->ir_ulock = XT_UNLOCK_READ;
-	}
+	if ((iref->ir_xlock = xlock))
+		XT_IPAGE_WRITE_LOCK(&block->cb_lock, ot->ot_thread->t_id);
+	else
+		XT_IPAGE_READ_LOCK(&block->cb_lock);
 
 	IDX_CAC_UNLOCK(seg, ot->ot_thread);
 
@@ -1358,18 +1442,31 @@ xtPublic xtBool xt_ind_fetch(XTOpenTablePtr ot, xtIndexNodeID address, XTPageLoc
 	 * As a result, we need to pass a pointer to both the
 	 * cache block and the cache block data:
 	 */
+	iref->ir_updated = FALSE;
 	iref->ir_block = block;
 	iref->ir_branch = (XTIdxBranchDPtr) block->cb_data;
 	return OK;
 }
 
-xtPublic xtBool xt_ind_release(XTOpenTablePtr ot, XTIndexPtr ind, XTPageUnlockType XT_UNUSED(utype), XTIndReferencePtr iref)
+xtPublic xtBool xt_ind_release(XTOpenTablePtr ot, XTIndexPtr ind, XTPageUnlockType XT_NDEBUG_UNUSED(utype), XTIndReferencePtr iref)
 {
 	register XTIndBlockPtr	block;
 
 	block = iref->ir_block;
 
-	if (utype == XT_UNLOCK_R_UPDATE || utype == XT_UNLOCK_W_UPDATE) {
+#ifdef DEBUG
+	ASSERT_NS(iref->ir_xlock != 2);
+	ASSERT_NS(iref->ir_updated != 2);
+	if (iref->ir_updated)
+		ASSERT_NS(utype == XT_UNLOCK_R_UPDATE || utype == XT_UNLOCK_W_UPDATE);
+	else
+		ASSERT_NS(utype == XT_UNLOCK_READ || utype == XT_UNLOCK_WRITE);
+	if (iref->ir_xlock)
+		ASSERT_NS(utype == XT_UNLOCK_WRITE || utype == XT_UNLOCK_W_UPDATE);
+	else
+		ASSERT_NS(utype == XT_UNLOCK_READ || utype == XT_UNLOCK_R_UPDATE);
+#endif
+	if (iref->ir_updated) {
 		/* The page was update: */
 		ASSERT_NS(block->cb_state == IDX_CAC_BLOCK_CLEAN || block->cb_state == IDX_CAC_BLOCK_DIRTY);
 		block->cp_flush_seq = ot->ot_table->tab_ind_flush_seq;
@@ -1386,16 +1483,10 @@ xtPublic xtBool xt_ind_release(XTOpenTablePtr ot, XTIndexPtr ind, XTPageUnlockTy
 		}
 	}
 
+	XT_IPAGE_UNLOCK(&block->cb_lock, iref->ir_xlock);
 #ifdef DEBUG
-	if (utype == XT_UNLOCK_W_UPDATE)
-		utype = XT_UNLOCK_WRITE;
-	else if (utype == XT_UNLOCK_R_UPDATE)
-		utype = XT_UNLOCK_READ;
-	ASSERT_NS(iref->ir_ulock == utype);
-#endif
-	xt_atomicrwlock_unlock(&block->cb_lock, iref->ir_ulock == XT_UNLOCK_WRITE ? TRUE : FALSE);
-#ifdef DEBUG
-	iref->ir_ulock = XT_UNLOCK_NONE;
+	iref->ir_xlock = 2;
+	iref->ir_updated = 2;
 #endif
 	return OK;
 }
@@ -1484,24 +1575,3 @@ xtPublic void xt_ind_unreserve(XTOpenTablePtr ot)
 		xt_ind_free_reserved(ot);
 }
 
-xtPublic void xt_load_indices(XTThreadPtr self, XTOpenTablePtr ot)
-{
-	register XTTableHPtr	tab = ot->ot_table;
-	register XTIndBlockPtr	block;
-	DcSegmentPtr			seg;
-	xtIndexNodeID			id;
-
-	xt_lock_mutex_ns(&tab->tab_ind_flush_lock);
-
-	for (id=1; id < XT_NODE_ID(tab->tab_ind_eof); id++) {
-		if (!(block = ind_cac_fetch(ot, id, &seg, TRUE))) {
-			xt_unlock_mutex_ns(&tab->tab_ind_flush_lock);
-			xt_throw(self);
-		}
-		IDX_CAC_UNLOCK(seg, ot->ot_thread);
-	}
-
-	xt_unlock_mutex_ns(&tab->tab_ind_flush_lock);
-}
-
-
diff --git a/storage/pbxt/src/cache_xt.h b/storage/pbxt/src/cache_xt.h
index d113bb2f907..fcd05a8fe34 100644
--- a/storage/pbxt/src/cache_xt.h
+++ b/storage/pbxt/src/cache_xt.h
@@ -45,8 +45,46 @@ struct XTIdxReadBuffer;
 #define IDX_CAC_BLOCK_CLEAN				1
 #define IDX_CAC_BLOCK_DIRTY				2
 
-typedef enum XTPageLockType { XT_LOCK_READ, XT_LOCK_WRITE, XT_XLOCK_LEAF };
-typedef enum XTPageUnlockType { XT_UNLOCK_NONE, XT_UNLOCK_READ, XT_UNLOCK_WRITE, XT_UNLOCK_R_UPDATE, XT_UNLOCK_W_UPDATE };
+#ifdef XT_NO_ATOMICS
+#define XT_IPAGE_USE_PTHREAD_RW
+#else
+//#define XT_IPAGE_USE_ATOMIC_RW
+#define XT_IPAGE_USE_SPINXSLOCK
+//#define XT_IPAGE_USE_SKEW_RW
+#endif
+
+#ifdef XT_IPAGE_USE_ATOMIC_RW
+#define XT_IPAGE_LOCK_TYPE				XTAtomicRWLockRec
+#define XT_IPAGE_INIT_LOCK(s, i)		xt_atomicrwlock_init_with_autoname(s, i)
+#define XT_IPAGE_FREE_LOCK(s, i)		xt_atomicrwlock_free(s, i)	
+#define XT_IPAGE_READ_LOCK(i)			xt_atomicrwlock_slock(i)
+#define XT_IPAGE_WRITE_LOCK(i, o)		xt_atomicrwlock_xlock(i, o)
+#define XT_IPAGE_UNLOCK(i, x)			xt_atomicrwlock_unlock(i, x)
+#elif defined(XT_IPAGE_USE_PTHREAD_RW)
+#define XT_IPAGE_LOCK_TYPE				xt_rwlock_type
+#define XT_IPAGE_INIT_LOCK(s, i)		xt_init_rwlock(s, i)
+#define XT_IPAGE_FREE_LOCK(s, i)		xt_free_rwlock(i)	
+#define XT_IPAGE_READ_LOCK(i)			xt_slock_rwlock_ns(i)
+#define XT_IPAGE_WRITE_LOCK(i, s)		xt_xlock_rwlock_ns(i)
+#define XT_IPAGE_UNLOCK(i, x)			xt_unlock_rwlock_ns(i)
+#elif defined(XT_IPAGE_USE_SPINXSLOCK)
+#define XT_IPAGE_LOCK_TYPE				XTSpinXSLockRec
+#define XT_IPAGE_INIT_LOCK(s, i)		xt_spinxslock_init_with_autoname(s, i)
+#define XT_IPAGE_FREE_LOCK(s, i)		xt_spinxslock_free(s, i)	
+#define XT_IPAGE_READ_LOCK(i)			xt_spinxslock_slock(i)
+#define XT_IPAGE_WRITE_LOCK(i, o)		xt_spinxslock_xlock(i, o)
+#define XT_IPAGE_UNLOCK(i, x)			xt_spinxslock_unlock(i, x)
+#else // XT_IPAGE_USE_SKEW_RW
+#define XT_IPAGE_LOCK_TYPE				XTSkewRWLockRec
+#define XT_IPAGE_INIT_LOCK(s, i)		xt_skewrwlock_init_with_autoname(s, i)
+#define XT_IPAGE_FREE_LOCK(s, i)		xt_skewrwlock_free(s, i)	
+#define XT_IPAGE_READ_LOCK(i)			xt_skewrwlock_slock(i)
+#define XT_IPAGE_WRITE_LOCK(i, o)		xt_skewrwlock_xlock(i, o)
+#define XT_IPAGE_UNLOCK(i, x)			xt_skewrwlock_unlock(i, x)
+#endif
+
+enum XTPageLockType { XT_LOCK_READ, XT_LOCK_WRITE, XT_XLOCK_LEAF, XT_XLOCK_DEL_LEAF };
+enum XTPageUnlockType { XT_UNLOCK_NONE, XT_UNLOCK_READ, XT_UNLOCK_WRITE, XT_UNLOCK_R_UPDATE, XT_UNLOCK_W_UPDATE };
 
 /* A block is X locked if it is being changed or freed.
  * A block is S locked if it is being read.
@@ -64,10 +102,11 @@ typedef struct XTIndBlock {
 	struct XTIndBlock	*cb_mr_used;					/* More recently used blocks. */
 	struct XTIndBlock	*cb_lr_used;					/* Less recently used blocks. */
 	/* Protected by cb_lock: */
-	XTAtomicRWLockRec	cb_lock;
+	XT_IPAGE_LOCK_TYPE	cb_lock;
 	xtWord1				cb_state;						/* Block status. */
 	xtWord2				cb_handle_count;				/* TRUE if this page is referenced by a handle. */
 	xtWord2				cp_flush_seq;
+	xtWord2				cp_del_count;					/* Number of deleted entries. */
 #ifdef XT_USE_DIRECT_IO_ON_INDEX
 	xtWord1				*cb_data;
 #else
@@ -76,16 +115,18 @@ typedef struct XTIndBlock {
 } XTIndBlockRec, *XTIndBlockPtr;
 
 typedef struct XTIndReference {
-	XTPageUnlockType		ir_ulock;
+	xtBool					ir_xlock;					/* Set to TRUE if the cache block is X locked. */
+	xtBool					ir_updated;					/* Set to TRUE if the cache block is updated. */
 	XTIndBlockPtr			ir_block;
 	XTIdxBranchDPtr			ir_branch;
 } XTIndReferenceRec, *XTIndReferencePtr;
 
 typedef struct XTIndFreeBlock {
+	XTDiskValue1			if_zero1_1;					/* Must be set to zero. */
+	XTDiskValue1			if_zero2_1;					/* Must be set to zero. */
 	XTDiskValue1			if_status_1;
 	XTDiskValue1			if_unused1_1;
-	XTDiskValue2			if_unused2_2;
-	XTDiskValue4			if_unused3_4;
+	XTDiskValue4			if_unused2_4;
 	XTDiskValue8			if_next_block_8;
 } XTIndFreeBlockRec, *XTIndFreeBlockPtr;
 
@@ -116,14 +157,13 @@ xtInt8			xt_ind_get_size();
 xtBool			xt_ind_write(struct XTOpenTable *ot, XTIndexPtr ind, xtIndexNodeID offset, size_t size, xtWord1 *data);
 xtBool			xt_ind_write_cache(struct XTOpenTable *ot, xtIndexNodeID offset, size_t size, xtWord1 *data);
 xtBool			xt_ind_clean(struct XTOpenTable *ot, XTIndexPtr ind, xtIndexNodeID offset);
-xtBool			xt_ind_read_bytes(struct XTOpenTable *ot, xtIndexNodeID offset, size_t size, xtWord1 *data);
+xtBool			xt_ind_read_bytes(struct XTOpenTable *ot, XTIndexPtr ind, xtIndexNodeID offset, size_t size, xtWord1 *data);
 void			xt_ind_check_cache(XTIndexPtr ind);
 xtBool			xt_ind_reserve(struct XTOpenTable *ot, u_int count, XTIdxBranchDPtr not_this);
 void			xt_ind_free_reserved(struct XTOpenTable *ot);
 void			xt_ind_unreserve(struct XTOpenTable *ot);
-void			xt_load_indices(XTThreadPtr self, struct XTOpenTable *ot);
 
-xtBool			xt_ind_fetch(struct XTOpenTable *ot, xtIndexNodeID node, XTPageLockType ltype, XTIndReferencePtr iref);
+xtBool			xt_ind_fetch(struct XTOpenTable *ot, XTIndexPtr ind, xtIndexNodeID node, XTPageLockType ltype, XTIndReferencePtr iref);
 xtBool			xt_ind_release(struct XTOpenTable *ot, XTIndexPtr ind, XTPageUnlockType utype, XTIndReferencePtr iref);
 
 void			xt_ind_lock_handle(XTIndHandlePtr handle);
diff --git a/storage/pbxt/src/ccutils_xt.cc b/storage/pbxt/src/ccutils_xt.cc
index 1d93e4c34b3..2f31061ac21 100644
--- a/storage/pbxt/src/ccutils_xt.cc
+++ b/storage/pbxt/src/ccutils_xt.cc
@@ -29,7 +29,7 @@
 #include "ccutils_xt.h"
 #include "bsearch_xt.h"
 
-static int ccu_compare_object(XTThreadPtr XT_UNUSED(self), register const void XT_UNUSED(*thunk), register const void *a, register const void *b)
+static int ccu_compare_object(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	XTObject *obj_ptr = (XTObject *) b;
 
diff --git a/storage/pbxt/src/database_xt.cc b/storage/pbxt/src/database_xt.cc
index d15ffc6d06e..c2231f6e854 100644
--- a/storage/pbxt/src/database_xt.cc
+++ b/storage/pbxt/src/database_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <string.h>
 #include <stdio.h>
 
@@ -240,7 +244,7 @@ static void db_hash_free(XTThreadPtr self, void *data)
 	xt_heap_release(self, (XTDatabaseHPtr) data);
 }
 
-static int db_cmp_db_id(struct XTThread XT_UNUSED(*self), register const void XT_UNUSED(*thunk), register const void *a, register const void *b)
+static int db_cmp_db_id(struct XTThread *XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtDatabaseID	db_id = *((xtDatabaseID *) a);
 	XTDatabaseHPtr	*db_ptr = (XTDatabaseHPtr *) b;
@@ -346,7 +350,7 @@ static void db_finalize(XTThreadPtr self, void *x)
 	}
 }
 
-static void db_onrelease(XTThreadPtr self, void XT_UNUSED(*x))
+static void db_onrelease(XTThreadPtr self, void *XT_UNUSED(x))
 {
 	/* Signal threads waiting for exclusive use of the database: */
 	if (xt_db_open_databases)	// The database may already be closed.
@@ -612,7 +616,7 @@ xtPublic void xt_drop_database(XTThreadPtr self, XTDatabaseHPtr	db)
 xtPublic void xt_open_database(XTThreadPtr self, char *path, xtBool multi_path)
 {
 	XTDatabaseHPtr db;
-	
+
 	/* We cannot get a database, without unusing the current
 	 * first. The reason is that the restart process will
 	 * partially set the current database!
@@ -621,7 +625,7 @@ xtPublic void xt_open_database(XTThreadPtr self, char *path, xtBool multi_path)
 	db = xt_get_database(self, path, multi_path);
 	pushr_(xt_heap_release, db);
 	xt_use_database(self, db, XT_FOR_USER);
-	freer_(); // xt_heap_release(self, db);	
+	freer_();	// xt_heap_release(self, db);	
 }
 
 /* This function can only be called if you do not already have a database in
@@ -638,6 +642,12 @@ xtPublic void xt_use_database(XTThreadPtr self, XTDatabaseHPtr db, int what_for)
 
 	xt_heap_reference(self, db);
 	self->st_database = db;
+#ifdef XT_WAIT_FOR_CLEANUP
+	self->st_last_xact = 0;
+	for (int i=0; i<XT_MAX_XACT_BEHIND; i++) {
+		self->st_prev_xact[i] = db->db_xn_curr_id;
+	}
+#endif
 	xt_xn_init_thread(self, what_for);
 }
 
@@ -1117,15 +1127,18 @@ xtPublic void xt_db_return_table_to_pool_ns(XTOpenTablePtr ot)
 	XTDatabaseHPtr		db = ot->ot_table->tab_db;
 	xtBool				flush_table = TRUE;
 
+	/* No open table returned to the pool should still
+	 * have a cache handle!
+	 */
+	ASSERT_NS(!ot->ot_ind_rhandle);
 	xt_lock_mutex_ns(&db->db_ot_pool.opt_lock);
 
 	if (!(table_pool = db_get_open_table_pool(db, ot->ot_table->tab_id)))
 		goto failed;
 
 	if (table_pool->opt_locked && !table_pool->opt_flushing) {
-		table_pool->opt_total_open--;
 		/* Table will be closed below: */
-		if (table_pool->opt_total_open > 0)
+		if (table_pool->opt_total_open > 1)
 			flush_table = FALSE;
 	}
 	else {
@@ -1151,14 +1164,21 @@ xtPublic void xt_db_return_table_to_pool_ns(XTOpenTablePtr ot)
 		ot = NULL;
 	}
 
+	if (ot) {
+		xt_unlock_mutex_ns(&db->db_ot_pool.opt_lock);
+		xt_close_table(ot, flush_table, FALSE);
+
+		/* assume that table_pool cannot be invalidated in between as we have table_pool->opt_total_open > 0 */
+		xt_lock_mutex_ns(&db->db_ot_pool.opt_lock);
+		table_pool->opt_total_open--;
+	}
+
 	db_free_open_table_pool(NULL, table_pool);
 
 	if (!xt_broadcast_cond_ns(&db->db_ot_pool.opt_cond))
 		goto failed;
 	xt_unlock_mutex_ns(&db->db_ot_pool.opt_lock);
-	if (ot)
-		xt_close_table(ot, flush_table, FALSE);
-
+	
 	return;
 
 	failed:
diff --git a/storage/pbxt/src/datadic_xt.cc b/storage/pbxt/src/datadic_xt.cc
index c2ac186cf6f..87bdf7c8b3d 100644
--- a/storage/pbxt/src/datadic_xt.cc
+++ b/storage/pbxt/src/datadic_xt.cc
@@ -26,6 +26,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <ctype.h>
 #include <errno.h>
 
@@ -433,7 +437,7 @@ class XTTokenizer {
 	XTToken *nextToken(XTThreadPtr self, c_char *keyword, XTToken *tk);
 };
 
-void ri_free_token(XTThreadPtr self __attribute__((unused)), XTToken *tk)
+void ri_free_token(XTThreadPtr XT_UNUSED(self), XTToken *tk)
 {
 	delete tk;
 }
@@ -524,6 +528,13 @@ XTToken *XTTokenizer::nextToken(XTThreadPtr self)
 						break;
 					tkn_curr_pos++;
 				}
+				/* TODO: Unless sql_mode == 'NO_BACKSLASH_ESCAPES'!!! */
+				if (*tkn_curr_pos == '\\') {
+					if (*(tkn_curr_pos+1) == quote) {
+						if (quote == '"' || quote == '\'')
+							tkn_curr_pos++;
+					}
+				}
 				tkn_curr_pos++;
 			}
 			
@@ -639,7 +650,7 @@ class XTParseTable : public XTObject {
 	int parseKeyAction(XTThreadPtr self);	
 	void parseCreateTable(XTThreadPtr self);
 	void parseAddTableItem(XTThreadPtr self);
-	void parseQualifiedName(XTThreadPtr self, char *name);
+	void parseQualifiedName(XTThreadPtr self, char *parent_name, char *name);
 	void parseTableName(XTThreadPtr self, bool alterTable);
 	void parseExpression(XTThreadPtr self, bool allow_reserved);
 	void parseBrackets(XTThreadPtr self);
@@ -667,53 +678,53 @@ class XTParseTable : public XTObject {
 		memset(&pt_sbuffer, 0, sizeof(XTStringBufferRec));
 	}
 
-	virtual void finalize(XTThreadPtr self __attribute__((unused))) {
+	virtual void finalize(XTThreadPtr XT_UNUSED(self)) {
 		if (pt_tokenizer)
 			delete pt_tokenizer;
 		xt_sb_set_size(NULL, &pt_sbuffer, 0);
 	}
 
 	// Hooks to receive output from the parser:
-	virtual void setTableName(XTThreadPtr self __attribute__((unused)), char *name __attribute__((unused)), bool alterTable __attribute__((unused))) {
+	virtual void setTableName(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(name), bool XT_UNUSED(alterTable)) {
 	}
-	virtual void addColumn(XTThreadPtr self __attribute__((unused)), char *col_name __attribute__((unused)), char *old_col_name __attribute__((unused))) {
+	virtual void addColumn(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(col_name), char *XT_UNUSED(old_col_name)) {
 	}
 	virtual void setDataType(XTThreadPtr self, char *cstring) {
 		if (cstring) 
 			xt_free(self, cstring);
 	}
-	virtual void setNull(XTThreadPtr self __attribute__((unused)), bool nullOK __attribute__((unused))) {
+	virtual void setNull(XTThreadPtr XT_UNUSED(self), bool XT_UNUSED(nullOK)) {
 	}
-	virtual void setAutoInc(XTThreadPtr self __attribute__((unused)), bool autoInc __attribute__((unused))) {
+	virtual void setAutoInc(XTThreadPtr XT_UNUSED(self), bool XT_UNUSED(autoInc)) {
 	}
 	
 	/* Add a contraint. If lastColumn is TRUE then add the contraint 
 	 * to the last column. If not, expect addListedColumn() to be called.
 	 */
-	virtual void addConstraint(XTThreadPtr self __attribute__((unused)), char *name __attribute__((unused)), u_int type __attribute__((unused)), bool lastColumn __attribute__((unused))) {
+	virtual void addConstraint(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(name), u_int XT_UNUSED(type), bool XT_UNUSED(lastColumn)) {
 	}
 	
 	/* Move the last column created. If symbol is NULL then move the column to the
 	 * first position, else move it to the position just after the given column.
 	 */
-	virtual void moveColumn(XTThreadPtr self __attribute__((unused)), char *col_name __attribute__((unused))) {
+	virtual void moveColumn(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(col_name)) {
 	}
 
-	virtual void dropColumn(XTThreadPtr self __attribute__((unused)), char *col_name __attribute__((unused))) {
+	virtual void dropColumn(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(col_name)) {
 	}
 
-	virtual void dropConstraint(XTThreadPtr self __attribute__((unused)), char *name __attribute__((unused)), u_int type __attribute__((unused))) {
+	virtual void dropConstraint(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(name), u_int XT_UNUSED(type)) {
 	}
 
-	virtual void setIndexName(XTThreadPtr self __attribute__((unused)), char *name __attribute__((unused))) {
+	virtual void setIndexName(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(name)) {
 	}
-	virtual void addListedColumn(XTThreadPtr self __attribute__((unused)), char *index_col_name __attribute__((unused))) {
+	virtual void addListedColumn(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(index_col_name)) {
 	}
-	virtual void setReferencedTable(XTThreadPtr self __attribute__((unused)), char *ref_table __attribute__((unused))) {
+	virtual void setReferencedTable(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(ref_schema), char *XT_UNUSED(ref_table)) {
 	}
-	virtual void addReferencedColumn(XTThreadPtr self __attribute__((unused)), char *index_col_name __attribute__((unused))) {
+	virtual void addReferencedColumn(XTThreadPtr XT_UNUSED(self), char *XT_UNUSED(index_col_name)) {
 	}
-	virtual void setActions(XTThreadPtr self __attribute__((unused)), int on_delete __attribute__((unused)), int on_update __attribute__((unused))) {
+	virtual void setActions(XTThreadPtr XT_UNUSED(self), int XT_UNUSED(on_delete), int XT_UNUSED(on_update)) {
 	}
 
 	virtual void parseTable(XTThreadPtr self, bool convert, char *sql);	
@@ -859,7 +870,7 @@ void XTParseTable::parseAddTableItem(XTThreadPtr self)
 	if (pt_current->isKeyWord("CONSTRAINT")) {
 		pt_current = pt_tokenizer->nextToken(self);
 		if (pt_current->isIdentifier())
-			parseQualifiedName(self, name);
+			parseQualifiedName(self, NULL, name);
 	}
 
 	if (pt_current->isReservedWord(XT_TK_PRIMARY)) {
@@ -974,13 +985,15 @@ void XTParseTable::parseMoveColumn(XTThreadPtr self)
 		char	name[XT_IDENTIFIER_NAME_SIZE];
 
 		pt_current = pt_tokenizer->nextToken(self);
-		parseQualifiedName(self, name);
+		parseQualifiedName(self, NULL, name);
 		moveColumn(self, name);
 	}
 }
 
-void XTParseTable::parseQualifiedName(XTThreadPtr self, char *name)
+void XTParseTable::parseQualifiedName(XTThreadPtr self, char *parent_name, char *name)
 {
+	if (parent_name)
+		parent_name[0] = '\0';
 	/* Should be an identifier by I have this example:
 	 * CREATE TABLE t1 ( comment CHAR(32) ASCII NOT NULL, koi8_ru_f CHAR(32) CHARACTER SET koi8r NOT NULL default '' ) CHARSET=latin5;
 	 *
@@ -990,6 +1003,8 @@ void XTParseTable::parseQualifiedName(XTThreadPtr self, char *name)
 		raiseError(self, pt_current, XT_ERR_ID_TOO_LONG);
 	pt_current = pt_tokenizer->nextToken(self);
 	while (pt_current->isKeyWord(".")) {
+		if (parent_name)
+			xt_strcpy(XT_IDENTIFIER_NAME_SIZE,parent_name, name);
 		pt_current = pt_tokenizer->nextToken(self);
 		/* Accept anything after the DOT! */
 		if (pt_current->getString(name, XT_IDENTIFIER_NAME_SIZE) >= XT_IDENTIFIER_NAME_SIZE)
@@ -1002,7 +1017,7 @@ void XTParseTable::parseTableName(XTThreadPtr self, bool alterTable)
 {
 	char name[XT_IDENTIFIER_NAME_SIZE];
 
-	parseQualifiedName(self, name);
+	parseQualifiedName(self, NULL, name);
 	setTableName(self, name, alterTable);
 }
 
@@ -1011,7 +1026,7 @@ void XTParseTable::parseColumnDefinition(XTThreadPtr self, char *old_col_name)
 	char col_name[XT_IDENTIFIER_NAME_SIZE];
 
 	// column_definition
-	parseQualifiedName(self, col_name);
+	parseQualifiedName(self, NULL, col_name);
 	addColumn(self, col_name, old_col_name);
 	parseDataType(self);
 
@@ -1111,7 +1126,7 @@ u_int XTParseTable::columnList(XTThreadPtr self, bool index_cols)
 	pt_current->expectKeyWord(self, "(");
 	do {
 		pt_current = pt_tokenizer->nextToken(self);
-		parseQualifiedName(self, name);
+		parseQualifiedName(self, NULL, name);
 		addListedColumn(self, name);
 		cols++;
 		if (index_cols) {
@@ -1135,19 +1150,20 @@ void XTParseTable::parseReferenceDefinition(XTThreadPtr self, u_int req_cols)
 	int		on_delete = XT_KEY_ACTION_DEFAULT;
 	int		on_update = XT_KEY_ACTION_DEFAULT;
 	char	name[XT_IDENTIFIER_NAME_SIZE];
+	char	parent_name[XT_IDENTIFIER_NAME_SIZE];
 	u_int	cols = 0;
 
 	// REFERENCES tbl_name
 	pt_current = pt_tokenizer->nextToken(self, "REFERENCES", pt_current);
-	parseQualifiedName(self, name);
-	setReferencedTable(self, name);
+	parseQualifiedName(self, parent_name, name);
+	setReferencedTable(self, parent_name[0] ? parent_name : NULL, name);
 
 	// [ (index_col_name,...) ]
 	if (pt_current->isKeyWord("(")) {
 		pt_current->expectKeyWord(self, "(");
 		do {
 			pt_current = pt_tokenizer->nextToken(self);
-			parseQualifiedName(self, name);
+			parseQualifiedName(self, NULL, name);
 			addReferencedColumn(self, name);
 			cols++;
 			if (cols > req_cols)
@@ -1219,7 +1235,7 @@ void XTParseTable::parseAlterTable(XTThreadPtr self)
 			if (pt_current->isReservedWord(XT_TK_COLUMN))
 				pt_current = pt_tokenizer->nextToken(self);
 
-			parseQualifiedName(self, old_col_name);
+			parseQualifiedName(self, NULL, old_col_name);
 			parseColumnDefinition(self, old_col_name);
 			parseMoveColumn(self);
 		}
@@ -1251,7 +1267,7 @@ void XTParseTable::parseAlterTable(XTThreadPtr self)
 			else {
 				if (pt_current->isReservedWord(XT_TK_COLUMN))
 					pt_current = pt_tokenizer->nextToken(self);
-				parseQualifiedName(self, name);
+				parseQualifiedName(self, NULL, name);
 				dropColumn(self, name);
 			}
 		}
@@ -1259,7 +1275,7 @@ void XTParseTable::parseAlterTable(XTThreadPtr self)
 			pt_current = pt_tokenizer->nextToken(self);
 			if (pt_current->isKeyWord("TO"))
 				pt_current = pt_tokenizer->nextToken(self);
-			parseQualifiedName(self, name);
+			parseQualifiedName(self, NULL, name);
 		}
 		else
 			/* Just ignore the syntax until the next , */
@@ -1284,7 +1300,7 @@ void XTParseTable::parseCreateIndex(XTThreadPtr self)
 	else if (pt_current->isKeyWord("SPACIAL"))
 		pt_current = pt_tokenizer->nextToken(self);
 	pt_current = pt_tokenizer->nextToken(self, "INDEX", pt_current);
-	parseQualifiedName(self, name);
+	parseQualifiedName(self, NULL, name);
 	optionalIndexType(self);
 	pt_current = pt_tokenizer->nextToken(self, "ON", pt_current);
 	parseTableName(self, true);
@@ -1299,7 +1315,7 @@ void XTParseTable::parseDropIndex(XTThreadPtr self)
 
 	pt_current = pt_tokenizer->nextToken(self, "DROP", pt_current);
 	pt_current = pt_tokenizer->nextToken(self, "INDEX", pt_current);
-	parseQualifiedName(self, name);
+	parseQualifiedName(self, NULL, name);
 	pt_current = pt_tokenizer->nextToken(self, "ON", pt_current);
 	parseTableName(self, true);
 	dropConstraint(self, name, XT_DD_INDEX);
@@ -1340,7 +1356,7 @@ class XTCreateTable : public XTParseTable {
 	virtual void addConstraint(XTThreadPtr self, char *name, u_int type, bool lastColumn);
 	virtual void dropConstraint(XTThreadPtr self, char *name, u_int type);
 	virtual void addListedColumn(XTThreadPtr self, char *index_col_name);
-	virtual void setReferencedTable(XTThreadPtr self, char *ref_table);
+	virtual void setReferencedTable(XTThreadPtr self, char *ref_schema, char *ref_table);
 	virtual void addReferencedColumn(XTThreadPtr self, char *index_col_name);
 	virtual void setActions(XTThreadPtr self, int on_delete, int on_update);
 
@@ -1535,23 +1551,31 @@ void XTCreateTable::addListedColumn(XTThreadPtr self, char *index_col_name)
 	}
 }
 
-void XTCreateTable::setReferencedTable(XTThreadPtr self, char *ref_table)
+void XTCreateTable::setReferencedTable(XTThreadPtr self, char *ref_schema, char *ref_table)
 {
 	XTDDForeignKey	*fk = (XTDDForeignKey *) ct_curr_constraint;
 	char			path[PATH_MAX];
 
-	xt_strcpy(PATH_MAX, path, ct_tab_path->ps_path);
-	xt_remove_last_name_of_path(path);
-	if (ct_convert) {
-		char	buffer[XT_IDENTIFIER_NAME_SIZE];
-		size_t	len;
-
-		myxt_static_convert_identifier(self, ct_charset, ref_table, buffer, XT_IDENTIFIER_NAME_SIZE);
-		len = strlen(path);
-		myxt_static_convert_table_name(self, buffer, &path[len], PATH_MAX - len);
-	}
-	else
+	if (ref_schema) {
+		xt_strcpy(PATH_MAX,path, ".");
+		xt_add_dir_char(PATH_MAX, path);
+		xt_strcat(PATH_MAX, path, ref_schema);
+		xt_add_dir_char(PATH_MAX, path);
 		xt_strcat(PATH_MAX, path, ref_table);
+	} else {
+		xt_strcpy(PATH_MAX, path, ct_tab_path->ps_path);
+		xt_remove_last_name_of_path(path);
+		if (ct_convert) {
+			char	buffer[XT_IDENTIFIER_NAME_SIZE];
+			size_t	len;
+
+			myxt_static_convert_identifier(self, ct_charset, ref_table, buffer, XT_IDENTIFIER_NAME_SIZE);
+			len = strlen(path);
+			myxt_static_convert_table_name(self, buffer, &path[len], PATH_MAX - len);
+		}
+		else
+			xt_strcat(PATH_MAX, path, ref_table);
+	}
 
 	fk->fk_ref_tab_name = (XTPathStrPtr) xt_dup_string(self, path);
 }
@@ -1578,7 +1602,7 @@ void XTCreateTable::addReferencedColumn(XTThreadPtr self, char *index_col_name)
 		fk->fk_ref_cols.clone(self, &fk->co_cols);
 }
 
-void XTCreateTable::setActions(XTThreadPtr self __attribute__((unused)), int on_delete, int on_update)
+void XTCreateTable::setActions(XTThreadPtr XT_UNUSED(self), int on_delete, int on_update)
 {
 	XTDDForeignKey	*fk = (XTDDForeignKey *) ct_curr_constraint;
 
@@ -1773,6 +1797,7 @@ bool XTDDTableRef::checkReference(xtWord1 *before_buf, XTThreadPtr thread)
 	XTIdxSearchKeyRec	search_key;
 	xtXactID			xn_id;
 	XTXactWaitRec		xw;
+	bool				ok = false;
 
 	if (!(loc_ind = tr_fkey->getReferenceIndexPtr()))
 		return false;
@@ -1792,40 +1817,42 @@ bool XTDDTableRef::checkReference(xtWord1 *before_buf, XTThreadPtr thread)
 
 	/* Search for the key in the child (referencing) table: */
 	if (!(ot = xt_db_open_table_using_tab(tr_fkey->co_table->dt_table, thread)))
-		goto failed;
+		return false;
 
 	retry:
 	if (!xt_idx_search(ot, ind, &search_key))
-		goto failed;
+		goto done;
 		
 	while (ot->ot_curr_rec_id && search_key.sk_on_key) {
 		switch (xt_tab_maybe_committed(ot, ot->ot_curr_rec_id, &xn_id, &ot->ot_curr_row_id, &ot->ot_curr_updated)) {
 			case XT_MAYBE:				
 				xw.xw_xn_id = xn_id;
 				if (!xt_xn_wait_for_xact(thread, &xw, NULL))
-					goto failed;
+					goto done;
 				goto retry;
 			case XT_ERR:
-				goto failed;
+				goto done;
 			case TRUE:
 				/* We found a matching child: */
 				xt_register_ixterr(XT_REG_CONTEXT, XT_ERR_ROW_IS_REFERENCED, tr_fkey->co_name);
-				goto failed;
-				break;
+				goto done;
 			case FALSE:
 				if (!xt_idx_next(ot, ind, &search_key))
-					goto failed;
+					goto done;
 				break;
 		}
 	}
 
 	/* No matching children, all OK: */
-	xt_db_return_table_to_pool_ns(ot);
-	return true;
+	ok = true;
 
-	failed:
+	done:
+	if (ot->ot_ind_rhandle) {
+		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, thread);
+		ot->ot_ind_rhandle = NULL;
+	}
 	xt_db_return_table_to_pool_ns(ot);
-	return false;
+	return ok;
 }
 
 /*
@@ -1962,6 +1989,10 @@ bool XTDDTableRef::modifyRow(XTOpenTablePtr XT_UNUSED(ref_ot), xtWord1 *before_b
 	}
 
 	/* No matching children, all OK: */
+	if (ot->ot_ind_rhandle) {
+		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, thread);
+		ot->ot_ind_rhandle = NULL;
+	}
 	xt_db_return_table_to_pool_ns(ot);
 
 	success:
@@ -1971,6 +2002,10 @@ bool XTDDTableRef::modifyRow(XTOpenTablePtr XT_UNUSED(ref_ot), xtWord1 *before_b
 	return true;
 
 	failed_2:
+	if (ot->ot_ind_rhandle) {
+		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, thread);
+		ot->ot_ind_rhandle = NULL;
+	}
 	xt_db_return_table_to_pool_ns(ot);
 
 	failed:
@@ -2055,8 +2090,13 @@ void XTDDForeignKey::finalize(XTThreadPtr self)
 
 void XTDDForeignKey::loadString(XTThreadPtr self, XTStringBufferPtr sb)
 {
+	char schema_name[XT_IDENTIFIER_NAME_SIZE];
+	
 	XTDDConstraint::loadString(self, sb);
 	xt_sb_concat(self, sb, " REFERENCES `");
+	xt_2nd_last_name_of_path(XT_IDENTIFIER_NAME_SIZE, schema_name, fk_ref_tab_name->ps_path);
+	xt_sb_concat(self, sb, schema_name);
+	xt_sb_concat(self, sb, "`.`");
 	xt_sb_concat(self, sb, xt_last_name_of_path(fk_ref_tab_name->ps_path));
 	xt_sb_concat(self, sb, "` ");
 
@@ -2288,6 +2328,10 @@ bool XTDDForeignKey::insertRow(xtWord1 *before_buf, xtWord1 *rec_buf, XTThreadPt
 				goto failed_2;
 			case TRUE:
 				/* We found a matching parent: */
+				if (ot->ot_ind_rhandle) {
+					xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, thread);
+					ot->ot_ind_rhandle = NULL;
+				}
 				xt_db_return_table_to_pool_ns(ot);
 				goto success;
 			case FALSE:
@@ -2300,6 +2344,10 @@ bool XTDDForeignKey::insertRow(xtWord1 *before_buf, xtWord1 *rec_buf, XTThreadPt
 	xt_register_ixterr(XT_REG_CONTEXT, XT_ERR_NO_REFERENCED_ROW, co_name);
 
 	failed_2:
+	if (ot->ot_ind_rhandle) {
+		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, thread);
+		ot->ot_ind_rhandle = NULL;
+	}
 	xt_db_return_table_to_pool_ns(ot);
 
 	failed:
@@ -2867,9 +2915,33 @@ bool XTDDTable::updateRow(XTOpenTablePtr ot, xtWord1 *before, xtWord1 *after)
 	return ok;
 }
 
-xtBool XTDDTable::checkCanDrop()
+/*
+ * drop_db parameter is TRUE if we are dropping the schema of this table. In this case
+ * we return TRUE if the table has only refs to the tables from its own schema
+ */
+xtBool XTDDTable::checkCanDrop(xtBool drop_db)
 {
 	/* no refs or references only itself */
-	return (dt_trefs == NULL) || 
-		(dt_trefs->tr_next == NULL) && (dt_trefs->tr_fkey->co_table == this);
+	if ((dt_trefs == NULL) || ((dt_trefs->tr_next == NULL) && (dt_trefs->tr_fkey->co_table == this)))
+		return TRUE;
+
+	if (!drop_db) 
+		return FALSE;
+	
+	const char *this_schema = xt_last_2_names_of_path(dt_table->tab_name->ps_path);
+	size_t this_schema_sz = xt_last_name_of_path(dt_table->tab_name->ps_path) - this_schema;
+	XTDDTableRef *tr = dt_trefs;
+
+	while (tr) {
+		const char *tab_path = tr->tr_fkey->co_table->dt_table->tab_name->ps_path;
+		const char *tab_schema = xt_last_2_names_of_path(tab_path);
+		size_t tab_schema_sz = xt_last_name_of_path(tab_path) - tab_schema;
+
+		if (this_schema_sz != tab_schema_sz || strncmp(this_schema, tab_schema, tab_schema_sz))
+			return FALSE;
+		
+		tr = tr->tr_next;
+	}
+
+	return TRUE;
 }
diff --git a/storage/pbxt/src/datadic_xt.h b/storage/pbxt/src/datadic_xt.h
index 825914b60f3..72565ce1212 100644
--- a/storage/pbxt/src/datadic_xt.h
+++ b/storage/pbxt/src/datadic_xt.h
@@ -137,6 +137,7 @@ class XTDDColumnRef : public XTObject {
 		return new_obj;
 	}
 
+	virtual void init(XTThreadPtr self) { XTObject::init(self); }
 	virtual void init(XTThreadPtr self, XTObject *obj);
 	virtual void finalize(XTThreadPtr self);
 };
@@ -156,6 +157,7 @@ class XTDDConstraint : public XTObject {
 		co_ind_name(NULL) {
 	}
 
+	virtual void init(XTThreadPtr self) { XTObject::init(self); }
 	virtual void init(XTThreadPtr self, XTObject *obj);
 	virtual void finalize(XTThreadPtr self) {
 		if (co_name)
@@ -198,6 +200,7 @@ class XTDDIndex : public XTDDConstraint {
 		return new_obj;
 	}
 
+        virtual void init(XTThreadPtr self) { XTDDConstraint::init(self); };
 	virtual void init(XTThreadPtr self, XTObject *obj);
 	struct XTIndex *getIndexPtr();
 };
@@ -230,6 +233,7 @@ class XTDDForeignKey : public XTDDIndex {
 		return new_obj;
 	}
 
+        virtual void init(XTThreadPtr self) { XTDDIndex::init(self); }
 	virtual void init(XTThreadPtr self, XTObject *obj);
 	virtual void finalize(XTThreadPtr self);
 	virtual void loadString(XTThreadPtr self, XTStringBufferPtr sb);
@@ -284,7 +288,7 @@ class XTDDTable : public XTObject {
 	XTDDIndex *findReferenceIndex(XTDDForeignKey *fk);
 	bool insertRow(struct XTOpenTable *rec_ot, xtWord1 *buffer);
 	bool checkNoAction(struct XTOpenTable *ot, xtRecordID rec_id);
-	xtBool checkCanDrop();
+	xtBool checkCanDrop(xtBool drop_db);
 	bool deleteRow(struct XTOpenTable *rec_ot, xtWord1 *buffer);
 	void deleteAllRows(XTThreadPtr self);
 	bool updateRow(struct XTOpenTable *rec_ot, xtWord1 *before, xtWord1 *after);
diff --git a/storage/pbxt/src/datalog_xt.cc b/storage/pbxt/src/datalog_xt.cc
index dc9423e7eac..10390975f1e 100644
--- a/storage/pbxt/src/datalog_xt.cc
+++ b/storage/pbxt/src/datalog_xt.cc
@@ -301,7 +301,7 @@ static xtBool dl_write_log_header(XTDataLogFilePtr data_log, XTOpenFilePtr of, x
 	return OK;
 }
 
-static void dl_free_seq_read(XTThreadPtr self __attribute__((unused)), XTDataSeqReadPtr seq_read)
+static void dl_free_seq_read(XTThreadPtr XT_UNUSED(self), XTDataSeqReadPtr seq_read)
 {
 	seq_read->sl_seq_exit();
 }
@@ -452,7 +452,7 @@ xtBool XTDataLogCache::dls_set_log_state(XTDataLogFilePtr data_log, int state)
 	return FAILED;
 }
 
-static int dl_cmp_log_id(XTThreadPtr XT_UNUSED(self), register const void XT_UNUSED(*thunk), register const void *a, register const void *b)
+static int dl_cmp_log_id(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtLogID			log_id_a = *((xtLogID *) a);
 	xtLogID			log_id_b = *((xtLogID *) b);
@@ -1149,7 +1149,7 @@ xtBool XTDataLogBuffer::dlb_flush_log(xtBool commit, XTThreadPtr thread)
 	return OK;
 }
 
-xtBool XTDataLogBuffer::dlb_write_thru_log(xtLogID log_id __attribute__((unused)), xtLogOffset log_offset, size_t size, xtWord1 *data, XTThreadPtr thread)
+xtBool XTDataLogBuffer::dlb_write_thru_log(xtLogID XT_NDEBUG_UNUSED(log_id), xtLogOffset log_offset, size_t size, xtWord1 *data, XTThreadPtr thread)
 {
 	ASSERT_NS(log_id == dlb_data_log->dlf_log_id);
 
@@ -1166,7 +1166,7 @@ xtBool XTDataLogBuffer::dlb_write_thru_log(xtLogID log_id __attribute__((unused)
 	return OK;
 }
 
-xtBool XTDataLogBuffer::dlb_append_log(xtLogID log_id __attribute__((unused)), xtLogOffset log_offset, size_t size, xtWord1 *data, XTThreadPtr thread)
+xtBool XTDataLogBuffer::dlb_append_log(xtLogID XT_NDEBUG_UNUSED(log_id), xtLogOffset log_offset, size_t size, xtWord1 *data, XTThreadPtr thread)
 {
 	ASSERT_NS(log_id == dlb_data_log->dlf_log_id);
 
@@ -1357,7 +1357,7 @@ xtBool XTDataLogBuffer::dlb_delete_log(xtLogID log_id, xtLogOffset log_offset, s
  * Delete all the extended data belonging to a particular
  * table.
  */
-xtPublic void xt_dl_delete_ext_data(XTThreadPtr self, XTTableHPtr tab, xtBool missing_ok __attribute__((unused)), xtBool have_table_lock)
+xtPublic void xt_dl_delete_ext_data(XTThreadPtr self, XTTableHPtr tab, xtBool XT_UNUSED(missing_ok), xtBool have_table_lock)
 {
 	XTOpenTablePtr	ot;
 	xtRecordID		page_rec_id, offs_rec_id;
diff --git a/storage/pbxt/src/discover_xt.cc b/storage/pbxt/src/discover_xt.cc
index 074132d47cb..fb325d78f7f 100644
--- a/storage/pbxt/src/discover_xt.cc
+++ b/storage/pbxt/src/discover_xt.cc
@@ -1282,9 +1282,11 @@ warn:
 #endif // LOCK_OPEN_HACK_REQUIRED
 
 //------------------------------
-int xt_create_table_frm(handlerton *hton, THD* thd, const char *db, const char *name, DT_FIELD_INFO *info, DT_KEY_INFO *keys __attribute__((unused)), xtBool skip_existing)
+int xt_create_table_frm(handlerton *hton, THD* thd, const char *db, const char *name, DT_FIELD_INFO *info, DT_KEY_INFO *XT_UNUSED(keys), xtBool skip_existing)
 {
 #ifdef DRIZZLED
+    drizzled::message::Table table_proto;
+
 	static const char *ext = ".dfe";
 	static const int ext_len = 4;
 #else
@@ -1329,8 +1331,7 @@ int xt_create_table_frm(handlerton *hton, THD* thd, const char *db, const char *
 			info->field_flags,
             COLUMN_FORMAT_TYPE_FIXED,
 		    NULL /*default_value*/, NULL /*on_update_value*/, &comment, NULL /*change*/,
-            NULL /*interval_list*/, info->field_charset,
-			NULL /*vcol_info*/))
+            NULL /*interval_list*/, info->field_charset))
 #else
 		if (add_field_to_list(thd, &field_name, info->field_type, field_length_ptr, info->field_decimal_length,
 			info->field_flags,
@@ -1365,7 +1366,10 @@ int xt_create_table_frm(handlerton *hton, THD* thd, const char *db, const char *
 	
 	/* Create an internal temp table */
 #ifdef DRIZZLED
-	if (mysql_create_table_no_lock(thd, db, name, &mylex.create_info, &mylex.alter_info, 1, 0, false)) 
+    table_proto.set_name(name);
+    table_proto.set_type(drizzled::message::Table::STANDARD);
+
+	if (mysql_create_table_no_lock(thd, db, name, &mylex.create_info, &table_proto, &mylex.alter_info, 1, 0, false)) 
 		goto error;
 #else
 	if (mysql_create_table_no_lock(thd, db, name, &mylex.create_info, &mylex.alter_info, 1, 0)) 
diff --git a/storage/pbxt/src/filesys_xt.cc b/storage/pbxt/src/filesys_xt.cc
index 5ca36cd9244..5b8bdb583e0 100644
--- a/storage/pbxt/src/filesys_xt.cc
+++ b/storage/pbxt/src/filesys_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #ifndef XT_WIN
 #include <unistd.h>
 #include <dirent.h>
@@ -127,11 +131,11 @@ static void fs_close_fmap(XTThreadPtr self, XTFileMemMapPtr mm)
 		mm->mm_start = NULL;
 	}
 #endif
-	xt_rwmutex_free(self, &mm->mm_lock);
+	FILE_MAP_FREE_LOCK(self, &mm->mm_lock);
 	xt_free(self, mm);
 }
 
-static void fs_free_file(XTThreadPtr self, void *thunk __attribute__((unused)), void *item)
+static void fs_free_file(XTThreadPtr self, void *XT_UNUSED(thunk), void *item)
 {
 	XTFilePtr	file_ptr = *((XTFilePtr *) item);
 
@@ -148,17 +152,13 @@ static void fs_free_file(XTThreadPtr self, void *thunk __attribute__((unused)),
 		file_ptr->fil_filedes = XT_NULL_FD;
 	}
 
-	if (file_ptr->fil_memmap) {
-		fs_close_fmap(self, file_ptr->fil_memmap);
-		file_ptr->fil_memmap = NULL;
-	}
-
 #ifdef DEBUG_TRACE_FILES
 	PRINTF("%s: free file: (%d) %s\n", self->t_name, (int) file_ptr->fil_id, 
 		file_ptr->fil_path ? xt_last_2_names_of_path(file_ptr->fil_path) : "?");
 #endif
 
 	if (!file_ptr->fil_ref_count) {
+		ASSERT_NS(!file_ptr->fil_handle_count);
 		/* Flush any cache before this file is invalid: */
 		if (file_ptr->fil_path) {
 			xt_free(self, file_ptr->fil_path);
@@ -169,7 +169,7 @@ static void fs_free_file(XTThreadPtr self, void *thunk __attribute__((unused)),
 	}
 }
 
-static int fs_comp_file(XTThreadPtr self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int fs_comp_file(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	char		*file_name = (char *) a;
 	XTFilePtr	file_ptr = *((XTFilePtr *) b);
@@ -177,7 +177,7 @@ static int fs_comp_file(XTThreadPtr self __attribute__((unused)), register const
 	return strcmp(file_name, file_ptr->fil_path);
 }
 
-static int fs_comp_file_ci(XTThreadPtr self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int fs_comp_file_ci(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	char		*file_name = (char *) a;
 	XTFilePtr	file_ptr = *((XTFilePtr *) b);
@@ -938,6 +938,29 @@ xtBool xt_pread_file(XTOpenFilePtr of, off_t offset, size_t size, size_t min_siz
 	return OK;
 }
 
+xtPublic xtBool xt_lock_file_ptr(XTOpenFilePtr of, xtWord1 **data, off_t offset, size_t size, XTIOStatsPtr stat, XTThreadPtr thread)
+{
+	size_t red_size;
+
+	if (!*data) {
+		if (!(*data = (xtWord1 *) xt_malloc_ns(size)))
+			return FAILED;
+	}
+
+	if (!xt_pread_file(of, offset, size, 0, *data, &red_size, stat, thread))
+		return FAILED;
+	
+	//if (red_size < size)
+	//	memset();
+	return OK;
+}
+
+xtPublic void xt_unlock_file_ptr(XTOpenFilePtr XT_UNUSED(of), xtWord1 *data, XTThreadPtr XT_UNUSED(thread))
+{
+	if (data)
+		xt_free_ns(data);
+}
+
 /* ----------------------------------------------------------------------
  * Directory operations
  */
@@ -949,7 +972,13 @@ XTOpenDirPtr xt_dir_open(XTThreadPtr self, c_char *path, c_char *filter)
 {
 	XTOpenDirPtr	od;
 
-	pushsr_(od, xt_dir_close, (XTOpenDirPtr) xt_calloc(self, sizeof(XTOpenDirRec)));
+#ifdef XT_SOLARIS
+	/* see the comment in filesys_xt.h */
+	size_t sz = pathconf(path, _PC_NAME_MAX) + sizeof(XTOpenDirRec) + 1;
+#else
+	size_t sz = sizeof(XTOpenDirRec);
+#endif
+	pushsr_(od, xt_dir_close, (XTOpenDirPtr) xt_calloc(self, sz));
 
 #ifdef XT_WIN
 	size_t			len;
@@ -976,7 +1005,6 @@ XTOpenDirPtr xt_dir_open(XTThreadPtr self, c_char *path, c_char *filter)
 	if (!od->od_dir)
 		xt_throw_ferrno(XT_CONTEXT, errno, path);
 #endif
-
 	popr_(); // Discard xt_dir_close(od)
 	return od;
 }
@@ -1097,7 +1125,7 @@ xtBool xt_dir_next(XTThreadPtr self, XTOpenDirPtr od)
 }
 #endif
 
-char *xt_dir_name(XTThreadPtr self __attribute__((unused)), XTOpenDirPtr od)
+char *xt_dir_name(XTThreadPtr XT_UNUSED(self), XTOpenDirPtr od)
 {
 #ifdef XT_WIN
 	return od->od_data.cFileName;
@@ -1106,8 +1134,9 @@ char *xt_dir_name(XTThreadPtr self __attribute__((unused)), XTOpenDirPtr od)
 #endif
 }
 
-xtBool xt_dir_is_file(XTThreadPtr self __attribute__((unused)), XTOpenDirPtr od)
+xtBool xt_dir_is_file(XTThreadPtr self, XTOpenDirPtr od)
 {
+	(void) self;
 #ifdef XT_WIN
 	if (od->od_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)
 		return FALSE;
@@ -1228,7 +1257,7 @@ xtPublic XTMapFilePtr xt_open_fmap(XTThreadPtr self, char *file, size_t grow_siz
 			/* NULL is the value returned on error! */
 			mm->mm_mapdes = NULL;
 #endif
-			xt_rwmutex_init_with_autoname(self, &mm->mm_lock);
+			FILE_MAP_INIT_LOCK(self, &mm->mm_lock);
 			mm->mm_length = fs_seek_eof(self, map->fr_file->fil_filedes, map->fr_file);
 			if (sizeof(size_t) == 4 && mm->mm_length >= (off_t) 0xFFFFFFFF)
 				xt_throw_ixterr(XT_CONTEXT, XT_ERR_FILE_TOO_LONG, map->fr_file->fil_path);
@@ -1257,21 +1286,19 @@ xtPublic XTMapFilePtr xt_open_fmap(XTThreadPtr self, char *file, size_t grow_siz
 
 xtPublic void xt_close_fmap(XTThreadPtr self, XTMapFilePtr map)
 {
+	ASSERT_NS(!map->mf_slock_count);
 	if (map->fr_file) {
-		xt_fs_release_file(self, map->fr_file);
-
 		xt_sl_lock(self, fs_globals.fsg_open_files);
-		pushr_(xt_sl_unlock, fs_globals.fsg_open_files);
-		
+		pushr_(xt_sl_unlock, fs_globals.fsg_open_files);		
 		map->fr_file->fil_handle_count--;
-		if (!map->fr_file->fil_handle_count)
-			fs_free_file(self, NULL, &map->fr_file);
-
+		if (!map->fr_file->fil_handle_count) {
+			fs_close_fmap(self, map->fr_file->fil_memmap);
+			map->fr_file->fil_memmap = NULL;
+		}
 		freer_();
-
+		
+		xt_fs_release_file(self, map->fr_file);
 		map->fr_file = NULL;
-
-
 	}
 	map->mf_memmap = NULL;
 	xt_free(self, map);
@@ -1367,16 +1394,19 @@ static xtBool fs_remap_file(XTMapFilePtr map, off_t offset, size_t size, XTIOSta
 xtPublic xtBool xt_pwrite_fmap(XTMapFilePtr map, off_t offset, size_t size, void *data, XTIOStatsPtr stat, XTThreadPtr thread)
 {
 	XTFileMemMapPtr mm = map->mf_memmap;
+#ifndef FILE_MAP_USE_PTHREAD_RW
 	xtThreadID		thd_id = thread->t_id;
+#endif
 
 #ifdef DEBUG_TRACE_MAP_IO
 	xt_trace("/* %s */ pbxt_fmap_writ(\"%s\", %lu, %lu);\n", xt_trace_clock_diff(NULL), map->fr_file->fil_path, (u_long) offset, (u_long) size);
 #endif
-	xt_rwmutex_slock(&mm->mm_lock, thd_id);
+	ASSERT_NS(!map->mf_slock_count);
+	FILE_MAP_READ_LOCK(&mm->mm_lock, thd_id);
 	if (!mm->mm_start || offset + (off_t) size > mm->mm_length) {
-		xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 
-		xt_rwmutex_xlock(&mm->mm_lock, thd_id);
+		FILE_MAP_WRITE_LOCK(&mm->mm_lock, thd_id);
 		if (!fs_remap_file(map, offset, size, stat))
 			goto failed;
 	}
@@ -1396,29 +1426,32 @@ xtPublic xtBool xt_pwrite_fmap(XTMapFilePtr map, off_t offset, size_t size, void
 	memcpy(mm->mm_start + offset, data, size);
 #endif
 
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	stat->ts_write += size;
 	return OK;
 
 	failed:
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	return FAILED;
 }
 
 xtPublic xtBool xt_pread_fmap_4(XTMapFilePtr map, off_t offset, xtWord4 *value, XTIOStatsPtr stat, XTThreadPtr thread)
 {
 	XTFileMemMapPtr	mm = map->mf_memmap;
+#ifndef FILE_MAP_USE_PTHREAD_RW
 	xtThreadID		thd_id = thread->t_id;
+#endif
 
 #ifdef DEBUG_TRACE_MAP_IO
 	xt_trace("/* %s */ pbxt_fmap_read_4(\"%s\", %lu, 4);\n", xt_trace_clock_diff(NULL), map->fr_file->fil_path, (u_long) offset);
 #endif
-	xt_rwmutex_slock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_READ_LOCK(&mm->mm_lock, thd_id);
 	if (!mm->mm_start) {
-		xt_rwmutex_unlock(&mm->mm_lock, thd_id);
-		xt_rwmutex_xlock(&mm->mm_lock, thd_id);
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
+		FILE_MAP_WRITE_LOCK(&mm->mm_lock, thd_id);
 		if (!fs_remap_file(map, 0, 0, stat)) {
-			xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+			FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 			return FAILED;
 		}
 	}
@@ -1436,7 +1469,7 @@ xtPublic xtBool xt_pread_fmap_4(XTMapFilePtr map, off_t offset, xtWord4 *value,
 		}
 		__except(EXCEPTION_EXECUTE_HANDLER)
 		{
-			xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+			FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 			return xt_register_ferrno(XT_REG_CONTEXT, GetExceptionCode(), xt_file_path(map));
 		}
 #else
@@ -1444,7 +1477,8 @@ xtPublic xtBool xt_pread_fmap_4(XTMapFilePtr map, off_t offset, xtWord4 *value,
 #endif
 	}
 
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	stat->ts_read += 4;
 	return OK;
 }
@@ -1452,7 +1486,9 @@ xtPublic xtBool xt_pread_fmap_4(XTMapFilePtr map, off_t offset, xtWord4 *value,
 xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_t min_size, void *data, size_t *red_size, XTIOStatsPtr stat, XTThreadPtr thread)
 {
 	XTFileMemMapPtr	mm = map->mf_memmap;
+#ifndef FILE_MAP_USE_PTHREAD_RW
 	xtThreadID		thd_id = thread->t_id;
+#endif
 	size_t			tfer;
 
 #ifdef DEBUG_TRACE_MAP_IO
@@ -1461,6 +1497,8 @@ xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_
 	/* NOTE!! The file map may already be locked,
 	 * by a call to xt_lock_fmap_ptr()!
 	 *
+	 * 20.05.2009: This problem should be fixed now with mf_slock_count!
+	 *
 	 * This can occur during a sequential scan:
 	 * xt_pread_fmap()  Line 1330
 	 * XTTabCache::tc_read_direct()  Line 361
@@ -1491,13 +1529,16 @@ xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_
 	 * As a result, the slock must be able to handle
 	 * nested calls to lock/unlock.
 	 */
-	xt_rwmutex_slock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_READ_LOCK(&mm->mm_lock, thd_id);
 	tfer = size;
 	if (!mm->mm_start) {
-		xt_rwmutex_unlock(&mm->mm_lock, thd_id);
-		xt_rwmutex_xlock(&mm->mm_lock, thd_id);
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
+		ASSERT_NS(!map->mf_slock_count);
+		FILE_MAP_WRITE_LOCK(&mm->mm_lock, thd_id);
 		if (!fs_remap_file(map, 0, 0, stat)) {
-			xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+			if (!map->mf_slock_count)
+				FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 			return FAILED;
 		}
 	}
@@ -1514,7 +1555,8 @@ xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_
 		}
 		__except(EXCEPTION_EXECUTE_HANDLER)
 		{
-			xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+			if (!map->mf_slock_count)
+				FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 			return xt_register_ferrno(XT_REG_CONTEXT, GetExceptionCode(), xt_file_path(map));
 		}
 #else
@@ -1522,7 +1564,8 @@ xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_
 #endif
 	}
 
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	if (tfer < min_size)
 		return xt_register_ferrno(XT_REG_CONTEXT, ESPIPE, xt_file_path(map));
 
@@ -1535,18 +1578,23 @@ xtPublic xtBool xt_pread_fmap(XTMapFilePtr map, off_t offset, size_t size, size_
 xtPublic xtBool xt_flush_fmap(XTMapFilePtr map, XTIOStatsPtr stat, XTThreadPtr thread)
 {
 	XTFileMemMapPtr	mm = map->mf_memmap;
+#ifndef FILE_MAP_USE_PTHREAD_RW
 	xtThreadID		thd_id = thread->t_id;
+#endif
 	xtWord8			s;
 
 #ifdef DEBUG_TRACE_MAP_IO
 	xt_trace("/* %s */ pbxt_fmap_sync(\"%s\");\n", xt_trace_clock_diff(NULL), map->fr_file->fil_path);
 #endif
-	xt_rwmutex_slock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_READ_LOCK(&mm->mm_lock, thd_id);
 	if (!mm->mm_start) {
-		xt_rwmutex_unlock(&mm->mm_lock, thd_id);
-		xt_rwmutex_xlock(&mm->mm_lock, thd_id);
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
+		ASSERT_NS(!map->mf_slock_count);
+		FILE_MAP_WRITE_LOCK(&mm->mm_lock, thd_id);
 		if (!fs_remap_file(map, 0, 0, stat)) {
-			xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+			if (!map->mf_slock_count)
+				FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 			return FAILED;
 		}
 	}
@@ -1562,7 +1610,8 @@ xtPublic xtBool xt_flush_fmap(XTMapFilePtr map, XTIOStatsPtr stat, XTThreadPtr t
 		goto failed;
 	}
 #endif
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	s = stat->ts_flush_start;
 	stat->ts_flush_start = 0;
 	stat->ts_flush_time += xt_trace_clock() - s;
@@ -1570,22 +1619,27 @@ xtPublic xtBool xt_flush_fmap(XTMapFilePtr map, XTIOStatsPtr stat, XTThreadPtr t
 	return OK;
 
 	failed:
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	s = stat->ts_flush_start;
 	stat->ts_flush_start = 0;
 	stat->ts_flush_time += xt_trace_clock() - s;
 	return FAILED;
 }
 
-xtPublic xtWord1 *xt_lock_fmap_ptr(XTMapFilePtr map, off_t offset, size_t size, XTIOStatsPtr stat, XTThreadPtr XT_UNUSED(thread))
+xtPublic xtWord1 *xt_lock_fmap_ptr(XTMapFilePtr map, off_t offset, size_t size, XTIOStatsPtr stat, XTThreadPtr thread)
 {
 	XTFileMemMapPtr	mm = map->mf_memmap;
+#ifndef FILE_MAP_USE_PTHREAD_RW
 	xtThreadID		thd_id = thread->t_id;
+#endif
 
-	xt_rwmutex_slock(&mm->mm_lock, thd_id);
+	if (!map->mf_slock_count)
+		FILE_MAP_READ_LOCK(&mm->mm_lock, thd_id);
+	map->mf_slock_count++;
 	if (!mm->mm_start) {
-		xt_rwmutex_unlock(&mm->mm_lock, thd_id);
-		xt_rwmutex_xlock(&mm->mm_lock, thd_id);
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
+		FILE_MAP_WRITE_LOCK(&mm->mm_lock, thd_id);
 		if (!fs_remap_file(map, 0, 0, stat))
 			goto failed;
 	}
@@ -1599,13 +1653,17 @@ xtPublic xtWord1 *xt_lock_fmap_ptr(XTMapFilePtr map, off_t offset, size_t size,
 	return mm->mm_start + offset;
 
 	failed:
-	xt_rwmutex_unlock(&mm->mm_lock, thd_id);
+	map->mf_slock_count--;
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&mm->mm_lock, thd_id);
 	return NULL;
 }
 
 xtPublic void xt_unlock_fmap_ptr(XTMapFilePtr map, XTThreadPtr thread)
 {
-	xt_rwmutex_unlock(&map->mf_memmap->mm_lock, thread->t_id);
+	map->mf_slock_count--;
+	if (!map->mf_slock_count)
+		FILE_MAP_UNLOCK(&map->mf_memmap->mm_lock, thread->t_id);
 }
 
 /* ----------------------------------------------------------------------
diff --git a/storage/pbxt/src/filesys_xt.h b/storage/pbxt/src/filesys_xt.h
index ebc4f474fc9..585ed9bd8d0 100644
--- a/storage/pbxt/src/filesys_xt.h
+++ b/storage/pbxt/src/filesys_xt.h
@@ -76,13 +76,60 @@ xtBool			xt_fs_rename(struct XTThread *self, char *from_path, char *to_path);
 #define XT_NULL_FD	(-1)
 #endif
 
+/* Note, this lock must be re-entrant,
+ * The only lock that satifies this is
+ * FILE_MAP_USE_RWMUTEX!
+ *
+ * 20.05.2009: This problem should be fixed now with mf_slock_count!
+ *
+ * The lock need no longer be re-entrant
+ */
+#ifdef XT_NO_ATOMICS
+#define FILE_MAP_USE_PTHREAD_RW
+#else
+//#define FILE_MAP_USE_RWMUTEX
+//#define FILE_MAP_USE_PTHREAD_RW
+//#define IDX_USE_SPINXSLOCK
+#define FILE_MAP_USE_XSMUTEX
+#endif
+
+#ifdef FILE_MAP_USE_XSMUTEX
+#define FILE_MAP_LOCK_TYPE				XTXSMutexRec
+#define FILE_MAP_INIT_LOCK(s, i)		xt_xsmutex_init_with_autoname(s, i)
+#define FILE_MAP_FREE_LOCK(s, i)		xt_xsmutex_free(s, i)	
+#define FILE_MAP_READ_LOCK(i, o)		xt_xsmutex_slock(i, o)
+#define FILE_MAP_WRITE_LOCK(i, o)		xt_xsmutex_xlock(i, o)
+#define FILE_MAP_UNLOCK(i, o)			xt_xsmutex_unlock(i, o)
+#elif defined(FILE_MAP_USE_PTHREAD_RW)
+#define FILE_MAP_LOCK_TYPE				xt_rwlock_type
+#define FILE_MAP_INIT_LOCK(s, i)		xt_init_rwlock(s, i)
+#define FILE_MAP_FREE_LOCK(s, i)		xt_free_rwlock(i)	
+#define FILE_MAP_READ_LOCK(i, o)		xt_slock_rwlock_ns(i)
+#define FILE_MAP_WRITE_LOCK(i, o)		xt_xlock_rwlock_ns(i)
+#define FILE_MAP_UNLOCK(i, o)			xt_unlock_rwlock_ns(i)
+#elif defined(FILE_MAP_USE_RWMUTEX)
+#define FILE_MAP_LOCK_TYPE				XTRWMutexRec
+#define FILE_MAP_INIT_LOCK(s, i)		xt_rwmutex_init_with_autoname(s, i)
+#define FILE_MAP_FREE_LOCK(s, i)		xt_rwmutex_free(s, i)	
+#define FILE_MAP_READ_LOCK(i, o)		xt_rwmutex_slock(i, o)
+#define FILE_MAP_WRITE_LOCK(i, o)		xt_rwmutex_xlock(i, o)
+#define FILE_MAP_UNLOCK(i, o)			xt_rwmutex_unlock(i, o)
+#elif defined(FILE_MAP_USE_SPINXSLOCK)
+#define FILE_MAP_LOCK_TYPE				XTSpinXSLockRec
+#define FILE_MAP_INIT_LOCK(s, i)		xt_spinxslock_init_with_autoname(s, i)
+#define FILE_MAP_FREE_LOCK(s, i)		xt_spinxslock_free(s, i)	
+#define FILE_MAP_READ_LOCK(i, o)		xt_spinxslock_slock(i, o)
+#define FILE_MAP_WRITE_LOCK(i, o)		xt_spinxslock_xlock(i, o)
+#define FILE_MAP_UNLOCK(i, o)			xt_spinxslock_unlock(i, o)
+#endif
+
 typedef struct XTFileMemMap {
 	xtWord1				*mm_start;			/* The in-memory start of the map. */
 #ifdef XT_WIN
 	HANDLE				mm_mapdes;
 #endif
 	off_t				mm_length;			/* The length of the file map. */
-	XTRWMutexRec		mm_lock;			/* The file map R/W lock. */
+	FILE_MAP_LOCK_TYPE	mm_lock;			/* The file map R/W lock. */
 	size_t				mm_grow_size;		/* The amount by which the map file is increased. */
 } XTFileMemMapRec, *XTFileMemMapPtr;
 
@@ -127,6 +174,9 @@ xtBool			xt_pwrite_file(XTOpenFilePtr of, off_t offset, size_t size, void *data,
 xtBool			xt_pread_file(XTOpenFilePtr of, off_t offset, size_t size, size_t min_size, void *data, size_t *red_size, struct XTIOStats *timer, struct XTThread *thread);
 xtBool			xt_flush_file(XTOpenFilePtr of, struct XTIOStats *timer, struct XTThread *thread);
 
+xtBool			xt_lock_file_ptr(XTOpenFilePtr of, xtWord1 **data, off_t offset, size_t size, struct XTIOStats *timer, struct XTThread *thread);
+void			xt_unlock_file_ptr(XTOpenFilePtr of, xtWord1 *data, struct XTThread *thread);
+
 typedef struct XTOpenDir {
 	char				*od_path;
 #ifdef XT_WIN
@@ -134,8 +184,14 @@ typedef struct XTOpenDir {
 	WIN32_FIND_DATA		od_data;
 #else
 	char				*od_filter;
-	struct dirent		od_entry;
 	DIR					*od_dir;
+	/* WARNING: Solaris requires od_entry.d_name member to have size at least as returned
+	 * by pathconf() function on per-directory basis. This makes it impossible to statically
+	 * pre-set the size. So xt_dir_open on Solaris dynamically allocates space as needed. 
+	 *
+	 * This also means that the od_entry member should always be last in the XTOpenDir structure.
+	 */
+	struct dirent		od_entry;
 #endif
 } XTOpenDirRec, *XTOpenDirPtr;
 
@@ -147,6 +203,7 @@ xtBool			xt_dir_is_file(struct XTThread *self, XTOpenDirPtr od);
 off_t			xt_dir_file_size(struct XTThread *self, XTOpenDirPtr od);
 
 typedef struct XTMapFile : public XTFileRef {
+	u_int				mf_slock_count;
 	XTFileMemMapPtr		mf_memmap;
 } XTMapFileRec, *XTMapFilePtr;
 
diff --git a/storage/pbxt/src/ha_pbxt.cc b/storage/pbxt/src/ha_pbxt.cc
index 86cea271e0d..e266e2325ee 100644
--- a/storage/pbxt/src/ha_pbxt.cc
+++ b/storage/pbxt/src/ha_pbxt.cc
@@ -91,20 +91,22 @@ extern "C" char **session_query(Session *session);
 //#define PRINT_STATEMENTS
 #endif
 
+#ifndef DRIZZLED
 static handler	*pbxt_create_handler(handlerton *hton, TABLE_SHARE *table, MEM_ROOT *mem_root);
 static int		pbxt_init(void *p);
 static int		pbxt_end(void *p);
-#ifndef DRIZZLED
 static int		pbxt_panic(handlerton *hton, enum ha_panic_function flag);
-#endif
 static void		pbxt_drop_database(handlerton *hton, char *path);
 static int		pbxt_close_connection(handlerton *hton, THD* thd);
 static int		pbxt_commit(handlerton *hton, THD *thd, bool all);
 static int		pbxt_rollback(handlerton *hton, THD *thd, bool all);
+#endif
 static void		ha_aquire_exclusive_use(XTThreadPtr self, XTSharePtr share, ha_pbxt *mine);
 static void		ha_release_exclusive_use(XTThreadPtr self, XTSharePtr share);
 static void		ha_close_open_tables(XTThreadPtr self, XTSharePtr share, ha_pbxt *mine);
 
+extern void		xt_xres_start_database_recovery(XTThreadPtr self);
+
 #ifdef TRACE_STATEMENTS
 
 #ifdef PRINT_STATEMENTS
@@ -167,7 +169,7 @@ xtBool					pbxt_crash_debug = FALSE;
 /* Variables for pbxt share methods */
 static xt_mutex_type	pbxt_database_mutex;		// Prevent a database from being opened while it is being dropped
 static XTHashTabPtr		pbxt_share_tables;			// Hash used to track open tables
-static XTDatabaseHPtr	pbxt_database = NULL;		// The global open database
+XTDatabaseHPtr			pbxt_database = NULL;		// The global open database
 static char				*pbxt_index_cache_size;
 static char				*pbxt_record_cache_size;
 static char				*pbxt_log_cache_size;
@@ -178,6 +180,7 @@ static char				*pbxt_checkpoint_frequency;
 static char				*pbxt_data_log_threshold;
 static char				*pbxt_data_file_grow_size;
 static char				*pbxt_row_file_grow_size;
+static int				pbxt_max_threads;
 
 #ifdef DEBUG
 #define XT_SHARE_LOCK_WAIT		5000
@@ -454,7 +457,7 @@ xtPublic void xt_ha_close_global_database(XTThreadPtr self)
  * operation to make sure it does not occur while
  * some other thread is doing a "closeall".
  */
-xtPublic void xt_ha_open_database_of_table(XTThreadPtr self, XTPathStrPtr table_path __attribute__((unused)))
+xtPublic void xt_ha_open_database_of_table(XTThreadPtr self, XTPathStrPtr XT_UNUSED(table_path))
 {
 #ifdef XT_USE_GLOBAL_DB
 	if (!self->st_database) {
@@ -574,7 +577,7 @@ xtPublic XTThreadPtr xt_ha_thd_to_self(THD *thd)
 }
 
 /* The first bit is 1. */
-static u_int ha_get_max_bit(MY_BITMAP *map)
+static u_int ha_get_max_bit(MX_BITMAP *map)
 {
 	my_bitmap_map	*data_ptr = map->bitmap;
 	my_bitmap_map	*end_ptr = map->last_word_ptr;
@@ -676,7 +679,7 @@ xtPublic int xt_ha_pbxt_to_mysql_error(int xt_err)
 	return(-1);			// Unknown error
 }
 
-xtPublic int xt_ha_pbxt_thread_error_for_mysql(THD *thd __attribute__((unused)), const XTThreadPtr self, int ignore_dup_key)
+xtPublic int xt_ha_pbxt_thread_error_for_mysql(THD *XT_UNUSED(thd), const XTThreadPtr self, int ignore_dup_key)
 {
 	int xt_err = self->t_exception.e_xt_err;
 
@@ -979,9 +982,13 @@ static void ha_exit(XTThreadPtr self)
 /*
  * Outout the PBXT status. Return FALSE on error.
  */
-static bool pbxt_show_status(handlerton *hton __attribute__((unused)), THD* thd, 
+#ifdef DRIZZLED
+bool PBXTStorageEngine::show_status(Session *thd, stat_print_fn *stat_print, enum ha_stat_type)
+#else
+static bool pbxt_show_status(handlerton *XT_UNUSED(hton), THD* thd, 
                           stat_print_fn* stat_print,
-                          enum ha_stat_type stat_type __attribute__((unused)))
+                          enum ha_stat_type XT_UNUSED(stat_type))
+#endif
 {
 	XTThreadPtr			self;	
 	int					err = 0;
@@ -997,6 +1004,9 @@ static bool pbxt_show_status(handlerton *hton __attribute__((unused)), THD* thd,
 	xt_trace("// %s - dump\n", xt_trace_clock_diff(NULL));
 	xt_dump_trace();
 #endif
+#ifdef XT_TRACK_CONNECTIONS
+	xt_dump_conn_tracking();
+#endif
 
 	try_(a) {
 		myxt_get_status(self, &strbuf);
@@ -1020,7 +1030,11 @@ static bool pbxt_show_status(handlerton *hton __attribute__((unused)), THD* thd,
  *
  * return 1 on error, else 0.
  */
+#ifdef DRIZZLED
+static int pbxt_init(PluginRegistry &registry)
+#else
 static int pbxt_init(void *p)
+#endif
 {
 	int init_err = 0;
 
@@ -1045,22 +1059,22 @@ static int pbxt_init(void *p)
 
  		xt_p_mutex_init_with_autoname(&pbxt_database_mutex, NULL);
 
+#ifdef DRIZZLED
+		pbxt_hton= new PBXTStorageEngine(std::string("PBXT"));
+		registry.add(pbxt_hton);
+#else
 		pbxt_hton = (handlerton *) p;
 		pbxt_hton->state = SHOW_OPTION_YES;
-#ifndef DRIZZLED
 		pbxt_hton->db_type = DB_TYPE_PBXT; // Wow! I have my own!
-#endif
 		pbxt_hton->close_connection = pbxt_close_connection; /* close_connection, cleanup thread related data. */
 		pbxt_hton->commit = pbxt_commit; /* commit */
 		pbxt_hton->rollback = pbxt_rollback; /* rollback */
 		pbxt_hton->create = pbxt_create_handler; /* Create a new handler */
 		pbxt_hton->drop_database = pbxt_drop_database; /* Drop a database */
-#ifndef DRIZZLED
 		pbxt_hton->panic = pbxt_panic; /* Panic call */
-#endif
 		pbxt_hton->show_status = pbxt_show_status;
 		pbxt_hton->flags = HTON_NO_FLAGS; /* HTON_CAN_RECREATE - Without this flags TRUNCATE uses delete_all_rows() */
-
+#endif
 		if (!xt_init_logging())					/* Initialize logging */
 			goto error_1;
 
@@ -1082,9 +1096,13 @@ static int pbxt_init(void *p)
 		 * +1 Free'er thread
 		 * +1 Temporary thread (e.g. TempForClose, TempForEnd)
 		 */
-		self = xt_init_threading(max_connections + 7);				/* Create the main self: */
+#ifndef DRIZZLED
+		if (pbxt_max_threads == 0)
+			pbxt_max_threads = max_connections + 7;
+#endif
+		self = xt_init_threading(pbxt_max_threads);				/* Create the main self: */
 		if (!self)
-			goto error_4;
+			goto error_3;
 
  		pbxt_inited = true;
 
@@ -1148,9 +1166,7 @@ static int pbxt_init(void *p)
 							xt_throw(self);
 					}
 
-					xt_open_database(self, mysql_real_data_home, TRUE);
-					pbxt_database = self->st_database;
-					xt_heap_reference(self, pbxt_database);
+					xt_xres_start_database_recovery(self);
 				}
 				catch_(b) {
 					if (!curr_thd && thd)
@@ -1205,32 +1221,37 @@ static int pbxt_init(void *p)
 			 * I have to stop the freeer here because it was
 			 * started before opening the database.
 			 */
-			pbxt_call_exit(self);
-			pbxt_inited = FALSE;
-			xt_exit_threading(self);
-			goto error_4;
+
+			/* {FREEER-HANG-ON-INIT-ERROR}
+			 * pbxt_init is called with LOCK_plugin and if it fails and tries to exit
+			 * the freeer here it hangs because the freeer calls THD::~THD which tries
+			 * to aquire the same lock and hangs. OTOH MySQL calls pbxt_end() after
+			 * an unsuccessful call to pbxt_init, so we defer cleaup, except 
+			 * releasing 'self'
+			 */
+			xt_free_thread(self);
+			goto error_3;
 		}
 		xt_free_thread(self);
  	}
 	XT_RETURN(init_err);
 
-	error_4:
-	xt_exit_memory();
-
 	error_3:
 #ifdef XT_STREAMING
 	xt_exit_streaming();
 
 	error_2:
 #endif
-	xt_exit_logging();
 
 	error_1:
-	xt_p_mutex_destroy(&pbxt_database_mutex);		
 	XT_RETURN(1);
 }
 
-static int pbxt_end(void *p __attribute__((unused)))
+#ifdef DRIZZLED
+static int pbxt_end(PluginRegistry &registry)
+#else
+static int pbxt_end(void *)
+#endif
 {
 	XTThreadPtr		self;
 	int				err = 0;
@@ -1241,7 +1262,7 @@ static int pbxt_end(void *p __attribute__((unused)))
 		XTExceptionRec	e;
 
 		/* This flag also means "shutting down". */
-		pbxt_inited = FALSE;
+		pbxt_inited = FALSE; 
 		self = xt_create_thread("TempForEnd", FALSE, TRUE, &e);
 		if (self) {
 			self->t_main = TRUE;
@@ -1249,6 +1270,9 @@ static int pbxt_end(void *p __attribute__((unused)))
 		}
 	}
 
+#ifdef DRIZZLED
+	registry.remove(pbxt_hton);
+#endif
 	XT_RETURN(err);
 }
 
@@ -1262,8 +1286,14 @@ static int pbxt_panic(handlerton *hton, enum ha_panic_function flag)
 /*
  * Kill the PBXT thread associated with the MySQL thread.
  */
+#ifdef DRIZZLED
+int PBXTStorageEngine::close_connection(Session *thd)
+{
+	PBXTStorageEngine * const hton = this;
+#else
 static int pbxt_close_connection(handlerton *hton, THD* thd)
 {
+#endif
 	XTThreadPtr		self;
 #ifdef XT_STREAMING
 	XTExceptionRec	e;
@@ -1290,7 +1320,11 @@ static int pbxt_close_connection(handlerton *hton, THD* thd)
  * when the last PBXT table was removed from the 
  * database.
  */
-static void pbxt_drop_database(handlerton *hton __attribute__((unused)), char *path __attribute__((unused)))
+#ifdef DRIZZLED
+void PBXTStorageEngine::drop_database(char *)
+#else
+static void pbxt_drop_database(handlerton *XT_UNUSED(hton), char *XT_UNUSED(path))
+#endif
 {
 	XT_TRACE_CALL();
 }
@@ -1317,8 +1351,14 @@ static void pbxt_drop_database(handlerton *hton __attribute__((unused)), char *p
  * pbxt_thr is a pointer the the PBXT thread structure.
  *
  */
+#ifdef DRIZZLED
+int PBXTStorageEngine::commit(Session *thd, bool all)
+{
+	PBXTStorageEngine * const hton = this;
+#else
 static int pbxt_commit(handlerton *hton, THD *thd, bool all)
 {
+#endif
 	int			err = 0;
 	XTThreadPtr	self;
 
@@ -1343,8 +1383,14 @@ static int pbxt_commit(handlerton *hton, THD *thd, bool all)
 	return err;
 }
 
+#ifdef DRIZZLED
+int PBXTStorageEngine::rollback(Session *thd, bool all)
+{
+	PBXTStorageEngine * const hton = this;
+#else
 static int pbxt_rollback(handlerton *hton, THD *thd, bool all)
 {
+#endif
 	int			err = 0;
 	XTThreadPtr	self;
 
@@ -1377,8 +1423,14 @@ static int pbxt_rollback(handlerton *hton, THD *thd, bool all)
 	return 0;
 }
 
+#ifdef DRIZZLED
+handler *PBXTStorageEngine::create(TABLE_SHARE *table, MEM_ROOT *mem_root)
+{
+	PBXTStorageEngine * const hton = this;
+#else
 static handler *pbxt_create_handler(handlerton *hton, TABLE_SHARE *table, MEM_ROOT *mem_root)
 {
+#endif
 	if (table && XTSystemTableShare::isSystemTable(table->path.str))
 		return new (mem_root) ha_xtsys(hton, table);
 	else
@@ -1513,7 +1565,11 @@ static void ha_close_open_tables(XTThreadPtr self, XTSharePtr share, ha_pbxt *mi
 	freer_(); // xt_unlock_mutex(share->sh_ex_mutex)
 }
 
-static void ha_release_exclusive_use(XTThreadPtr self __attribute__((unused)), XTSharePtr share)
+#ifdef PBXT_ALLOW_PRINTING
+static void ha_release_exclusive_use(XTThreadPtr self, XTSharePtr share)
+#else
+static void ha_release_exclusive_use(XTThreadPtr XT_UNUSED(self), XTSharePtr share)
+#endif
 {
 	XT_PRINT1(self, "ha_release_exclusive_use %s PBXT X UNLOCK\n", share->sh_table_path->ps_path);
 	xt_lock_mutex_ns((xt_mutex_type *) share->sh_ex_mutex);
@@ -1629,11 +1685,23 @@ ST_FIELD_INFO pbxt_statistics_fields_info[]=
 	{ 0,		0,	MYSQL_TYPE_STRING,		0, 0, 0, SKIP_OPEN_TABLE}
 };
 
+#ifdef DRIZZLED
+static InfoSchemaTable	*pbxt_statistics_table;
+
+int pbxt_init_statitics(PluginRegistry &registry)
+#else
 int pbxt_init_statitics(void *p)
+#endif
 {
-	ST_SCHEMA_TABLE *schema = (ST_SCHEMA_TABLE *) p;
-	schema->fields_info = pbxt_statistics_fields_info;
-	schema->fill_table = pbxt_statistics_fill_table;
+#ifdef DRIZZLED
+	pbxt_statistics_table = (InfoSchemaTable *)xt_calloc_ns(sizeof(InfoSchemaTable));
+	pbxt_statistics_table->table_name= "PBXT_STATISTICS";
+	registry.add(pbxt_statistics_table);
+#else
+	ST_SCHEMA_TABLE *pbxt_statistics_table = (ST_SCHEMA_TABLE *) p;
+#endif
+	pbxt_statistics_table->fields_info = pbxt_statistics_fields_info;
+	pbxt_statistics_table->fill_table = pbxt_statistics_fill_table;
 
 #if defined(XT_WIN) && defined(XT_COREDUMP)
 	void register_crash_filter();
@@ -1645,8 +1713,16 @@ int pbxt_init_statitics(void *p)
 	return 0;
 }
 
-int pbxt_exit_statitics(void *p __attribute__((unused)))
+#ifdef DRIZZLED
+int pbxt_exit_statitics(PluginRegistry &registry)
+#else
+int pbxt_exit_statitics(void *XT_UNUSED(p))
+#endif
 {
+#ifdef DRIZZLED
+	registry.remove(pbxt_statistics_table);
+	xt_free_ns(pbxt_statistics_table);
+#endif
 	return(0);
 }
 
@@ -1765,7 +1841,7 @@ MX_TABLE_TYPES_T ha_pbxt::table_flags() const
  */
 #define FLAGS_ARE_READ_DYNAMICALLY
 
-MX_ULONG_T ha_pbxt::index_flags(uint inx __attribute__((unused)), uint part __attribute__((unused)), bool all_parts __attribute__((unused))) const
+MX_ULONG_T ha_pbxt::index_flags(uint XT_UNUSED(inx), uint XT_UNUSED(part), bool XT_UNUSED(all_parts)) const
 {
 	/* It would be nice if the dynamic version of this function works,
 	 * but it does not. MySQL loads this information when the table is openned,
@@ -1876,7 +1952,7 @@ void ha_pbxt::internal_close(THD *thd, struct XTThread *self)
  * Called from handler.cc by handler::ha_open(). The server opens all tables by
  * calling ha_open() which then calls the handler specific open().
  */
-int ha_pbxt::open(const char *table_path, int mode __attribute__((unused)), uint test_if_locked __attribute__((unused)))
+int ha_pbxt::open(const char *table_path, int XT_UNUSED(mode), uint XT_UNUSED(test_if_locked))
 {
 	THD			*thd = current_thd;
 	int			err = 0;
@@ -2104,9 +2180,9 @@ void ha_pbxt::init_auto_increment(xtWord8 min_auto_inc)
 }
 
 void ha_pbxt::get_auto_increment(MX_ULONGLONG_T offset, MX_ULONGLONG_T increment,
-                                 MX_ULONGLONG_T nb_desired_values __attribute__((unused)),
+                                 MX_ULONGLONG_T XT_UNUSED(nb_desired_values),
                                  MX_ULONGLONG_T *first_value,
-                                 MX_ULONGLONG_T *nb_reserved_values __attribute__((unused)))
+                                 MX_ULONGLONG_T *nb_reserved_values)
 {
 	register XTTableHPtr	tab;
 	MX_ULONGLONG_T			nr, nr_less_inc;
@@ -2491,7 +2567,7 @@ int ha_pbxt::delete_row(const byte * buf)
  * commit;
  */
 
-int ha_pbxt::xt_index_in_range(register XTOpenTablePtr ot __attribute__((unused)), register XTIndexPtr ind,
+int ha_pbxt::xt_index_in_range(register XTOpenTablePtr XT_UNUSED(ot), register XTIndexPtr ind,
 	register XTIdxSearchKeyPtr search_key, xtWord1 *buf)
 {
 	/* If search key is given, this means we want an exact match. */
@@ -2698,7 +2774,7 @@ int ha_pbxt::xt_index_prev_read(XTOpenTablePtr ot, XTIndexPtr ind, xtBool key_on
 	return ha_log_pbxt_thread_error_for_mysql(FALSE);
 }
 
-int ha_pbxt::index_init(uint idx, bool sorted __attribute__((unused)))
+int ha_pbxt::index_init(uint idx, bool XT_UNUSED(sorted))
 {
 	XTIndexPtr ind;
 
@@ -2715,7 +2791,8 @@ int ha_pbxt::index_init(uint idx, bool sorted __attribute__((unused)))
 
 	/* The number of columns required: */
 	if (pb_open_tab->ot_is_modify) {
-		pb_open_tab->ot_cols_req = table->read_set->n_bits;
+
+		pb_open_tab->ot_cols_req = table->read_set->MX_BIT_SIZE();
 #ifdef XT_PRINT_INDEX_OPT
 		ind = (XTIndexPtr) pb_share->sh_dic_keys[idx];
 
@@ -2764,10 +2841,10 @@ int ha_pbxt::index_init(uint idx, bool sorted __attribute__((unused)))
 		 * seem to have this problem!
 		 */
 		ind = (XTIndexPtr) pb_share->sh_dic_keys[idx];
-		if (bitmap_is_subset(table->read_set, &ind->mi_col_map))
+		if (MX_BIT_IS_SUBSET(table->read_set, &ind->mi_col_map))
 			pb_key_read = TRUE;
 #ifdef XT_PRINT_INDEX_OPT
-		printf("index_init %s index %d cols req=%d/%d read_bits=%X write_bits=%X index_bits=%X converage=%d\n", pb_open_tab->ot_table->tab_name->ps_path, (int) idx, pb_open_tab->ot_cols_req, table->read_set->n_bits, (int) *table->read_set->bitmap, (int) *table->write_set->bitmap, (int) *ind->mi_col_map.bitmap, (int) (bitmap_is_subset(table->read_set, &ind->mi_col_map) != 0));
+		printf("index_init %s index %d cols req=%d/%d read_bits=%X write_bits=%X index_bits=%X converage=%d\n", pb_open_tab->ot_table->tab_name->ps_path, (int) idx, pb_open_tab->ot_cols_req, table->read_set->MX_BIT_SIZE(), (int) *table->read_set->bitmap, (int) *table->write_set->bitmap, (int) *ind->mi_col_map.bitmap, (int) (MX_BIT_IS_SUBSET(table->read_set, &ind->mi_col_map) != 0));
 #endif
 	}
 	
@@ -2845,7 +2922,7 @@ void ha_return_row(XTOpenTablePtr ot, u_int index)
 }
 #endif
 
-int ha_pbxt::index_read_xt(byte * buf, uint idx, const byte *key, uint key_len __attribute__((unused)), enum ha_rkey_function find_flag __attribute__((unused)))
+int ha_pbxt::index_read_xt(byte * buf, uint idx, const byte *key, uint key_len, enum ha_rkey_function find_flag)
 {
 	int					err = 0;
 	XTIndexPtr			ind;
@@ -2887,9 +2964,12 @@ int ha_pbxt::index_read_xt(byte * buf, uint idx, const byte *key, uint key_len _
 			xt_idx_prep_key(ind, &search_key, ((find_flag == HA_READ_AFTER_KEY) ? XT_SEARCH_AFTER_KEY : 0) | prefix, (xtWord1 *) key, key_len);
 			if (!xt_idx_search(pb_open_tab, ind, &search_key))
 				err = ha_log_pbxt_thread_error_for_mysql(pb_ignore_dup_key);
-			else
+			else {
 				err = xt_index_next_read(pb_open_tab, ind, pb_key_read,
 					(find_flag == HA_READ_KEY_EXACT || find_flag == HA_READ_PREFIX) ? &search_key : NULL, buf);
+				if (err == HA_ERR_END_OF_FILE && find_flag == HA_READ_AFTER_KEY)
+					err = HA_ERR_KEY_NOT_FOUND;			
+			}
 			break;
 	}
 
@@ -2913,13 +2993,13 @@ int ha_pbxt::index_read_xt(byte * buf, uint idx, const byte *key, uint key_len _
  * row if available. If the key value is null, begin at the first key of the
  * index.
  */
-int ha_pbxt::index_read(byte * buf, const byte * key, uint key_len __attribute__((unused)), enum ha_rkey_function find_flag __attribute__((unused)))
+int ha_pbxt::index_read(byte * buf, const byte * key, uint key_len, enum ha_rkey_function find_flag)
 {
 	//statistic_increment(ha_read_key_count,&LOCK_status);
 	return index_read_xt(buf, active_index, key, key_len, find_flag);
 }
 
-int ha_pbxt::index_read_idx(byte * buf, uint idx, const byte *key, uint key_len __attribute__((unused)), enum ha_rkey_function find_flag __attribute__((unused)))
+int ha_pbxt::index_read_idx(byte * buf, uint idx, const byte *key, uint key_len, enum ha_rkey_function find_flag)
 {
 	//statistic_increment(ha_read_key_count,&LOCK_status);
 	return index_read_xt(buf, idx, key, key_len, find_flag);
@@ -3147,9 +3227,24 @@ int ha_pbxt::rnd_init(bool scan)
 	XT_PRINT1(pb_open_tab->ot_thread, "ha_pbxt::rnd_init %s\n", pb_share->sh_table_path->ps_path);
 	XT_DISABLED_TRACE(("seq scan tx=%d\n", (int) pb_open_tab->ot_thread->st_xact_data->xd_start_xn_id));
 
+	/* Call xt_tab_seq_exit() to make sure the resources used by the previous
+	 * scan are freed. In particular make sure cache page ref count is decremented.
+	 * This is needed as rnd_init() can be called mulitple times w/o matching calls 
+	 * to rnd_end(). Our experience is that currently this is done in queries like:
+	 *
+	 * SELECT t1.c1,t2.c1 FROM t1 LEFT JOIN t2 USING (c1);
+	 * UPDATE t1 LEFT JOIN t2 USING (c1) SET t1.c1 = t2.c1 WHERE t1.c1 = t2.c1;
+	 *
+	 * when scanning inner tables. It is important to understand that in such case
+	 * multiple calls to rnd_init() are not semantically equal to a new query. For
+	 * example we cannot make row locks permanent as we do in rnd_end(), as 
+	 * ha_pbxt::unlock_row still can be called.
+	 */
+	xt_tab_seq_exit(pb_open_tab);
+
 	/* The number of columns required: */
 	if (pb_open_tab->ot_is_modify)
-		pb_open_tab->ot_cols_req = table->read_set->n_bits;
+		pb_open_tab->ot_cols_req = table->read_set->MX_BIT_SIZE();
 	else {
 		pb_open_tab->ot_cols_req = ha_get_max_bit(table->read_set);
 
@@ -3243,7 +3338,7 @@ int ha_pbxt::rnd_next(byte *buf)
  *
  * Called from filesort.cc, sql_select.cc, sql_delete.cc and sql_update.cc.
  */
-void ha_pbxt::position(const byte *record __attribute__((unused)))
+void ha_pbxt::position(const byte *XT_UNUSED(record))
 {
 	XT_TRACE_CALL();
 	ASSERT_NS(pb_ex_in_use);
@@ -3434,10 +3529,15 @@ int ha_pbxt::info(uint flag)
 #endif
 
 #endif // SAFE_MUTEX
+#ifdef DRIZZLED
+			set_prefix(share->keys_in_use, share->keys);
+			share->keys_for_keyread&= share->keys_in_use;
+#else
 			share->keys_in_use.set_prefix(share->keys);
 			//share->keys_in_use.intersect_extended(info.key_map);
 			share->keys_for_keyread.intersect(share->keys_in_use);
 			//share->db_record_offset = info.record_offset;
+#endif
 			for (u_int i = 0; i < share->keys; i++) {
 				ind = pb_share->sh_dic_keys[i];
 
@@ -3445,7 +3545,7 @@ int ha_pbxt::info(uint flag)
 				if (ind->mi_seg_count == 1 && (ind->mi_flags & HA_NOSAME))
 					rec_per_key = 1;
 				else {
-					
+					rec_per_key = 1;	
 				}
 				for (u_int j = 0; j < table->key_info[i].key_parts; j++)
 	 				table->key_info[i].rec_per_key[j] = (ulong) rec_per_key;
@@ -3570,6 +3670,8 @@ int ha_pbxt::extra(enum ha_extra_function operation)
 				if (pb_open_tab)
 					pb_open_tab->ot_table->tab_locks.xt_make_lock_permanent(pb_open_tab, &self->st_lock_list);
 			}
+			if (pb_open_tab)
+				pb_open_tab->ot_for_update = 0;
 			break;
 		case HA_EXTRA_KEYREAD:
 			/* This means we so not need to read the entire record. */
@@ -3706,6 +3808,12 @@ int ha_pbxt::delete_all_rows()
 		 */
 		ha_close_share(self, pb_share);
 
+		/* MySQL documentation requires us to reset auto increment value to 1
+		 * on truncate even if the table was created with a different value. 
+		 * This is also consistent with other engines.
+		 */
+		dic.dic_min_auto_inc = 1;
+
 		xt_create_table(self, (XTPathStrPtr) path, &dic);
 		if (!pb_table_locked)
 			freer_(); // ha_release_exclusive_use(pb_share)
@@ -3737,7 +3845,7 @@ int ha_pbxt::delete_all_rows()
  * now agree with the MyISAM strategy.
  * 
  */
-int ha_pbxt::analyze(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt __attribute__((unused)))
+int ha_pbxt::analyze(THD *thd, HA_CHECK_OPT *XT_UNUSED(check_opt))
 {
 	int				err = 0;
 	XTDatabaseHPtr	db;
@@ -3819,7 +3927,7 @@ int ha_pbxt::analyze(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt _
 	XT_RETURN(err);
 }
 
-int ha_pbxt::repair(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt __attribute__((unused)))
+int ha_pbxt::repair(THD *XT_UNUSED(thd), HA_CHECK_OPT *XT_UNUSED(check_opt))
 {
 	return(HA_ADMIN_TRY_ALTER);
 }
@@ -3828,7 +3936,7 @@ int ha_pbxt::repair(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt __
  * This is mapped to "ALTER TABLE tablename TYPE=PBXT", which rebuilds
  * the table in MySQL.
  */
-int ha_pbxt::optimize(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt __attribute__((unused)))
+int ha_pbxt::optimize(THD *XT_UNUSED(thd), HA_CHECK_OPT *XT_UNUSED(check_opt))
 {
 	return(HA_ADMIN_TRY_ALTER);
 }
@@ -3837,7 +3945,7 @@ int ha_pbxt::optimize(THD *thd __attribute__((unused)), HA_CHECK_OPT *check_opt
 extern int pbxt_mysql_trace_on;
 #endif
 
-int ha_pbxt::check(THD* thd, HA_CHECK_OPT* check_opt __attribute__((unused)))
+int ha_pbxt::check(THD* thd, HA_CHECK_OPT* XT_UNUSED(check_opt))
 {
 	int				err = 0;
 	XTThreadPtr		self;
@@ -3993,8 +4101,10 @@ xtPublic int ha_pbxt::external_lock(THD *thd, int lock_type)
 			 * (or update statement) just saw.
 			 */
 			if (pb_open_tab) {
-				if (pb_open_tab->ot_for_update)
+				if (pb_open_tab->ot_for_update) {
 					self->st_visible_time = self->st_database->db_xn_end_time;
+					pb_open_tab->ot_for_update = 0;
+				}
 
 				if (pb_share->sh_recalc_selectivity) {
 					if ((pb_share->sh_table->tab_row_eof_id - 1 - pb_share->sh_table->tab_row_fnum) >= 200) {
@@ -4079,10 +4189,15 @@ xtPublic int ha_pbxt::external_lock(THD *thd, int lock_type)
 			pb_open_tab->ot_is_modify = FALSE;
 			if ((pb_open_tab->ot_for_update = (lock_type == F_WRLCK))) {
 				switch ((int) thd_sql_command(thd)) {
-					case SQLCOM_UPDATE:
-					case SQLCOM_UPDATE_MULTI:
 					case SQLCOM_DELETE:
 					case SQLCOM_DELETE_MULTI:
+						/* turn DELETE IGNORE into normal DELETE. The IGNORE option causes problems because 
+						 * when a record is deleted we add an xlog record which we cannot "rollback" later
+						 * when we find that an FK-constraint has failed. 
+						 */
+						thd->lex->ignore = false;
+					case SQLCOM_UPDATE:
+					case SQLCOM_UPDATE_MULTI:
 					case SQLCOM_REPLACE:
 					case SQLCOM_REPLACE_SELECT:
 					case SQLCOM_INSERT:
@@ -4290,7 +4405,9 @@ int ha_pbxt::start_stmt(THD *thd, thr_lock_type lock_type)
 	pb_open_tab->ot_for_update =
 		(lock_type != TL_READ && 
 		 lock_type != TL_READ_WITH_SHARED_LOCKS &&
+#ifndef DRIZZLED
 		 lock_type != TL_READ_HIGH_PRIORITY && 
+#endif
 		 lock_type != TL_READ_NO_INSERT);
 	pb_open_tab->ot_is_modify = FALSE;
 	if (pb_open_tab->ot_for_update) {
@@ -4557,9 +4674,12 @@ int ha_pbxt::delete_table(const char *table_path)
 {
 	THD				*thd = current_thd;
 	int				err = 0;
-	XTThreadPtr		self;
+	XTThreadPtr		self = NULL;
 	XTSharePtr		share;
 
+	STAT_TRACE(self, *thd_query(thd));
+	XT_PRINT1(self, "ha_pbxt::delete_table %s\n", table_path);
+
 	if (XTSystemTableShare::isSystemTable(table_path))
 		return delete_system_table(table_path);
 
@@ -4568,9 +4688,6 @@ int ha_pbxt::delete_table(const char *table_path)
 
 	self->st_ignore_fkeys = (thd_test_options(thd, OPTION_NO_FOREIGN_KEY_CHECKS)) != 0;
 
-	STAT_TRACE(self, *thd_query(thd));
-	XT_PRINT1(self, "ha_pbxt::delete_table %s\n", table_path);
-
 	try_(a) {
 		xt_ha_open_database_of_table(self, (XTPathStrPtr) table_path);
 
@@ -4586,16 +4703,23 @@ int ha_pbxt::delete_table(const char *table_path)
 			pushr_(ha_release_exclusive_use, share);
 			ha_close_open_tables(self, share, NULL);
 
-			xt_drop_table(self, (XTPathStrPtr) table_path);
+			xt_drop_table(self, (XTPathStrPtr) table_path, thd_sql_command(thd) == SQLCOM_DROP_DB);
 
 			freer_(); // ha_release_exclusive_use(share)
 			freer_(); // ha_unget_share(share)
 		}
 		catch_(b) {
-			/* If the table does not exist, just log the error and continue... */
+			/* In MySQL if the table does not exist, just log the error and continue. This is
+ 			 * needed to delete table in the case when CREATE TABLE fails and no PBXT disk
+ 			 * structures were created. 
+ 			 * Drizzle unlike MySQL iterates over all handlers and tries to delete table. It
+ 			 * stops after when a handler returns TRUE, so in Drizzle we need to report error.  
+			 */
+#ifndef DRIZZLED
 			if (self->t_exception.e_xt_err == XT_ERR_TABLE_NOT_FOUND)
 				xt_log_and_clear_exception(self);
 			else
+#endif
 				throw_();
 		}
 		cont_(b);
@@ -4619,6 +4743,10 @@ int ha_pbxt::delete_table(const char *table_path)
 	}
 	catch_(a) {
 		err = xt_ha_pbxt_thread_error_for_mysql(thd, self, pb_ignore_dup_key);
+#ifdef DRIZZLED
+		if (err == HA_ERR_NO_SUCH_TABLE)
+			err = ENOENT;
+#endif
 	}
 	cont_(a);
 
@@ -4736,7 +4864,7 @@ int ha_pbxt::rename_table(const char *from, const char *to)
 	XT_RETURN(err);
 }
 
-int ha_pbxt::rename_system_table(const char *from __attribute__((unused)), const char *to __attribute__((unused)))
+int ha_pbxt::rename_system_table(const char *XT_UNUSED(from), const char *XT_UNUSED(to))
 {
 	return ER_NOT_SUPPORTED_YET;
 }
@@ -4771,7 +4899,7 @@ double ha_pbxt::scan_time()
 /*
  * The next method will never be called if you do not implement indexes.
  */
-double ha_pbxt::read_time(uint index __attribute__((unused)), uint ranges, ha_rows rows)
+double ha_pbxt::read_time(uint XT_UNUSED(index), uint ranges, ha_rows rows)
 {
 	double result = rows2double(ranges+rows);
 	return result;
@@ -4945,7 +5073,7 @@ void ha_pbxt::free_foreign_key_create_info(char* str)
 	xt_free(NULL, str);
 }
 
-bool ha_pbxt::get_error_message(int error __attribute__((unused)), String *buf)
+bool ha_pbxt::get_error_message(int XT_UNUSED(error), String *buf)
 {
 	THD				*thd = current_thd;
 	int				err = 0;
@@ -5104,9 +5232,9 @@ struct st_mysql_sys_var
 #endif
 
 #ifdef USE_CONST_SAVE
-static void pbxt_record_cache_size_func(THD *thd __attribute__((unused)), struct st_mysql_sys_var *var, void *tgt, const void *save)
+static void pbxt_record_cache_size_func(THD *XT_UNUSED(thd), struct st_mysql_sys_var *var, void *tgt, const void *save)
 #else
-static void pbxt_record_cache_size_func(THD *thd __attribute__((unused)), struct st_mysql_sys_var *var, void *tgt, void *save)
+static void pbxt_record_cache_size_func(THD *XT_UNUSED(thd), struct st_mysql_sys_var *var, void *tgt, void *save)
 #endif
 {
 	xtInt8	record_cache_size;
@@ -5215,6 +5343,18 @@ static MYSQL_SYSVAR_INT(sweeper_priority, xt_db_sweeper_priority,
 	"Determines the priority of the background sweeper process, 0 = low (default), 1 = normal (same as user threads), 2 = high.",
 	NULL, NULL, XT_PRIORITY_LOW, XT_PRIORITY_LOW, XT_PRIORITY_HIGH, 1);
 
+#ifdef DRIZZLED
+static MYSQL_SYSVAR_INT(max_threads, pbxt_max_threads,
+	PLUGIN_VAR_OPCMDARG,
+	"The maximum number of threads used by PBXT",
+	NULL, NULL, 500, 20, 20000, 1);
+#else
+static MYSQL_SYSVAR_INT(max_threads, pbxt_max_threads,
+	PLUGIN_VAR_OPCMDARG,
+	"The maximum number of threads used by PBXT, 0 = set according to MySQL max_connections.",
+	NULL, NULL, 0, 0, 20000, 1);
+#endif
+
 static struct st_mysql_sys_var* pbxt_system_variables[] = {
   MYSQL_SYSVAR(index_cache_size),
   MYSQL_SYSVAR(record_cache_size),
@@ -5231,6 +5371,7 @@ static struct st_mysql_sys_var* pbxt_system_variables[] = {
   MYSQL_SYSVAR(auto_increment_mode),
   MYSQL_SYSVAR(offline_log_function),
   MYSQL_SYSVAR(sweeper_priority),
+  MYSQL_SYSVAR(max_threads),
   NULL
 };
 #endif
@@ -5241,8 +5382,8 @@ drizzle_declare_plugin(pbxt)
 mysql_declare_plugin(pbxt)
 #endif
 {
-	MYSQL_STORAGE_ENGINE_PLUGIN,
 #ifndef DRIZZLED
+	MYSQL_STORAGE_ENGINE_PLUGIN,
 	&pbxt_storage_engine,
 #endif
 	"PBXT",
@@ -5266,8 +5407,8 @@ mysql_declare_plugin(pbxt)
 	NULL						/* config options                  */
 },
 {
-	MYSQL_INFORMATION_SCHEMA_PLUGIN,
 #ifndef DRIZZLED
+	MYSQL_INFORMATION_SCHEMA_PLUGIN,
 	&pbxt_statitics,
 #endif
 	"PBXT_STATISTICS",
diff --git a/storage/pbxt/src/ha_pbxt.h b/storage/pbxt/src/ha_pbxt.h
index 6f6a194de12..d48bcd34147 100644
--- a/storage/pbxt/src/ha_pbxt.h
+++ b/storage/pbxt/src/ha_pbxt.h
@@ -28,7 +28,7 @@
 #ifdef DRIZZLED
 #include <drizzled/common.h>
 #include <drizzled/handler.h>
-#include <drizzled/handlerton.h>
+#include <drizzled/plugin/storage_engine.h>
 #include <mysys/thr_lock.h>
 #else
 #include "mysql_priv.h"
@@ -51,6 +51,25 @@
 
 class ha_pbxt;
 
+#ifdef DRIZZLED
+
+class PBXTStorageEngine : public StorageEngine {
+public:
+	PBXTStorageEngine(std::string name_arg)
+	: StorageEngine(name_arg, HTON_NO_FLAGS) {}
+
+	/* override */ int close_connection(Session *);
+	/* override */ int commit(Session *, bool);
+	/* override */ int rollback(Session *, bool);
+	/* override */ handler *create(TABLE_SHARE *, MEM_ROOT *);
+	/* override */ void drop_database(char *);
+	/* override */ bool show_status(Session *, stat_print_fn *, enum ha_stat_type);
+};
+
+typedef PBXTStorageEngine handlerton;
+
+#endif
+
 extern handlerton *pbxt_hton;
 
 /*
diff --git a/storage/pbxt/src/ha_xtsys.cc b/storage/pbxt/src/ha_xtsys.cc
index 1c76d13379a..c76f60267be 100644
--- a/storage/pbxt/src/ha_xtsys.cc
+++ b/storage/pbxt/src/ha_xtsys.cc
@@ -75,7 +75,7 @@ const char **ha_xtsys::bas_ext() const
 	return ha_pbms_exts;
 }
 
-int ha_xtsys::open(const char *table_path, int mode __attribute__((unused)), uint test_if_locked __attribute__((unused)))
+int ha_xtsys::open(const char *table_path, int XT_UNUSED(mode), uint XT_UNUSED(test_if_locked))
 {
 	THD				*thd = current_thd;
 	XTExceptionRec	e;
@@ -141,7 +141,7 @@ int ha_xtsys::close(void)
 	return err;
 }
 
-int ha_xtsys::rnd_init(bool scan __attribute__((unused)))
+int ha_xtsys::rnd_init(bool XT_UNUSED(scan))
 {
 	int err = 0;
 
@@ -185,7 +185,7 @@ int ha_xtsys::rnd_pos(byte * buf, byte *pos)
 	return err;
 }
 
-int ha_xtsys::info(uint flag __attribute__((unused)))
+int ha_xtsys::info(uint XT_UNUSED(flag))
 {
 	return 0;
 }
@@ -211,7 +211,7 @@ int ha_xtsys::external_lock(THD *thd, int lock_type)
 	return err;
 }
 
-THR_LOCK_DATA **ha_xtsys::store_lock(THD *thd __attribute__((unused)), THR_LOCK_DATA **to, enum thr_lock_type lock_type)
+THR_LOCK_DATA **ha_xtsys::store_lock(THD *XT_UNUSED(thd), THR_LOCK_DATA **to, enum thr_lock_type lock_type)
 {
 	if (lock_type != TL_IGNORE && ha_lock.type == TL_UNLOCK)
 		ha_lock.type = lock_type;
@@ -220,13 +220,13 @@ THR_LOCK_DATA **ha_xtsys::store_lock(THD *thd __attribute__((unused)), THR_LOCK_
 }
 
 /* Note: ha_pbxt::delete_system_table is called instead. */
-int ha_xtsys::delete_table(const char *table_path __attribute__((unused)))
+int ha_xtsys::delete_table(const char *XT_UNUSED(table_path))
 {
 	/* Should never be called */
 	return 0;
 }
 
-int ha_xtsys::create(const char *name __attribute__((unused)), TABLE *table_arg __attribute__((unused)), HA_CREATE_INFO *create_info __attribute__((unused)))
+int ha_xtsys::create(const char *XT_UNUSED(name), TABLE *XT_UNUSED(table_arg), HA_CREATE_INFO *XT_UNUSED(create_info))
 {
 	/* Allow the table to be created.
 	 * This is required after a dump is restored.
@@ -234,7 +234,7 @@ int ha_xtsys::create(const char *name __attribute__((unused)), TABLE *table_arg
 	return 0;
 }
 
-bool ha_xtsys::get_error_message(int error __attribute__((unused)), String *buf)
+bool ha_xtsys::get_error_message(int XT_UNUSED(error), String *buf)
 {
 	THD				*thd = current_thd;
 	XTExceptionRec	e;
diff --git a/storage/pbxt/src/ha_xtsys.h b/storage/pbxt/src/ha_xtsys.h
index 66a4b5a5dfa..598abe0938f 100644
--- a/storage/pbxt/src/ha_xtsys.h
+++ b/storage/pbxt/src/ha_xtsys.h
@@ -59,7 +59,7 @@ public:
 
 	const char *table_type() const { return "PBXT"; }
 
-	const char *index_type(uint inx __attribute__((unused))) {
+	const char *index_type(uint XT_UNUSED(inx)) {
 		return "NONE";
 	}
 
@@ -69,7 +69,7 @@ public:
 		return HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE;
 	}
 
-	MX_ULONG_T index_flags(uint inx __attribute__((unused)), uint part __attribute__((unused)), bool all_parts __attribute__((unused))) const {
+	MX_ULONG_T index_flags(uint XT_UNUSED(inx), uint XT_UNUSED(part), bool XT_UNUSED(all_parts)) const {
 		return (HA_READ_NEXT | HA_READ_PREV | HA_READ_RANGE | HA_KEYREAD_ONLY);
 	}
 	uint	max_supported_keys()			const { return 512; }
diff --git a/storage/pbxt/src/hashtab_xt.cc b/storage/pbxt/src/hashtab_xt.cc
index 3708f071ac5..80ba86a5248 100644
--- a/storage/pbxt/src/hashtab_xt.cc
+++ b/storage/pbxt/src/hashtab_xt.cc
@@ -115,7 +115,7 @@ xtPublic void xt_ht_put(XTThreadPtr self, XTHashTabPtr ht, void *data)
 	popr_();
 }
 
-xtPublic void *xt_ht_get(XTThreadPtr self __attribute__((unused)), XTHashTabPtr ht, void *key)
+xtPublic void *xt_ht_get(XTThreadPtr XT_UNUSED(self), XTHashTabPtr ht, void *key)
 {
 	XTHashItemPtr	item;
 	xtHashValue		h;
@@ -239,14 +239,14 @@ xtPublic void xt_ht_signal(XTThreadPtr self, XTHashTabPtr ht)
 	xt_signal_cond(self, ht->ht_cond);
 }
 
-xtPublic void xt_ht_enum(struct XTThread *self __attribute__((unused)), XTHashTabPtr ht, XTHashEnumPtr en)
+xtPublic void xt_ht_enum(struct XTThread *XT_UNUSED(self), XTHashTabPtr ht, XTHashEnumPtr en)
 {
 	en->he_i = 0;
 	en->he_item = NULL;
 	en->he_ht = ht;
 }
 
-xtPublic void *xt_ht_next(struct XTThread *self __attribute__((unused)), XTHashEnumPtr en)
+xtPublic void *xt_ht_next(struct XTThread *XT_UNUSED(self), XTHashEnumPtr en)
 {
 	if (en->he_item) {
 		en->he_item = en->he_item->hi_next;
diff --git a/storage/pbxt/src/heap_xt.cc b/storage/pbxt/src/heap_xt.cc
index a88df833a4a..dcfa1dae11f 100644
--- a/storage/pbxt/src/heap_xt.cc
+++ b/storage/pbxt/src/heap_xt.cc
@@ -31,7 +31,7 @@
 #undef xt_heap_new
 #endif
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 xtPublic XTHeapPtr xt_mm_heap_new(XTThreadPtr self, size_t size, XTFinalizeFunc finalize, u_int line, c_char *file, xtBool track)
 #else
 xtPublic XTHeapPtr xt_heap_new(XTThreadPtr self, size_t size, XTFinalizeFunc finalize)
@@ -39,7 +39,7 @@ xtPublic XTHeapPtr xt_heap_new(XTThreadPtr self, size_t size, XTFinalizeFunc fin
 {
 	volatile XTHeapPtr	hp;
 	
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	hp = (XTHeapPtr) xt_mm_calloc(self, size, line, file);
 	hp->h_track = track;
 	if (track)
@@ -65,21 +65,21 @@ xtPublic XTHeapPtr xt_heap_new(XTThreadPtr self, size_t size, XTFinalizeFunc fin
 	return hp;
 }
 
-xtPublic void xt_check_heap(XTThreadPtr self __attribute__((unused)), XTHeapPtr hp __attribute__((unused)))
+xtPublic void xt_check_heap(XTThreadPtr XT_NDEBUG_UNUSED(self), XTHeapPtr XT_NDEBUG_UNUSED(hp))
 {
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	xt_mm_malloc_size(self, hp);
 #endif
 }
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 xtPublic void xt_mm_heap_reference(XTThreadPtr self, XTHeapPtr hp, u_int line, c_char *file)
 #else
 xtPublic void xt_heap_reference(XTThreadPtr, XTHeapPtr hp)
 #endif
 {
 	xt_spinlock_lock(&hp->h_lock);
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	if (hp->h_track)
 		printf("HEAP: +1 %d->%d %s:%d\n", (int) hp->h_ref_count, (int) hp->h_ref_count+1, file, (int) line);
 #endif
@@ -91,7 +91,7 @@ xtPublic void xt_heap_release(XTThreadPtr self, XTHeapPtr hp)
 {	
 	if (!hp)
 		return;
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	xt_spinlock_lock(&hp->h_lock);
 	ASSERT(hp->h_ref_count != 0);
 	xt_spinlock_unlock(&hp->h_lock);
@@ -100,7 +100,7 @@ xtPublic void xt_heap_release(XTThreadPtr self, XTHeapPtr hp)
 	if (hp->h_onrelease)
 		(*hp->h_onrelease)(self, hp);
 	if (hp->h_ref_count > 0) {
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	if (hp->h_track)
 		printf("HEAP: -1 %d->%d\n", (int) hp->h_ref_count, (int) hp->h_ref_count-1);
 #endif
@@ -116,12 +116,12 @@ xtPublic void xt_heap_release(XTThreadPtr self, XTHeapPtr hp)
 	xt_spinlock_unlock(&hp->h_lock);
 }
 
-xtPublic void xt_heap_set_release_callback(XTThreadPtr self __attribute__((unused)), XTHeapPtr hp, XTFinalizeFunc onrelease)
+xtPublic void xt_heap_set_release_callback(XTThreadPtr XT_UNUSED(self), XTHeapPtr hp, XTFinalizeFunc onrelease)
 {
 	hp->h_onrelease = onrelease;
 }
 
-xtPublic u_int xt_heap_get_ref_count(struct XTThread *self __attribute__((unused)), XTHeapPtr hp)
+xtPublic u_int xt_heap_get_ref_count(struct XTThread *XT_UNUSED(self), XTHeapPtr hp)
 {
 	return hp->h_ref_count;
 }
diff --git a/storage/pbxt/src/heap_xt.h b/storage/pbxt/src/heap_xt.h
index afad132e1e3..db7a6909f05 100644
--- a/storage/pbxt/src/heap_xt.h
+++ b/storage/pbxt/src/heap_xt.h
@@ -25,6 +25,7 @@
 
 #include "xt_defs.h"
 #include "lock_xt.h"
+#include "memory_xt.h"
 
 struct XTThread;
 
@@ -59,7 +60,7 @@ u_int		xt_heap_get_ref_count(struct XTThread *self, XTHeapPtr mem);
 
 void		xt_check_heap(struct XTThread *self, XTHeapPtr mem);
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 #define xt_heap_new(t, s, f)		xt_mm_heap_new(t, s, f, __LINE__, __FILE__, FALSE)
 #define xt_heap_new_track(t, s, f)	xt_mm_heap_new(t, s, f, __LINE__, __FILE__, TRUE)
 #define xt_heap_reference(t, s)		xt_mm_heap_reference(t, s, __LINE__, __FILE__)
diff --git a/storage/pbxt/src/index_xt.cc b/storage/pbxt/src/index_xt.cc
index 9cd9a966f74..cfd34c1afea 100644
--- a/storage/pbxt/src/index_xt.cc
+++ b/storage/pbxt/src/index_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <string.h>
 #include <stdio.h>
 #include <stddef.h>
@@ -52,7 +56,7 @@
 //#define CHECK_AND_PRINT
 //#define CHECK_NODE_REFERENCE
 //#define TRACE_FLUSH
-//#define CHECK_PRINTS_RECORD_REFERENCES
+#define CHECK_PRINTS_RECORD_REFERENCES
 #else
 #define MAX_SEARCH_DEPTH			100
 #endif
@@ -77,6 +81,7 @@ static u_int idx_check_index(XTOpenTablePtr ot, XTIndexPtr ind, xtBool with_lock
 #endif
 
 static xtBool idx_insert_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackPtr stack, XTIdxKeyValuePtr key_value, xtIndexNodeID branch);
+static xtBool idx_remove_lazy_deleted_item_in_node(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID current, XTIndReferencePtr iref, XTIdxKeyValuePtr key_value);
 
 #ifdef XT_TRACK_INDEX_UPDATES
 
@@ -163,7 +168,7 @@ static void track_dump_all(u_int max_block)
 
 #endif
 
-xtPublic void xt_ind_track_dump_block(XTTableHPtr tab __attribute__((unused)), xtIndexNodeID address __attribute__((unused)))
+xtPublic void xt_ind_track_dump_block(XTTableHPtr XT_UNUSED(tab), xtIndexNodeID XT_UNUSED(address))
 {
 #ifdef TRACK_ACTIVITY
 	u_int i = XT_NODE_ID(address)-1;
@@ -268,7 +273,7 @@ static xtBool idx_new_branch(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID *a
 
 	if ((XT_NODE_ID(wrote_pos) = XT_NODE_ID(tab->tab_ind_free))) {
 		/* Use the block on the free list: */
-		if (!xt_ind_read_bytes(ot, wrote_pos, sizeof(XTIndFreeBlockRec), (xtWord1 *) &free_block))
+		if (!xt_ind_read_bytes(ot, ind, wrote_pos, sizeof(XTIndFreeBlockRec), (xtWord1 *) &free_block))
 			goto failed;
 		XT_NODE_ID(tab->tab_ind_free) = (xtIndexNodeID) XT_GET_DISK_8(free_block.if_next_block_8);
 		xt_unlock_mutex_ns(&tab->tab_ind_lock);
@@ -343,7 +348,7 @@ static xtBool idx_free_branch(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID n
  * Simple compare functions
  */
 
-xtPublic int xt_compare_2_int4(XTIndexPtr ind __attribute__((unused)), uint key_length, xtWord1 *key_value, xtWord1 *b_value)
+xtPublic int xt_compare_2_int4(XTIndexPtr XT_UNUSED(ind), uint key_length, xtWord1 *key_value, xtWord1 *b_value)
 {
 	int r;
 
@@ -357,7 +362,7 @@ xtPublic int xt_compare_2_int4(XTIndexPtr ind __attribute__((unused)), uint key_
 	return r;
 }
 
-xtPublic int xt_compare_3_int4(XTIndexPtr ind __attribute__((unused)), uint key_length, xtWord1 *key_value, xtWord1 *b_value)
+xtPublic int xt_compare_3_int4(XTIndexPtr XT_UNUSED(ind), uint key_length, xtWord1 *key_value, xtWord1 *b_value)
 {
 	int r;
 
@@ -381,7 +386,7 @@ xtPublic int xt_compare_3_int4(XTIndexPtr ind __attribute__((unused)), uint key_
  * Tree branch sanning (searching nodes and leaves)
  */
 
-xtPublic void xt_scan_branch_single(struct XTTable *tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
+xtPublic void xt_scan_branch_single(struct XTTable *XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	u_int				branch_size;
@@ -522,7 +527,7 @@ xtPublic void xt_scan_branch_single(struct XTTable *tab __attribute__((unused)),
  * index (in the case of -1) or to the first value after the
  * the search key in the case of 1.
  */
-xtPublic void xt_scan_branch_fix(struct XTTable *tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
+xtPublic void xt_scan_branch_fix(struct XTTable *XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	u_int				branch_size;
@@ -619,7 +624,7 @@ xtPublic void xt_scan_branch_fix(struct XTTable *tab __attribute__((unused)), XT
 	result->sr_item.i_item_offset = node_ref_size + i * full_item_size;
 }
 
-xtPublic void xt_scan_branch_fix_simple(struct XTTable *tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
+xtPublic void xt_scan_branch_fix_simple(struct XTTable *XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	u_int				branch_size;
@@ -720,7 +725,7 @@ xtPublic void xt_scan_branch_fix_simple(struct XTTable *tab __attribute__((unuse
  * Variable length key values are stored as a sorted list. Since each list item has a variable length, we
  * must scan the list sequentially in order to find a key.
  */
-xtPublic void xt_scan_branch_var(struct XTTable *tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
+xtPublic void xt_scan_branch_var(struct XTTable *XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxKeyValuePtr value, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	u_int			branch_size;
@@ -816,7 +821,7 @@ xtPublic void xt_scan_branch_var(struct XTTable *tab __attribute__((unused)), XT
 }
 
 /* Go to the next item in the node. */
-static void idx_next_branch_item(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultRec *result)
+static void idx_next_branch_item(XTTableHPtr XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	xtWord1	*bitem;
@@ -834,7 +839,7 @@ static void idx_next_branch_item(XTTableHPtr tab __attribute__((unused)), XTInde
 	result->sr_branch = IDX_GET_NODE_REF(tab, bitem, result->sr_item.i_node_ref_size);
 }
 
-xtPublic void xt_prev_branch_item_fix(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind __attribute__((unused)), XTIdxBranchDPtr branch, register XTIdxResultRec *result)
+xtPublic void xt_prev_branch_item_fix(XTTableHPtr XT_UNUSED(tab), XTIndexPtr XT_UNUSED(ind), XTIdxBranchDPtr branch, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	ASSERT_NS(result->sr_item.i_item_offset >= result->sr_item.i_item_size + result->sr_item.i_node_ref_size + result->sr_item.i_node_ref_size);
@@ -843,7 +848,7 @@ xtPublic void xt_prev_branch_item_fix(XTTableHPtr tab __attribute__((unused)), X
 	result->sr_branch = IDX_GET_NODE_REF(tab, branch->tb_data + result->sr_item.i_item_offset, result->sr_item.i_node_ref_size);
 }
 
-xtPublic void xt_prev_branch_item_var(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultRec *result)
+xtPublic void xt_prev_branch_item_var(XTTableHPtr XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultRec *result)
 {
 	XT_NODE_TEMP;
 	xtWord1	*bitem;
@@ -865,7 +870,20 @@ xtPublic void xt_prev_branch_item_var(XTTableHPtr tab __attribute__((unused)), X
 	result->sr_item.i_item_offset = bitem - branch->tb_data;
 }
 
-static void idx_first_branch_item(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
+static void idx_reload_item_fix(XTIndexPtr XT_NDEBUG_UNUSED(ind), XTIdxBranchDPtr branch, register XTIdxResultPtr result)
+{
+	u_int branch_size;
+
+	branch_size = XT_GET_DISK_2(branch->tb_size_2);
+	ASSERT_NS(result->sr_item.i_node_ref_size == (XT_IS_NODE(branch_size) ? XT_NODE_REF_SIZE : 0));
+	ASSERT_NS(result->sr_item.i_item_size == ind->mi_key_size + XT_RECORD_REF_SIZE);
+	result->sr_item.i_total_size = XT_GET_BRANCH_DATA_SIZE(branch_size);
+	if (result->sr_item.i_item_offset > result->sr_item.i_total_size)
+		result->sr_item.i_item_offset = result->sr_item.i_total_size;
+	xt_get_res_record_ref(&branch->tb_data[result->sr_item.i_item_offset + result->sr_item.i_item_size - XT_RECORD_REF_SIZE], result); 
+}
+
+static void idx_first_branch_item(XTTableHPtr XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
 {
 	XT_NODE_TEMP;
 	u_int branch_size;
@@ -903,7 +921,7 @@ static void idx_first_branch_item(XTTableHPtr tab __attribute__((unused)), XTInd
 /*
  * Last means different things for leaf or node!
  */
-xtPublic void xt_last_branch_item_fix(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
+xtPublic void xt_last_branch_item_fix(XTTableHPtr XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
 {
 	XT_NODE_TEMP;
 	u_int branch_size;
@@ -935,7 +953,7 @@ xtPublic void xt_last_branch_item_fix(XTTableHPtr tab __attribute__((unused)), X
 	}
 }
 
-xtPublic void xt_last_branch_item_var(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
+xtPublic void xt_last_branch_item_var(XTTableHPtr XT_UNUSED(tab), XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result)
 {
 	XT_NODE_TEMP;
 	u_int	branch_size;
@@ -986,6 +1004,218 @@ xtPublic void xt_last_branch_item_var(XTTableHPtr tab __attribute__((unused)), X
 	}
 }
 
+xtPublic xtBool xt_idx_lazy_delete_on_leaf(XTIndexPtr ind, XTIndBlockPtr block, xtWord2 branch_size)
+{
+	ASSERT_NS(ind->mi_fix_key);
+	
+	/* Compact the leaf if more than half the items that fit on the page
+	 * are deleted: */
+	if (block->cp_del_count >= ind->mi_max_items/2)
+		return FALSE;
+
+	/* Compact the page if there is only 1 (or less) valid item left: */
+	if ((u_int) block->cp_del_count+1 >= ((u_int) branch_size - 2)/(ind->mi_key_size + XT_RECORD_REF_SIZE))
+		return FALSE;
+
+	return OK;
+}
+
+static xtBool idx_lazy_delete_on_node(XTIndexPtr ind, XTIndBlockPtr block, register XTIdxItemPtr item)
+{
+	ASSERT_NS(ind->mi_fix_key);
+	
+	/* Compact the node if more than 1/4 of the items that fit on the page
+	 * are deleted: */
+	if (block->cp_del_count >= ind->mi_max_items/4)
+		return FALSE;
+
+	/* Compact the page if there is only 1 (or less) valid item left: */
+	if ((u_int) block->cp_del_count+1 >= (item->i_total_size - item->i_node_ref_size)/(item->i_item_size + item->i_node_ref_size))
+		return FALSE;
+
+	return OK;
+}
+
+inline static xtBool idx_cmp_item_key_fix(XTIndReferencePtr iref, register XTIdxItemPtr item, XTIdxKeyValuePtr value)
+{
+	xtWord1 *data;
+
+	data = &iref->ir_branch->tb_data[item->i_item_offset];
+	return memcmp(data, value->sv_key, value->sv_length) == 0;
+}
+
+inline static void idx_set_item_key_fix(XTIndReferencePtr iref, register XTIdxItemPtr item, XTIdxKeyValuePtr value)
+{
+	xtWord1 *data;
+
+	data = &iref->ir_branch->tb_data[item->i_item_offset];
+	memcpy(data, value->sv_key, value->sv_length);
+	xt_set_val_record_ref(data + value->sv_length, value);
+	iref->ir_updated = TRUE;
+}
+
+inline static void idx_set_item_reference(XTIndReferencePtr iref, register XTIdxItemPtr item, xtRowID rec_id, xtRowID row_id)
+{
+	size_t	offset;
+	xtWord1	*data;
+
+	/* This is the offset of the reference in the item we found: */
+	offset = item->i_item_offset +item->i_item_size - XT_RECORD_REF_SIZE;
+	data = &iref->ir_branch->tb_data[offset];
+
+	xt_set_record_ref(data, rec_id, row_id);
+	iref->ir_updated = TRUE;
+}
+
+inline static void idx_set_item_row_id(XTIndReferencePtr iref, register XTIdxItemPtr item, xtRowID row_id)
+{
+	size_t	offset;
+	xtWord1	*data;
+
+	offset = 
+		/* This is the offset of the reference in the item we found: */
+		item->i_item_offset +item->i_item_size - XT_RECORD_REF_SIZE +
+		/* This is the offset of the row id in the reference: */
+		XT_RECORD_ID_SIZE;
+	data = &iref->ir_branch->tb_data[offset];
+
+	/* This update does not change the structure of page, so we do it without
+	 * copying the page before we write.
+	 */
+	XT_SET_DISK_4(data, row_id);
+	iref->ir_updated = TRUE;
+}
+
+inline static xtBool idx_is_item_deleted(register XTIdxBranchDPtr branch, register XTIdxItemPtr item)
+{
+	xtWord1	*data;
+
+	data = &branch->tb_data[item->i_item_offset + item->i_item_size - XT_RECORD_REF_SIZE + XT_RECORD_ID_SIZE];
+	return XT_GET_DISK_4(data) == (xtRowID) -1;
+}
+
+inline static void idx_set_item_deleted(XTIndReferencePtr iref, register XTIdxItemPtr item)
+{
+	idx_set_item_row_id(iref, item, (xtRowID) -1);
+	
+	/* This should be safe because there is only one thread,
+	 * the sweeper, that does this!
+	 *
+	 * Threads that decrement this value have an xlock on
+	 * the page, or the index.
+	 */
+	iref->ir_block->cp_del_count++;
+}
+
+/*
+ * {LAZY-DEL-INDEX-ITEMS}
+ * Do a lazy delete of an item by just setting the Row ID
+ * to the delete indicator: row ID -1.
+ */
+static void idx_lazy_delete_branch_item(XTOpenTablePtr ot, XTIndexPtr ind, XTIndReferencePtr iref, register XTIdxItemPtr item)
+{
+	idx_set_item_deleted(iref, item);
+	xt_ind_release(ot, ind, iref->ir_xlock ? XT_UNLOCK_W_UPDATE : XT_UNLOCK_R_UPDATE, iref);
+}
+
+/*
+ * This function compacts the leaf, but preserves the
+ * position of the item.
+ */
+static xtBool idx_compact_leaf(XTOpenTablePtr ot, XTIndexPtr ind, XTIndReferencePtr iref, register XTIdxItemPtr item)
+{
+	register XTIdxBranchDPtr branch = iref->ir_branch;
+	int		item_idx, count, i, idx;
+	u_int	size;
+	xtWord1	*s_data;
+	xtWord1	*d_data;
+	xtWord1	*data;
+	xtRowID	row_id;
+
+	if (iref->ir_block->cb_handle_count) {
+		if (!xt_ind_copy_on_write(iref)) {
+			xt_ind_release(ot, ind, iref->ir_xlock ? XT_UNLOCK_WRITE : XT_UNLOCK_READ, iref);
+			return FAILED;
+		}
+	}
+
+	ASSERT_NS(!item->i_node_ref_size);
+	ASSERT_NS(ind->mi_fix_key);
+	size = item->i_item_size;
+	count = item->i_total_size / size;
+	item_idx = item->i_item_offset / size;
+	s_data = d_data = branch->tb_data;
+	idx = 0;
+	for (i=0; i<count; i++) {
+		data = s_data + item->i_item_size - XT_RECORD_REF_SIZE + XT_RECORD_ID_SIZE;
+		row_id = XT_GET_DISK_4(data);
+		if (row_id == (xtRowID) -1) {
+			if (idx < item_idx)
+				item_idx--;
+		}
+		else {
+			if (d_data != s_data)
+				memcpy(d_data, s_data, size);
+			d_data += size;
+			idx++;
+		}
+		s_data += size;
+	}
+	iref->ir_block->cp_del_count = 0;
+	item->i_total_size = d_data - branch->tb_data;
+	ASSERT_NS(idx * size == item->i_total_size);
+	item->i_item_offset = item_idx * size;
+	XT_SET_DISK_2(branch->tb_size_2, XT_MAKE_BRANCH_SIZE(item->i_total_size, 0));
+	iref->ir_updated = TRUE;
+	return OK;
+}
+
+static xtBool idx_lazy_remove_leaf_item_right(XTOpenTablePtr ot, XTIndexPtr ind, XTIndReferencePtr iref, register XTIdxItemPtr item)
+{
+	register XTIdxBranchDPtr branch = iref->ir_branch;
+	int		item_idx, count, i;
+	u_int	size;
+	xtWord1	*s_data;
+	xtWord1	*d_data;
+	xtWord1	*data;
+	xtRowID	row_id;
+
+	ASSERT_NS(!item->i_node_ref_size);
+
+	if (iref->ir_block->cb_handle_count) {
+		if (!xt_ind_copy_on_write(iref)) {
+			xt_ind_release(ot, ind, XT_UNLOCK_WRITE, iref);
+			return FAILED;
+		}
+	}
+
+	ASSERT_NS(ind->mi_fix_key);
+	size = item->i_item_size;
+	count = item->i_total_size / size;
+	item_idx = item->i_item_offset / size;
+	s_data = d_data = branch->tb_data;
+	for (i=0; i<count; i++) {
+		if (i == item_idx)
+			item->i_item_offset = d_data - branch->tb_data;
+		else {
+			data = s_data + item->i_item_size - XT_RECORD_REF_SIZE + XT_RECORD_ID_SIZE;
+			row_id = XT_GET_DISK_4(data);
+			if (row_id != (xtRowID) -1) {
+				if (d_data != s_data)
+					memcpy(d_data, s_data, size);
+				d_data += size;
+			}
+		}
+		s_data += size;
+	}
+	iref->ir_block->cp_del_count = 0;
+	item->i_total_size = d_data - branch->tb_data;
+	XT_SET_DISK_2(branch->tb_size_2, XT_MAKE_BRANCH_SIZE(item->i_total_size, 0));
+	iref->ir_updated = TRUE;
+	xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, iref);
+	return OK;
+}
+
 /*
  * Remove an item and save to disk.
  */
@@ -1003,8 +1233,14 @@ static xtBool idx_remove_branch_item_right(XTOpenTablePtr ot, XTIndexPtr ind, xt
 	 * an Xlock on the cache block.
 	 */
 	if (iref->ir_block->cb_handle_count) {
-		if (!xt_ind_copy_on_write(iref))
+		if (!xt_ind_copy_on_write(iref)) {
+			xt_ind_release(ot, ind, item->i_node_ref_size ? XT_UNLOCK_READ : XT_UNLOCK_WRITE, iref);
 			return FAILED;
+		}
+	}
+	if (ind->mi_lazy_delete) {
+		if (idx_is_item_deleted(branch, item))
+			iref->ir_block->cp_del_count--;
 	}
 	/* Remove the node reference to the left of the item: */
 	memmove(&branch->tb_data[item->i_item_offset],
@@ -1013,18 +1249,28 @@ static xtBool idx_remove_branch_item_right(XTOpenTablePtr ot, XTIndexPtr ind, xt
 	item->i_total_size -= size;
 	XT_SET_DISK_2(branch->tb_size_2, XT_MAKE_BRANCH_SIZE(item->i_total_size, item->i_node_ref_size));
 	IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(address), (int) XT_GET_DISK_2(branch->tb_size_2));
+	iref->ir_updated = TRUE;
 	xt_ind_release(ot, ind, item->i_node_ref_size ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_W_UPDATE, iref);
 	return OK;
 }
 
-static xtBool idx_remove_branch_item_left(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID, XTIndReferencePtr iref, register XTIdxItemPtr item)
+static xtBool idx_remove_branch_item_left(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID, XTIndReferencePtr iref, register XTIdxItemPtr item, xtBool *lazy_delete_cleanup_required)
 {
 	register XTIdxBranchDPtr branch = iref->ir_branch;
 	u_int size = item->i_item_size + item->i_node_ref_size;
 
+	ASSERT_NS(item->i_node_ref_size);
 	if (iref->ir_block->cb_handle_count) {
-		if (!xt_ind_copy_on_write(iref))
+		if (!xt_ind_copy_on_write(iref)) {
+			xt_ind_release(ot, ind, item->i_node_ref_size ? XT_UNLOCK_READ : XT_UNLOCK_WRITE, iref);
 			return FAILED;
+		}
+	}
+	if (ind->mi_lazy_delete) {
+		if (idx_is_item_deleted(branch, item))
+			iref->ir_block->cp_del_count--;
+		if (lazy_delete_cleanup_required)
+			*lazy_delete_cleanup_required = idx_lazy_delete_on_node(ind, iref->ir_block, item);
 	}
 	/* Remove the node reference to the left of the item: */
 	memmove(&branch->tb_data[item->i_item_offset - item->i_node_ref_size],
@@ -1033,11 +1279,12 @@ static xtBool idx_remove_branch_item_left(XTOpenTablePtr ot, XTIndexPtr ind, xtI
 	item->i_total_size -= size;
 	XT_SET_DISK_2(branch->tb_size_2, XT_MAKE_BRANCH_SIZE(item->i_total_size, item->i_node_ref_size));
 	IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(address), (int) XT_GET_DISK_2(branch->tb_size_2));
+	iref->ir_updated = TRUE;
 	xt_ind_release(ot, ind, item->i_node_ref_size ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_W_UPDATE, iref);
 	return OK;
 }
 
-static void idx_insert_leaf_item(XTIndexPtr ind __attribute__((unused)), XTIdxBranchDPtr leaf, XTIdxKeyValuePtr value, XTIdxResultPtr result)
+static void idx_insert_leaf_item(XTIndexPtr XT_UNUSED(ind), XTIdxBranchDPtr leaf, XTIdxKeyValuePtr value, XTIdxResultPtr result)
 {
 	xtWord1 *item;
 
@@ -1053,7 +1300,7 @@ static void idx_insert_leaf_item(XTIndexPtr ind __attribute__((unused)), XTIdxBr
 	XT_SET_DISK_2(leaf->tb_size_2, XT_MAKE_LEAF_SIZE(result->sr_item.i_total_size));
 }
 
-static void idx_insert_node_item(XTTableHPtr tab __attribute__((unused)), XTIndexPtr ind __attribute__((unused)), XTIdxBranchDPtr leaf, XTIdxKeyValuePtr value, XTIdxResultPtr result, xtIndexNodeID branch)
+static void idx_insert_node_item(XTTableHPtr XT_UNUSED(tab), XTIndexPtr XT_UNUSED(ind), XTIdxBranchDPtr leaf, XTIdxKeyValuePtr value, XTIdxResultPtr result, xtIndexNodeID branch)
 {
 	xtWord1 *item;
 
@@ -1114,7 +1361,7 @@ static void idx_get_middle_branch_item(XTIndexPtr ind, XTIdxBranchDPtr branch, X
 	}
 }
 
-static size_t idx_write_branch_item(XTIndexPtr ind __attribute__((unused)), xtWord1 *item, XTIdxKeyValuePtr value)
+static size_t idx_write_branch_item(XTIndexPtr XT_UNUSED(ind), xtWord1 *item, XTIdxKeyValuePtr value)
 {
 	memcpy(item, value->sv_key, value->sv_length);
 	xt_set_val_record_ref(item + value->sv_length, value);
@@ -1133,23 +1380,38 @@ static xtBool idx_replace_node_key(XTOpenTablePtr ot, XTIndexPtr ind, IdxStackIt
 	xtWord1				key_buf[XT_INDEX_MAX_KEY_SIZE];
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
-	if (!xt_ind_fetch(ot, current, XT_LOCK_WRITE, &iref))
+	if (!xt_ind_fetch(ot, ind, current, XT_LOCK_WRITE, &iref))
 		return FAILED;
 	if (iref.ir_block->cb_handle_count) {
 		if (!xt_ind_copy_on_write(&iref))
 			goto failed_1;
 	}
+	if (ind->mi_lazy_delete) {
+		ASSERT_NS(item_size == item->i_pos.i_item_size);
+		if (idx_is_item_deleted(iref.ir_branch, &item->i_pos))
+			iref.ir_block->cp_del_count--;
+	}
 	memmove(&iref.ir_branch->tb_data[item->i_pos.i_item_offset + item_size],
 		&iref.ir_branch->tb_data[item->i_pos.i_item_offset + item->i_pos.i_item_size],
 		item->i_pos.i_total_size - item->i_pos.i_item_offset - item->i_pos.i_item_size);
 	memcpy(&iref.ir_branch->tb_data[item->i_pos.i_item_offset],
 		item_buf, item_size);
+	if (ind->mi_lazy_delete) {
+		if (idx_is_item_deleted(iref.ir_branch, &item->i_pos))
+			iref.ir_block->cp_del_count++;
+	}
 	item->i_pos.i_total_size = item->i_pos.i_total_size + item_size - item->i_pos.i_item_size;
 	XT_SET_DISK_2(iref.ir_branch->tb_size_2, XT_MAKE_NODE_SIZE(item->i_pos.i_total_size));
 	IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(current), (int) XT_GET_DISK_2(iref.ir_branch->tb_size_2));
+	iref.ir_updated = TRUE;
 
+#ifdef DEBUG
+	if (ind->mi_lazy_delete)
+		ASSERT_NS(item->i_pos.i_total_size <= XT_INDEX_PAGE_DATA_SIZE);
+#endif
 	if (item->i_pos.i_total_size <= XT_INDEX_PAGE_DATA_SIZE)
 		return xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
 
@@ -1184,6 +1446,7 @@ static xtBool idx_replace_node_key(XTOpenTablePtr ot, XTIndexPtr ind, IdxStackIt
 	/* Change the size of the old branch: */
 	XT_SET_DISK_2(iref.ir_branch->tb_size_2, XT_MAKE_NODE_SIZE(result.sr_item.i_item_offset));
 	IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(current), (int) XT_GET_DISK_2(iref.ir_branch->tb_size_2));
+	iref.ir_updated = TRUE;
 
 	xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
 
@@ -1237,7 +1500,8 @@ static xtBool idx_insert_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackP
 	XTIdxBranchDPtr		new_branch_ptr;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	/* Insert a new branch (key, data)... */
 	if (!(stack_item = idx_pop(stack))) {
@@ -1268,7 +1532,7 @@ static xtBool idx_insert_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackP
 	 * cache, and will remain in cache when we read again below for the
 	 * purpose of update.
 	 */
-	if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+	if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 		goto failed;
 	ASSERT_NS(XT_IS_NODE(XT_GET_DISK_2(iref.ir_branch->tb_size_2)));
 	ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, key_value, &result);
@@ -1280,6 +1544,7 @@ static xtBool idx_insert_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackP
 		}
 		idx_insert_node_item(ot->ot_table, ind, iref.ir_branch, key_value, &result, branch);
 		IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(current), (int) XT_GET_DISK_2(ot->ot_ind_wbuf.tb_size_2));
+		iref.ir_updated = TRUE;
 		ASSERT_NS(result.sr_item.i_total_size <= XT_INDEX_PAGE_DATA_SIZE);
 		xt_ind_release(ot, ind, XT_UNLOCK_R_UPDATE, &iref);
 		goto done_ok;
@@ -1314,6 +1579,7 @@ static xtBool idx_insert_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackP
 			goto failed_2;
 	}
 	memcpy(iref.ir_branch, &ot->ot_ind_wbuf, offsetof(XTIdxBranchDRec, tb_data) + result.sr_item.i_item_offset);
+	iref.ir_updated = TRUE;
 	xt_ind_release(ot, ind, XT_UNLOCK_R_UPDATE, &iref);
 
 	/* Insert the new branch into the parent node, using the new middle key value: */
@@ -1373,7 +1639,8 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 	XTXactWaitRec		xw;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	retry:
 	idx_newstack(&stack);
@@ -1385,7 +1652,7 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 	key_value->sv_flags = 0;
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref)) {
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref)) {
 			key_value->sv_flags = save_flags;
 			return FAILED;
 		}
@@ -1422,7 +1689,7 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 			while ((node = idx_pop(&stack))) {
 				if (node->i_pos.i_item_offset < node->i_pos.i_total_size) {
 					current = node->i_branch;
-					if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+					if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 						return FAILED;
 					xt_get_res_record_ref(&iref.ir_branch->tb_data[node->i_pos.i_item_offset + node->i_pos.i_item_size - XT_RECORD_REF_SIZE], &result);
 					result.sr_item = node->i_pos;
@@ -1439,6 +1706,11 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 			break;
 		}
 
+		if (ind->mi_lazy_delete) {
+			if (result.sr_row_id == (xtRowID) -1)
+				goto next_item;
+		}
+
 		switch (xt_tab_maybe_committed(ot, result.sr_rec_id, &xn_id, NULL, NULL)) {
 			case XT_MAYBE:
 				/* Record is not committed, wait for the transaction. */
@@ -1464,6 +1736,7 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 				break;
 		}
 
+		next_item:
 		idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
 
 		if (result.sr_item.i_node_ref_size) {
@@ -1473,7 +1746,7 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 				if (!idx_push(&stack, current, &result.sr_item))
 					return FAILED;
 				current = result.sr_branch;
-				if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+				if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 					return FAILED;
 				idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
 				if (!result.sr_item.i_node_ref_size)
@@ -1489,6 +1762,14 @@ static xtBool idx_check_duplicates(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyVa
 	return FAILED;
 }
 
+inline static void idx_still_on_key(XTIndexPtr ind, register XTIdxSearchKeyPtr search_key, register XTIdxBranchDPtr branch, register XTIdxItemPtr item)
+{
+	if (search_key && search_key->sk_on_key) {
+		search_key->sk_on_key = myxt_compare_key(ind, search_key->sk_key_value.sv_flags, search_key->sk_key_value.sv_length,
+			search_key->sk_key_value.sv_key, &branch->tb_data[item->i_item_offset]) == 0;
+	}
+}
+
 /*
  * Insert a value into the given index. Return FALSE if an error occurs.
  */
@@ -1506,9 +1787,11 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 	size_t				new_size;
 	xtBool				check_for_dups = ind->mi_flags & (HA_UNIQUE_CHECK | HA_NOSAME) && !allow_dups;
 	xtBool				lock_structure = FALSE;
+	xtBool				updated = FALSE;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 #ifdef CHECK_AND_PRINT
 	//idx_check_index(ot, ind, TRUE);
@@ -1559,6 +1842,7 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 		XT_INDEX_READ_LOCK(ind, ot);
 
 	retry:
+	/* Create a root node if required: */
 	if (!(XT_NODE_ID(current) = XT_NODE_ID(ind->mi_root))) {
 		/* Index is empty, create a new one: */
 		ASSERT_NS(lock_structure);
@@ -1575,8 +1859,9 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 		goto done_ok;
 	}
 
+	/* Search down the tree for the insertion point. */
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_XLOCK_LEAF, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_XLOCK_LEAF, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &key_value, &result);
 		if (result.sr_duplicate) {
@@ -1601,8 +1886,23 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 			}
 		}
 		if (result.sr_found) {
-			/* Node found, can happen during recovery of indexes! */
-			XTPageUnlockType utype;
+			/* Node found, can happen during recovery of indexes! 
+			 * We have found an exact match of both key and record.
+			 */
+			XTPageUnlockType	utype;
+			xtBool				overwrite = FALSE;
+
+			/* {LAZY-DEL-INDEX-ITEMS}
+			 * If the item has been lazy deleted, then just overwrite!
+			 */ 
+			if (result.sr_row_id == (xtRowID) -1) {
+				xtWord2 del_count;
+	
+				/* This is safe because we have an xlock on the leaf. */
+				if ((del_count = iref.ir_block->cp_del_count))
+					iref.ir_block->cp_del_count = del_count-1;
+				overwrite = TRUE;
+			}
 
 			if (!result.sr_row_id && row_id) {
 				/* {INDEX-RECOV_ROWID} Set the row-id
@@ -1610,20 +1910,11 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 				 * is not committed.
 				 * It will be removed later by the sweeper.
 				 */
-				size_t	offset;
-				xtWord1	*data;
-
-				offset = 
-					/* This is the offset of the reference in the item we found: */
-					result.sr_item.i_item_offset + result.sr_item.i_item_size - XT_RECORD_REF_SIZE +
-					/* This is the offset of the row id in the reference: */
-					4;
-				data = &iref.ir_branch->tb_data[offset];
-
-				/* This update does not change the structure of page, so we do it without
-				 * copying the page before we write.
-				 */
-				XT_SET_DISK_4(data, row_id);
+				overwrite = TRUE;
+			}
+
+			if (overwrite) {
+				idx_set_item_row_id(&iref, &result.sr_item, row_id);
 				utype = result.sr_item.i_node_ref_size ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_W_UPDATE;
 			}
 			else
@@ -1644,14 +1935,84 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 	/* Must be a leaf!: */
 	ASSERT_NS(!result.sr_item.i_node_ref_size);
 
+	updated = FALSE;
+	if (ind->mi_lazy_delete && iref.ir_block->cp_del_count) {
+		/* There are a number of possibilities:
+		 * - We could just replace a lazy deleted slot.
+		 * - We could compact and insert.
+		 * - We could just insert
+		 */
+
+		if (result.sr_item.i_item_offset > 0) {
+			/* Check if it can go into the previous node: */
+			XTIdxResultRec	t_res;
+
+			t_res.sr_item = result.sr_item;
+			xt_prev_branch_item_fix(ot->ot_table, ind, iref.ir_branch, &t_res);
+			if (t_res.sr_row_id != (xtRowID) -1)
+				goto try_current;
+
+			/* Yup, it can, but first check to see if it would be 
+			 * better to put it in the current node.
+			 * This is the case if the previous node key is not the
+			 * same as the key we are adding...
+			 */
+			if (result.sr_item.i_item_offset < result.sr_item.i_total_size &&
+				result.sr_row_id == (xtRowID) -1) {
+				if (!idx_cmp_item_key_fix(&iref, &t_res.sr_item, &key_value))
+					goto try_current;
+			}
+
+			idx_set_item_key_fix(&iref, &t_res.sr_item, &key_value);
+			iref.ir_block->cp_del_count--;
+			xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
+			goto done_ok;
+		}
+
+		try_current:
+		if (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+			if (result.sr_row_id == (xtRowID) -1) {
+				idx_set_item_key_fix(&iref, &result.sr_item, &key_value);
+				iref.ir_block->cp_del_count--;
+				xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
+				goto done_ok;
+			}
+		}
+
+		/* Check if we must compact... 
+		 * It makes no sense to split as long as there are lazy deleted items
+		 * in the page. So, delete them if a split would otherwise be required!
+		 */
+		ASSERT_NS(key_value.sv_length + XT_RECORD_REF_SIZE == result.sr_item.i_item_size);
+		if (result.sr_item.i_total_size + key_value.sv_length + XT_RECORD_REF_SIZE > XT_INDEX_PAGE_DATA_SIZE) {
+			if (!idx_compact_leaf(ot, ind, &iref, &result.sr_item))
+				goto failed;
+			updated = TRUE;
+		}
+		
+		/* Fall through to the insert code... */
+		/* NOTE: if there were no lazy deleted items in the leaf, then
+		 * idx_compact_leaf is a NOP. This is the only case in which it may not
+		 * fall through and do the insert below.
+		 *
+		 * Normally, if the cp_del_count is correct then the insert
+		 * will work below, and the assertion here will not fail.
+		 *
+		 * In this case, the xt_ind_release() will correctly indicate an update.
+		 */
+		ASSERT_NS(result.sr_item.i_total_size + key_value.sv_length + XT_RECORD_REF_SIZE <= XT_INDEX_PAGE_DATA_SIZE);
+	}
+
 	if (result.sr_item.i_total_size + key_value.sv_length + XT_RECORD_REF_SIZE <= XT_INDEX_PAGE_DATA_SIZE) {
 		if (iref.ir_block->cb_handle_count) {
 			if (!xt_ind_copy_on_write(&iref))
 				goto failed_1;
 		}
+
 		idx_insert_leaf_item(ind, iref.ir_branch, &key_value, &result);
 		IDX_TRACE("%d-> %x\n", (int) XT_NODE_ID(current), (int) XT_GET_DISK_2(ot->ot_ind_wbuf.tb_size_2));
 		ASSERT_NS(result.sr_item.i_total_size <= XT_INDEX_PAGE_DATA_SIZE);
+		iref.ir_updated = TRUE;
 		xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
 		goto done_ok;
 	}
@@ -1660,7 +2021,7 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 	 * Make sure we have a structural lock:
 	 */
 	if (!lock_structure) {
-		xt_ind_release(ot, ind, XT_UNLOCK_WRITE, &iref);
+		xt_ind_release(ot, ind, updated ? XT_UNLOCK_W_UPDATE : XT_UNLOCK_WRITE, &iref);
 		XT_INDEX_UNLOCK(ind, ot);
 		lock_structure = TRUE;
 		goto lock_and_retry;
@@ -1705,6 +2066,7 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 			goto failed_2;
 	}
 	memcpy(iref.ir_branch, &ot->ot_ind_wbuf, offsetof(XTIdxBranchDRec, tb_data) + result.sr_item.i_item_offset);
+	iref.ir_updated = TRUE;
 	xt_ind_release(ot, ind, XT_UNLOCK_W_UPDATE, &iref);
 
 	/* Insert the new branch into the parent node, using the new middle key value: */
@@ -1732,7 +2094,7 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 	idx_free_branch(ot, ind, new_branch);
 
 	failed_1:
-	xt_ind_release(ot, ind, XT_UNLOCK_WRITE, &iref);
+	xt_ind_release(ot, ind, updated ? XT_UNLOCK_W_UPDATE : XT_UNLOCK_WRITE, &iref);
 
 	failed:
 	XT_INDEX_UNLOCK(ind, ot);
@@ -1747,18 +2109,175 @@ xtPublic xtBool xt_idx_insert(XTOpenTablePtr ot, XTIndexPtr ind, xtRowID row_id,
 	return FAILED;
 }
 
+
+/* Remove the given item in the node.
+ * This is done by going down the tree to find a replacement
+ * for the deleted item!
+ */
+static xtBool idx_remove_item_in_node(XTOpenTablePtr ot, XTIndexPtr ind, IdxBranchStackPtr stack, XTIndReferencePtr iref, XTIdxKeyValuePtr key_value)
+{
+	IdxStackItemPtr		delete_node;
+	XTIdxResultRec		result;
+	xtIndexNodeID		current;
+	xtBool				lazy_delete_cleanup_required = FALSE;
+	IdxStackItemPtr		current_top;
+
+	delete_node = idx_top(stack);
+	current = delete_node->i_branch;
+	result.sr_item = delete_node->i_pos;
+
+	/* Follow the branch after this item: */
+	idx_next_branch_item(ot->ot_table, ind, iref->ir_branch, &result);
+	xt_ind_release(ot, ind, iref->ir_updated ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_READ, iref);
+
+	/* Go down the left-hand side until we reach a leaf: */
+	while (XT_NODE_ID(current)) {
+		current = result.sr_branch;
+		if (!xt_ind_fetch(ot, ind, current, XT_XLOCK_LEAF, iref))
+			return FAILED;
+		idx_first_branch_item(ot->ot_table, ind, iref->ir_branch, &result);
+		if (!result.sr_item.i_node_ref_size)
+			break;
+		xt_ind_release(ot, ind, XT_UNLOCK_READ, iref);
+		if (!idx_push(stack, current, &result.sr_item))
+			return FAILED;
+	}
+
+	ASSERT_NS(XT_NODE_ID(current));
+	ASSERT_NS(!result.sr_item.i_node_ref_size);
+
+	if (!xt_ind_reserve(ot, stack->s_top + 2, iref->ir_branch)) {
+		xt_ind_release(ot, ind, XT_UNLOCK_WRITE, iref);
+		return FAILED;
+	}
+	
+	/* This code removes lazy deleted items from the leaf,
+	 * before we promote an item to a leaf.
+	 * This is not essential, but prevents lazy deleted
+	 * items from being propogated up the tree.
+	 */
+	if (ind->mi_lazy_delete) {
+		if (iref->ir_block->cp_del_count) {
+			if (!idx_compact_leaf(ot, ind, iref, &result.sr_item))
+				return FAILED;
+		}
+	}
+
+	/* Crawl back up the stack trace, looking for a key
+	 * that can be used to replace the deleted key.
+	 *
+	 * Any empty nodes on the way up can be removed!
+	 */
+	if (result.sr_item.i_total_size > 0) {
+		/* There is a key in the leaf, extract it, and put it in the node: */
+		memcpy(key_value->sv_key, &iref->ir_branch->tb_data[result.sr_item.i_item_offset], result.sr_item.i_item_size);
+		/* This call also frees the iref.ir_branch page! */
+		if (!idx_remove_branch_item_right(ot, ind, current, iref, &result.sr_item))
+			return FAILED;
+		if (!idx_replace_node_key(ot, ind, delete_node, stack, result.sr_item.i_item_size, key_value->sv_key))
+			return FAILED;
+		goto done_ok;
+	}
+
+	xt_ind_release(ot, ind, iref->ir_updated ? XT_UNLOCK_W_UPDATE : XT_UNLOCK_WRITE, iref);
+
+	for (;;) {
+		/* The current node/leaf is empty, remove it: */
+		idx_free_branch(ot, ind, current);
+
+		current_top = idx_pop(stack);
+		current = current_top->i_branch;
+		if (!xt_ind_fetch(ot, ind, current, XT_XLOCK_LEAF, iref))
+			return FAILED;
+		
+		if (current_top == delete_node) {
+			/* All children have been removed. Delete the key and done: */
+			if (!idx_remove_branch_item_right(ot, ind, current, iref, &current_top->i_pos))
+				return FAILED;
+			goto done_ok;
+		}
+
+		if (current_top->i_pos.i_total_size > current_top->i_pos.i_node_ref_size) {
+			/* Save the key: */
+			memcpy(key_value->sv_key, &iref->ir_branch->tb_data[current_top->i_pos.i_item_offset], current_top->i_pos.i_item_size);
+			/* This function also frees the cache page: */
+			if (!idx_remove_branch_item_left(ot, ind, current, iref, &current_top->i_pos, &lazy_delete_cleanup_required))
+				return FAILED;
+			if (!idx_replace_node_key(ot, ind, delete_node, stack, current_top->i_pos.i_item_size, key_value->sv_key))
+				return FAILED;
+			/* */
+			if (lazy_delete_cleanup_required) {
+				if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, iref))
+					return FAILED;
+				if (!idx_remove_lazy_deleted_item_in_node(ot, ind, current, iref, key_value))
+					return FAILED;
+			}
+			goto done_ok;
+		}
+		xt_ind_release(ot, ind, current_top->i_pos.i_node_ref_size ? XT_UNLOCK_READ : XT_UNLOCK_WRITE, iref);
+	}
+
+	done_ok:
+#ifdef XT_TRACK_INDEX_UPDATES
+	ASSERT_NS(ot->ot_ind_reserved >= ot->ot_ind_reads);
+#endif
+	return OK;
+}
+
+/*
+ * This function assumes we have a lock on the structure of the index.
+ */
+static xtBool idx_remove_lazy_deleted_item_in_node(XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID current, XTIndReferencePtr iref, XTIdxKeyValuePtr key_value)
+{
+	IdxBranchStackRec	stack;
+	XTIdxResultRec		result;
+
+	/* Now remove all lazy deleted items in this node.... */
+	idx_first_branch_item(ot->ot_table, ind, (XTIdxBranchDPtr) iref->ir_block->cb_data, &result);
+
+	for (;;) {
+		while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+			if (result.sr_row_id == (xtRowID) -1)
+				goto remove_item;
+			idx_next_branch_item(ot->ot_table, ind, (XTIdxBranchDPtr) iref->ir_block->cb_data, &result);
+		}
+		break;
+
+		remove_item:
+
+		idx_newstack(&stack);
+		if (!idx_push(&stack, current, &result.sr_item)) {
+			xt_ind_release(ot, ind, iref->ir_updated ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_READ, iref);
+			return FAILED;
+		}
+
+		if (!idx_remove_item_in_node(ot, ind, &stack, iref, key_value))
+			return FAILED;
+
+		/* Go back up to the node we are trying to
+		 * free of things.
+		 */
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, iref))
+			return FAILED;
+		/* Load the data again: */
+		idx_reload_item_fix(ind, iref->ir_branch, &result);
+	}
+
+	xt_ind_release(ot, ind, iref->ir_updated ? XT_UNLOCK_R_UPDATE : XT_UNLOCK_READ, iref);
+	return OK;
+}
+
 static xtBool idx_delete(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyValuePtr key_value)
 {
 	IdxBranchStackRec	stack;
 	xtIndexNodeID		current;
 	XTIndReferenceRec	iref;
 	XTIdxResultRec		result;
-	IdxStackItemPtr		delete_node = NULL;
-	IdxStackItemPtr		current_top = NULL;
 	xtBool				lock_structure = FALSE;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	/* The index appears to have no root: */
 	if (!XT_NODE_ID(ind->mi_root))
@@ -1776,17 +2295,37 @@ static xtBool idx_delete(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyValuePtr key
 		goto done_ok;
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_XLOCK_LEAF, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_XLOCK_DEL_LEAF, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, key_value, &result);
 		if (!result.sr_item.i_node_ref_size) {
 			/* A leaf... */
 			if (result.sr_found) {
-				if (!idx_remove_branch_item_right(ot, ind, current, &iref, &result.sr_item))
-					goto failed;
+				if (ind->mi_lazy_delete) {
+					/* If the we have a W lock, then fetch decided that we
+					 * need to compact the page.
+					 * The decision is made by xt_idx_lazy_delete_on_leaf() 
+					 */
+					if (!iref.ir_xlock)
+						idx_lazy_delete_branch_item(ot, ind, &iref, &result.sr_item);
+					else {
+						if (!iref.ir_block->cp_del_count) {
+							if (!idx_remove_branch_item_right(ot, ind, current, &iref, &result.sr_item))
+								goto failed;
+						}
+						else {
+							if (!idx_lazy_remove_leaf_item_right(ot, ind, &iref, &result.sr_item))
+								goto failed;
+						}
+					}
+				}
+				else {
+					if (!idx_remove_branch_item_right(ot, ind, current, &iref, &result.sr_item))
+						goto failed;
+				}
 			}
 			else
-				xt_ind_release(ot, ind, XT_UNLOCK_WRITE, &iref);
+				xt_ind_release(ot, ind, iref.ir_xlock ? XT_UNLOCK_WRITE : XT_UNLOCK_READ, &iref);
 			goto done_ok;
 		}
 		if (!idx_push(&stack, current, &result.sr_item)) {
@@ -1803,6 +2342,35 @@ static xtBool idx_delete(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyValuePtr key
 	/* Must be a non-leaf!: */
 	ASSERT_NS(result.sr_item.i_node_ref_size);
 
+	if (ind->mi_lazy_delete) {
+		if (!idx_lazy_delete_on_node(ind, iref.ir_block, &result.sr_item)) {
+			/* We need to remove some items from this node: */
+
+			if (!lock_structure) {
+				xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
+				XT_INDEX_UNLOCK(ind, ot);
+				lock_structure = TRUE;
+				goto lock_and_retry;
+			}
+
+			idx_set_item_deleted(&iref, &result.sr_item);
+			if (!idx_remove_lazy_deleted_item_in_node(ot, ind, current, &iref, key_value))
+				goto failed;
+			goto done_ok;
+		}
+
+		if (!ot->ot_table->tab_dic.dic_no_lazy_delete) {
+			/* {LAZY-DEL-INDEX-ITEMS}
+			 * We just set item to deleted, this is a significant time
+			 * saver.
+			 * But this item can only be cleaned up when all
+			 * items on the node below are deleted.
+			 */
+			idx_lazy_delete_branch_item(ot, ind, &iref, &result.sr_item);
+			goto done_ok;
+		}
+	}
+
 	/* We will have to remove the key from a non-leaf node,
 	 * which means we are changing the structure of the index.
 	 * Make sure we have a structural lock:
@@ -1815,86 +2383,8 @@ static xtBool idx_delete(XTOpenTablePtr ot, XTIndexPtr ind, XTIdxKeyValuePtr key
 	}
 
 	/* This is the item we will have to replace: */
-	delete_node = idx_top(&stack);
-
-	/* Follow the branch after this item: */
-	idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
-	ASSERT_NS(XT_NODE_ID(current));
-	xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
-
-	/* Go down the left-hand side until we reach a leaf: */
-	while (XT_NODE_ID(current)) {
-		current = result.sr_branch;
-		if (!xt_ind_fetch(ot, current, XT_XLOCK_LEAF, &iref))
-			goto failed;
-		idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
-		if (!result.sr_item.i_node_ref_size)
-			break;
-		xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
-		if (!idx_push(&stack, current, &result.sr_item))
-			goto failed;
-	}
-
-	ASSERT_NS(XT_NODE_ID(current));
-	ASSERT_NS(!result.sr_item.i_node_ref_size);
-
-	if (!xt_ind_reserve(ot, stack.s_top + 2, iref.ir_branch)) {
-		xt_ind_release(ot, ind, XT_UNLOCK_WRITE, &iref);
+	if (!idx_remove_item_in_node(ot, ind, &stack, &iref, key_value))
 		goto failed;
-	}
-
-	/* Crawl back up the stack trace, looking for a key
-	 * that can be used to replace the deleted key.
-	 *
-	 * Any empty nodes on the way up can be removed!
-	 */
-	if (result.sr_item.i_total_size > 0) {
-		/* There is a key in the leaf, extract it, and put it in the node: */
-		memcpy(key_value->sv_key, &iref.ir_branch->tb_data[result.sr_item.i_item_offset], result.sr_item.i_item_size);
-		/* This call also frees the iref.ir_branch page! */
-		if (!idx_remove_branch_item_right(ot, ind, current, &iref, &result.sr_item))
-			goto failed;
-		if (!idx_replace_node_key(ot, ind, delete_node, &stack, result.sr_item.i_item_size, key_value->sv_key))
-			goto failed;
-		goto done_ok_2;
-	}
-
-	xt_ind_release(ot, ind, XT_UNLOCK_WRITE, &iref);
-
-	for (;;) {
-		/* The current node/leaf is empty, remove it: */
-		idx_free_branch(ot, ind, current);
-
-		current_top = idx_pop(&stack);
-		current = current_top->i_branch;
-		if (!xt_ind_fetch(ot, current, XT_XLOCK_LEAF, &iref))
-			goto failed;
-		
-		if (current_top == delete_node) {
-			/* All children have been removed. Delete the key and done: */
-			if (!idx_remove_branch_item_right(ot, ind, current, &iref, &current_top->i_pos))
-				goto failed;
-			goto done_ok_2;
-		}
-
-		if (current_top->i_pos.i_total_size > current_top->i_pos.i_node_ref_size) {
-			/* Save the key: */
-			memcpy(key_value->sv_key, &iref.ir_branch->tb_data[current_top->i_pos.i_item_offset], current_top->i_pos.i_item_size);
-			/* This function also frees the cache page: */
-			if (!idx_remove_branch_item_left(ot, ind, current, &iref, &current_top->i_pos))
-				goto failed;
-			if (!idx_replace_node_key(ot, ind, delete_node, &stack, current_top->i_pos.i_item_size, key_value->sv_key))
-				goto failed;
-			goto done_ok_2;
-		}
-		xt_ind_release(ot, ind, current_top->i_pos.i_node_ref_size ? XT_UNLOCK_READ : XT_UNLOCK_WRITE, &iref);
-	}
-
-
-	done_ok_2:
-#ifdef XT_TRACK_INDEX_UPDATES
-	ASSERT_NS(ot->ot_ind_reserved >= ot->ot_ind_reads);
-#endif
 
 	done_ok:
 	XT_INDEX_UNLOCK(ind, ot);
@@ -1945,7 +2435,8 @@ xtPublic xtBool xt_idx_update_row_id(XTOpenTablePtr ot, XTIndexPtr ind, xtRecord
 	xtWord1				key_buf[XT_INDEX_MAX_KEY_SIZE + XT_MAX_RECORD_REF_SIZE];
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 #ifdef CHECK_AND_PRINT
 	idx_check_index(ot, ind, TRUE);
@@ -1989,7 +2480,7 @@ xtPublic xtBool xt_idx_update_row_id(XTOpenTablePtr ot, XTIndexPtr ind, xtRecord
 		goto done_ok;
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &key_value, &result);
 		if (result.sr_found || !result.sr_item.i_node_ref_size)
@@ -1999,23 +2490,10 @@ xtPublic xtBool xt_idx_update_row_id(XTOpenTablePtr ot, XTIndexPtr ind, xtRecord
 	}
 
 	if (result.sr_found) {
-		size_t	offset;
-		xtWord1	*data;
-
-		offset = 
-			/* This is the offset of the reference in the item we found: */
-			result.sr_item.i_item_offset + result.sr_item.i_item_size - XT_RECORD_REF_SIZE +
-			/* This is the offset of the row id in the reference: */
-			4;
-		data = &iref.ir_branch->tb_data[offset];
-
-		/* This update does not change the structure of page, so we do it without
-		 * copying the page before we write.
-		 *
-		 * TODO: Check that concurrent reads can handle this!
+		/* TODO: Check that concurrent reads can handle this!
 		 * assuming the write is not atomic.
 		 */
-		XT_SET_DISK_4(data, row_id);
+		idx_set_item_row_id(&iref, &result.sr_item, row_id);
 		xt_ind_release(ot, ind, XT_UNLOCK_R_UPDATE, &iref);
 	}
 	else
@@ -2076,7 +2554,8 @@ xtPublic xtBool xt_idx_search(XTOpenTablePtr ot, XTIndexPtr ind, register XTIdxS
 	XTIdxResultRec		result;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	if (ot->ot_ind_rhandle) {
 		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, ot->ot_thread);
@@ -2110,7 +2589,7 @@ xtPublic xtBool xt_idx_search(XTOpenTablePtr ot, XTIndexPtr ind, register XTIdxS
 		goto done_ok;
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &search_key->sk_key_value, &result);
 		if (result.sr_found)
@@ -2124,6 +2603,17 @@ xtPublic xtBool xt_idx_search(XTOpenTablePtr ot, XTIndexPtr ind, register XTIdxS
 		current = result.sr_branch;
 	}
 
+	if (ind->mi_lazy_delete) {
+		ignore_lazy_deleted_items:
+		while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+			if (result.sr_row_id != (xtRowID) -1) {
+				idx_still_on_key(ind, search_key, iref.ir_branch, &result.sr_item);
+				break;
+			}
+			idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+		}
+	}
+
 	if (result.sr_item.i_item_offset == result.sr_item.i_total_size) {
 		IdxStackItemPtr node;
 
@@ -2134,12 +2624,39 @@ xtPublic xtBool xt_idx_search(XTOpenTablePtr ot, XTIndexPtr ind, register XTIdxS
 		xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
 		while ((node = idx_pop(&stack))) {
 			if (node->i_pos.i_item_offset < node->i_pos.i_total_size) {
-				xtRecordID rec_id;
-
-				if (!xt_ind_fetch(ot, node->i_branch, XT_LOCK_READ, &iref))
+				if (!xt_ind_fetch(ot, ind, node->i_branch, XT_LOCK_READ, &iref))
 					goto failed;
-				xt_get_record_ref(&iref.ir_branch->tb_data[node->i_pos.i_item_offset + node->i_pos.i_item_size - XT_RECORD_REF_SIZE], &rec_id, &ot->ot_curr_row_id);
-				ot->ot_curr_rec_id = rec_id;
+				xt_get_res_record_ref(&iref.ir_branch->tb_data[node->i_pos.i_item_offset + node->i_pos.i_item_size - XT_RECORD_REF_SIZE], &result);
+
+				if (ind->mi_lazy_delete) {
+					result.sr_item = node->i_pos;
+					if (result.sr_row_id == (xtRowID) -1) {
+						/* If this node pointer is lazy deleted, then
+						 * go down the next branch...
+						 */
+						idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+
+						/* Go down to the bottom: */
+						current = node->i_branch;
+						while (XT_NODE_ID(current)) {
+							xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
+							if (!idx_push(&stack, current, &result.sr_item))
+								goto failed;
+							current = result.sr_branch;
+							if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
+								goto failed;
+							idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+							if (!result.sr_item.i_node_ref_size)
+								break;
+						}
+
+						goto ignore_lazy_deleted_items;
+					}
+					idx_still_on_key(ind, search_key, iref.ir_branch, &result.sr_item);
+				}
+
+				ot->ot_curr_rec_id = result.sr_rec_id;
+				ot->ot_curr_row_id = result.sr_row_id;
 				ot->ot_ind_state = node->i_pos;
 
 				/* Convert the pointer to a handle which can be used in later operations: */
@@ -2180,14 +2697,16 @@ xtPublic xtBool xt_idx_search(XTOpenTablePtr ot, XTIndexPtr ind, register XTIdxS
 	//idx_check_index(ot, ind, TRUE);
 	//idx_check_on_key(ot);
 #endif
-	ASSERT_NS(iref.ir_ulock == XT_UNLOCK_NONE);
+	ASSERT_NS(iref.ir_xlock == 2);
+	ASSERT_NS(iref.ir_updated == 2);
 	return OK;
 
 	failed:
 	XT_INDEX_UNLOCK(ind, ot);
 	if (idx_out_of_memory_failure(ot))
 		goto retry_after_oom;
-	ASSERT_NS(iref.ir_ulock == XT_UNLOCK_NONE);
+	ASSERT_NS(iref.ir_xlock == 2);
+	ASSERT_NS(iref.ir_updated == 2);
 	return FAILED;
 }
 
@@ -2199,7 +2718,8 @@ xtPublic xtBool xt_idx_search_prev(XTOpenTablePtr ot, XTIndexPtr ind, register X
 	XTIdxResultRec		result;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	if (ot->ot_ind_rhandle) {
 		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, ot->ot_thread);
@@ -2232,7 +2752,7 @@ xtPublic xtBool xt_idx_search_prev(XTOpenTablePtr ot, XTIndexPtr ind, register X
 		goto done_ok;
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &search_key->sk_key_value, &result);
 		if (result.sr_found)
@@ -2249,17 +2769,43 @@ xtPublic xtBool xt_idx_search_prev(XTOpenTablePtr ot, XTIndexPtr ind, register X
 	if (result.sr_item.i_item_offset == 0) {
 		IdxStackItemPtr node;
 
-		/* We are at the end of a leaf node.
-		 * Go up the stack to find the start poition of the next key.
+		search_up_stack:
+		/* We are at the start of a leaf node.
+		 * Go up the stack to find the start position of the next key.
 		 * If we find none, then we are the end of the index.
 		 */
 		xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
 		while ((node = idx_pop(&stack))) {
 			if (node->i_pos.i_item_offset > node->i_pos.i_node_ref_size) {
-				if (!xt_ind_fetch(ot, node->i_branch, XT_LOCK_READ, &iref))
+				if (!xt_ind_fetch(ot, ind, node->i_branch, XT_LOCK_READ, &iref))
 					goto failed;
 				result.sr_item = node->i_pos;
 				ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+
+				if (ind->mi_lazy_delete) {
+					if (result.sr_row_id == (xtRowID) -1) {
+						/* Go down to the bottom, in order to scan the leaf backwards: */
+						current = node->i_branch;
+						while (XT_NODE_ID(current)) {
+							xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
+							if (!idx_push(&stack, current, &result.sr_item))
+								goto failed;
+							current = result.sr_branch;
+							if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
+								goto failed;
+							ind->mi_last_item(ot->ot_table, ind, iref.ir_branch, &result);
+							if (!result.sr_item.i_node_ref_size)
+								break;
+						}
+
+						/* If the leaf empty we have to go up the stack again... */
+						if (result.sr_item.i_total_size == 0)
+							goto search_up_stack;
+
+						goto scan_back_in_leaf;
+					}
+				}
+
 				goto record_found;
 			}
 		}
@@ -2269,6 +2815,16 @@ xtPublic xtBool xt_idx_search_prev(XTOpenTablePtr ot, XTIndexPtr ind, register X
 	/* We must just step once to the left in this leaf node... */
 	ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
 
+	if (ind->mi_lazy_delete) {
+		scan_back_in_leaf:
+		while (result.sr_row_id == (xtRowID) -1) {
+			if (result.sr_item.i_item_offset == 0)
+				goto search_up_stack;
+			ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+		}
+		idx_still_on_key(ind, search_key, iref.ir_branch, &result.sr_item);
+	}
+
 	record_found:
 	ot->ot_curr_rec_id = result.sr_rec_id;
 	ot->ot_curr_row_id = result.sr_row_id;
@@ -2330,34 +2886,47 @@ xtPublic xtBool xt_idx_next(register XTOpenTablePtr ot, register XTIndexPtr ind,
 	XTIndReferenceRec	iref;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	ASSERT_NS(ot->ot_ind_rhandle);
 	xt_ind_lock_handle(ot->ot_ind_rhandle);
-	if (!ot->ot_ind_state.i_node_ref_size && 
-		ot->ot_ind_state.i_item_offset < ot->ot_ind_state.i_total_size && 
+	result.sr_item = ot->ot_ind_state;
+	if (!result.sr_item.i_node_ref_size && 
+		result.sr_item.i_item_offset < result.sr_item.i_total_size && 
 		ot->ot_ind_rhandle->ih_cache_reference) {
-		key_value.sv_key = &ot->ot_ind_rhandle->ih_branch->tb_data[ot->ot_ind_state.i_item_offset];
-		key_value.sv_length = ot->ot_ind_state.i_item_size - XT_RECORD_REF_SIZE;
+		XTIdxItemRec prev_item;
+
+		key_value.sv_key = &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset];
+		key_value.sv_length = result.sr_item.i_item_size - XT_RECORD_REF_SIZE;
 
-		result.sr_item = ot->ot_ind_state;
+		prev_item = result.sr_item;
 		idx_next_branch_item(ot->ot_table, ind, ot->ot_ind_rhandle->ih_branch, &result);
+
+		if (ind->mi_lazy_delete) {
+			while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+				if (result.sr_row_id != (xtRowID) -1)
+					break;
+				prev_item = result.sr_item;
+				idx_next_branch_item(ot->ot_table, ind, ot->ot_ind_rhandle->ih_branch, &result);
+			}
+		}
+
 		if (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
 			/* Still on key? */
-			if (search_key && search_key->sk_on_key) {
-				search_key->sk_on_key = myxt_compare_key(ind, search_key->sk_key_value.sv_flags, search_key->sk_key_value.sv_length,
-					search_key->sk_key_value.sv_key, &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset]) == 0;
-			}
+			idx_still_on_key(ind, search_key, ot->ot_ind_rhandle->ih_branch, &result.sr_item);
 			xt_ind_unlock_handle(ot->ot_ind_rhandle);
 			goto checked_on_key;
 		}
+
+		result.sr_item = prev_item;
 	}
 
 	key_value.sv_flags = XT_SEARCH_WHOLE_KEY;
-	xt_get_record_ref(&ot->ot_ind_rhandle->ih_branch->tb_data[ot->ot_ind_state.i_item_offset + ot->ot_ind_state.i_item_size - XT_RECORD_REF_SIZE], &key_value.sv_rec_id, &key_value.sv_row_id);
+	xt_get_record_ref(&ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset + result.sr_item.i_item_size - XT_RECORD_REF_SIZE], &key_value.sv_rec_id, &key_value.sv_row_id);
 	key_value.sv_key = key_buf;
-	key_value.sv_length = ot->ot_ind_state.i_item_size - XT_RECORD_REF_SIZE;
-	memcpy(key_buf, &ot->ot_ind_rhandle->ih_branch->tb_data[ot->ot_ind_state.i_item_offset], key_value.sv_length);
+	key_value.sv_length = result.sr_item.i_item_size - XT_RECORD_REF_SIZE;
+	memcpy(key_buf, &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset], key_value.sv_length);
 	xt_ind_release_handle(ot->ot_ind_rhandle, TRUE, ot->ot_thread);
 	ot->ot_ind_rhandle = NULL;
 
@@ -2375,7 +2944,7 @@ xtPublic xtBool xt_idx_next(register XTOpenTablePtr ot, register XTIndexPtr ind,
 	}
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &key_value, &result);
 		if (result.sr_item.i_node_ref_size) {
@@ -2389,7 +2958,7 @@ xtPublic xtBool xt_idx_next(register XTOpenTablePtr ot, register XTIndexPtr ind,
 					if (!idx_push(&stack, current, &result.sr_item))
 						goto failed;
 					current = result.sr_branch;
-					if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+					if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 						goto failed;
 					idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
 					if (!result.sr_item.i_node_ref_size)
@@ -2416,6 +2985,15 @@ xtPublic xtBool xt_idx_next(register XTOpenTablePtr ot, register XTIndexPtr ind,
 		current = result.sr_branch;
 	}
 
+	if (ind->mi_lazy_delete) {
+		ignore_lazy_deleted_items:
+		while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+			if (result.sr_row_id != (xtRowID) -1)
+				break;
+			idx_next_branch_item(NULL, ind, iref.ir_branch, &result);
+		}
+	}
+
 	/* Check the current position in a leaf: */
 	if (result.sr_item.i_item_offset == result.sr_item.i_total_size) {
 		/* At the end: */
@@ -2428,10 +3006,37 @@ xtPublic xtBool xt_idx_next(register XTOpenTablePtr ot, register XTIndexPtr ind,
 		xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
 		while ((node = idx_pop(&stack))) {
 			if (node->i_pos.i_item_offset < node->i_pos.i_total_size) {
-				if (!xt_ind_fetch(ot, node->i_branch, XT_LOCK_READ, &iref))
+				if (!xt_ind_fetch(ot, ind, node->i_branch, XT_LOCK_READ, &iref))
 					goto failed;
 				result.sr_item = node->i_pos;
 				xt_get_res_record_ref(&iref.ir_branch->tb_data[result.sr_item.i_item_offset + result.sr_item.i_item_size - XT_RECORD_REF_SIZE], &result);
+
+				if (ind->mi_lazy_delete) {
+					if (result.sr_row_id == (xtRowID) -1) {
+						/* If this node pointer is lazy deleted, then
+						 * go down the next branch...
+						 */
+						idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+
+						/* Go down to the bottom: */
+						current = node->i_branch;
+						while (XT_NODE_ID(current)) {
+							xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
+							if (!idx_push(&stack, current, &result.sr_item))
+								goto failed;
+							current = result.sr_branch;
+							if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
+								goto failed;
+							idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+							if (!result.sr_item.i_node_ref_size)
+								break;
+						}
+
+						/* And scan the leaf... */
+						goto ignore_lazy_deleted_items;
+					}
+				}
+
 				goto unlock_check_on_key;
 			}
 		}
@@ -2503,32 +3108,39 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 	IdxStackItemPtr		node;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	ASSERT_NS(ot->ot_ind_rhandle);
 	xt_ind_lock_handle(ot->ot_ind_rhandle);
-	if (!ot->ot_ind_state.i_node_ref_size && ot->ot_ind_state.i_item_offset > 0) {
-		key_value.sv_key = &ot->ot_ind_rhandle->ih_branch->tb_data[ot->ot_ind_state.i_item_offset];
-		key_value.sv_length = ot->ot_ind_state.i_item_size - XT_RECORD_REF_SIZE;
+	result.sr_item = ot->ot_ind_state;
+	if (!result.sr_item.i_node_ref_size && result.sr_item.i_item_offset > 0) {
+		key_value.sv_key = &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset];
+		key_value.sv_length = result.sr_item.i_item_size - XT_RECORD_REF_SIZE;
 
-		result.sr_item = ot->ot_ind_state;
 		ind->mi_prev_item(ot->ot_table, ind, ot->ot_ind_rhandle->ih_branch, &result);
 
-		if (search_key && search_key->sk_on_key) {
-			search_key->sk_on_key = myxt_compare_key(ind, search_key->sk_key_value.sv_flags, search_key->sk_key_value.sv_length,
-				search_key->sk_key_value.sv_key, &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset]) == 0;
+		if (ind->mi_lazy_delete) {
+			while (result.sr_row_id == (xtRowID) -1) {
+				if (result.sr_item.i_item_offset == 0)
+					goto research;
+				ind->mi_prev_item(ot->ot_table, ind, ot->ot_ind_rhandle->ih_branch, &result);
+			}
 		}
 
+		idx_still_on_key(ind, search_key, ot->ot_ind_rhandle->ih_branch, &result.sr_item);
+
 		xt_ind_unlock_handle(ot->ot_ind_rhandle);
 		goto checked_on_key;
 	}
 
+	research:
 	key_value.sv_flags = XT_SEARCH_WHOLE_KEY;
 	key_value.sv_rec_id = ot->ot_curr_rec_id;
 	key_value.sv_row_id = 0;
 	key_value.sv_key = key_buf;
-	key_value.sv_length = ot->ot_ind_state.i_item_size - XT_RECORD_REF_SIZE;
-	memcpy(key_buf, &ot->ot_ind_rhandle->ih_branch->tb_data[ot->ot_ind_state.i_item_offset], key_value.sv_length);
+	key_value.sv_length = result.sr_item.i_item_size - XT_RECORD_REF_SIZE;
+	memcpy(key_buf, &ot->ot_ind_rhandle->ih_branch->tb_data[result.sr_item.i_item_offset], key_value.sv_length);
 	xt_ind_release_handle(ot->ot_ind_rhandle, TRUE, ot->ot_thread);
 	ot->ot_ind_rhandle = NULL;
 
@@ -2546,29 +3158,39 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 	}
 
 	while (XT_NODE_ID(current)) {
-		if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+		if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 			goto failed;
 		ind->mi_scan_branch(ot->ot_table, ind, iref.ir_branch, &key_value, &result);
 		if (result.sr_item.i_node_ref_size) {
 			if (result.sr_found) {
 				/* If we have found the key in a node: */
 
+				search_down_stack:
 				/* Go down to the bottom: */
 				while (XT_NODE_ID(current)) {
 					xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
 					if (!idx_push(&stack, current, &result.sr_item))
 						goto failed;
 					current = result.sr_branch;
-					if (!xt_ind_fetch(ot, current, XT_LOCK_READ, &iref))
+					if (!xt_ind_fetch(ot, ind, current, XT_LOCK_READ, &iref))
 						goto failed;
 					ind->mi_last_item(ot->ot_table, ind, iref.ir_branch, &result);
 					if (!result.sr_item.i_node_ref_size)
 						break;
 				}
 
-				/* Is the leaf not empty, then we are done... */
+				/* If the leaf empty we have to go up the stack again... */
 				if (result.sr_item.i_total_size == 0)
 					break;
+
+				if (ind->mi_lazy_delete) {
+					while (result.sr_row_id == (xtRowID) -1) {
+						if (result.sr_item.i_item_offset == 0)
+							goto search_up_stack;
+						ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+					}
+				}
+
 				goto unlock_check_on_key;
 			}
 		}
@@ -2580,6 +3202,15 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 			if (result.sr_item.i_item_offset == 0)
 				break;
 			ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+
+			if (ind->mi_lazy_delete) {
+				while (result.sr_row_id == (xtRowID) -1) {
+					if (result.sr_item.i_item_offset == 0)
+						goto search_up_stack;
+					ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+				}
+			}
+
 			goto unlock_check_on_key;
 		}
 		xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
@@ -2588,6 +3219,7 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 		current = result.sr_branch;
 	}
 
+	search_up_stack:
 	/* We are at the start of a leaf node.
 	 * Go up the stack to find the start poition of the next key.
 	 * If we find none, then we are the end of the index.
@@ -2595,10 +3227,18 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 	xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
 	while ((node = idx_pop(&stack))) {
 		if (node->i_pos.i_item_offset > node->i_pos.i_node_ref_size) {
-			if (!xt_ind_fetch(ot, node->i_branch, XT_LOCK_READ, &iref))
+			if (!xt_ind_fetch(ot, ind, node->i_branch, XT_LOCK_READ, &iref))
 				goto failed;
 			result.sr_item = node->i_pos;
 			ind->mi_prev_item(ot->ot_table, ind, iref.ir_branch, &result);
+
+			if (ind->mi_lazy_delete) {
+				if (result.sr_row_id == (xtRowID) -1) {
+					current = node->i_branch;
+					goto search_down_stack;
+				}
+			}
+
 			goto unlock_check_on_key;
 		}
 	}
@@ -2648,7 +3288,7 @@ xtPublic xtBool xt_idx_prev(register XTOpenTablePtr ot, register XTIndexPtr ind,
 }
 
 /* Return TRUE if the record matches the current index search! */
-xtPublic xtBool xt_idx_match_search(register XTOpenTablePtr ot __attribute__((unused)), register XTIndexPtr ind, register XTIdxSearchKeyPtr search_key, xtWord1 *buf, int mode)
+xtPublic xtBool xt_idx_match_search(register XTOpenTablePtr XT_UNUSED(ot), register XTIndexPtr ind, register XTIdxSearchKeyPtr search_key, xtWord1 *buf, int mode)
 {
 	int		r;
 	xtWord1	key_buf[XT_INDEX_MAX_KEY_SIZE];
@@ -2666,7 +3306,7 @@ xtPublic xtBool xt_idx_match_search(register XTOpenTablePtr ot __attribute__((un
 	return FALSE;
 }
 
-static void idx_set_index_selectivity(XTThreadPtr self __attribute__((unused)), XTOpenTablePtr ot, XTIndexPtr ind)
+static void idx_set_index_selectivity(XTThreadPtr self, XTOpenTablePtr ot, XTIndexPtr ind)
 {
 	static const xtRecordID MAX_RECORDS = 100;
 
@@ -2834,10 +3474,11 @@ static u_int idx_check_node(XTOpenTablePtr ot, XTIndexPtr ind, int depth, xtInde
 	XTIndReferenceRec	iref;
 
 #ifdef DEBUG
-	iref.ir_ulock = XT_UNLOCK_NONE;
+	iref.ir_xlock = 2;
+	iref.ir_updated = 2;
 #endif
 	ASSERT_NS(XT_NODE_ID(node) <= XT_NODE_ID(ot->ot_table->tab_ind_eof));
-	if (!xt_ind_fetch(ot, node, XT_LOCK_READ, &iref))
+	if (!xt_ind_fetch(ot, ind, node, XT_LOCK_READ, &iref))
 		return 0;
 
 	idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
@@ -2974,7 +3615,7 @@ xtPublic void xt_check_indices(XTOpenTablePtr ot)
 			track_block_exists(current);
 #endif
 			printf("%d ", (int) XT_NODE_ID(current));
-			if (!xt_ind_read_bytes(ot, current, sizeof(XTIndFreeBlockRec), (xtWord1 *) &free_block)) {
+			if (!xt_ind_read_bytes(ot, *ind, current, sizeof(XTIndFreeBlockRec), (xtWord1 *) &free_block)) {
 				xt_log_and_clear_exception_ns();
 				break;
 			}
@@ -3000,6 +3641,88 @@ xtPublic void xt_check_indices(XTOpenTablePtr ot)
 
 /*
  * -----------------------------------------------------------------------
+ * Load index
+ */
+
+static void idx_load_node(XTThreadPtr self, XTOpenTablePtr ot, XTIndexPtr ind, xtIndexNodeID node)
+{
+	XTIdxResultRec		result;
+	XTIndReferenceRec	iref;
+
+	ASSERT_NS(XT_NODE_ID(node) <= XT_NODE_ID(ot->ot_table->tab_ind_eof));
+	if (!xt_ind_fetch(ot, ind, node, XT_LOCK_READ, &iref))
+		xt_throw(self);
+
+	idx_first_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+	if (result.sr_item.i_node_ref_size)
+		idx_load_node(self, ot, ind, result.sr_branch);
+	while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+		idx_next_branch_item(ot->ot_table, ind, iref.ir_branch, &result);
+		if (result.sr_item.i_node_ref_size)
+			idx_load_node(self, ot, ind, result.sr_branch);
+	}
+
+	xt_ind_release(ot, ind, XT_UNLOCK_READ, &iref);
+}
+
+xtPublic void xt_load_indices(XTThreadPtr self, XTOpenTablePtr ot)
+{
+	register XTTableHPtr	tab = ot->ot_table;
+	XTIndexPtr				*ind_ptr;
+	XTIndexPtr				ind;
+	xtIndexNodeID			current;
+
+	xt_lock_mutex(self, &tab->tab_ind_flush_lock);
+	pushr_(xt_unlock_mutex, &tab->tab_ind_flush_lock);
+
+	ind_ptr = tab->tab_dic.dic_keys;
+	for (u_int k=0; k<tab->tab_dic.dic_key_count; k++, ind_ptr++) {
+		ind = *ind_ptr;
+		XT_INDEX_WRITE_LOCK(ind, ot);
+		if ((XT_NODE_ID(current) = XT_NODE_ID(ind->mi_root)))
+			idx_load_node(self, ot, ind, current);
+		XT_INDEX_UNLOCK(ind, ot);
+	}
+
+	freer_(); // xt_unlock_mutex(&tab->tab_ind_flush_lock)
+}
+
+/*
+ * -----------------------------------------------------------------------
+ * Count the number of deleted entries in a node:
+ */
+
+/*
+ * {LAZY-DEL-INDEX-ITEMS}
+ *
+ * Use this function to count the number of deleted items 
+ * in a node when it is loaded.
+ *
+ * The count helps us decide of the node should be "packed".
+ */
+xtPublic void xt_ind_count_deleted_items(XTTableHPtr tab, XTIndexPtr ind, XTIndBlockPtr block)
+{
+	XTIdxResultRec		result;
+	int					del_count = 0;
+	xtWord2				branch_size;
+
+	branch_size = XT_GET_DISK_2(((XTIdxBranchDPtr) block->cb_data)->tb_size_2);
+
+	/* This is possible when reading free pages. */
+	if (XT_GET_INDEX_BLOCK_LEN(branch_size) < 2 || XT_GET_INDEX_BLOCK_LEN(branch_size) > XT_INDEX_PAGE_SIZE)
+		return;
+
+	idx_first_branch_item(tab, ind, (XTIdxBranchDPtr) block->cb_data, &result);
+	while (result.sr_item.i_item_offset < result.sr_item.i_total_size) {
+		if (result.sr_row_id == (xtRowID) -1)
+			del_count++;
+		idx_next_branch_item(tab, ind, (XTIdxBranchDPtr) block->cb_data, &result);
+	}
+	block->cp_del_count = del_count;
+}
+
+/*
+ * -----------------------------------------------------------------------
  * Index consistant flush
  */
 
@@ -3408,7 +4131,7 @@ void XTIndexLogPool::ilp_init(struct XTThread *self, struct XTDatabase *db, size
 	xt_throw(self);
 }
 
-void XTIndexLogPool::ilp_close(struct XTThread *self __attribute__((unused)), xtBool lock)
+void XTIndexLogPool::ilp_close(struct XTThread *XT_UNUSED(self), xtBool lock)
 {
 	XTIndexLogPtr	il;
 
@@ -3570,7 +4293,7 @@ xtBool XTIndexLog::il_require_space(size_t bytes, XTThreadPtr thread)
 	return OK;
 }
 
-xtBool XTIndexLog::il_write_byte(struct XTOpenTable *ot __attribute__((unused)), xtWord1 byte)
+xtBool XTIndexLog::il_write_byte(struct XTOpenTable *ot, xtWord1 byte)
 {
 	if (!il_require_space(1, ot->ot_thread))
 		return FAILED;
@@ -3579,7 +4302,7 @@ xtBool XTIndexLog::il_write_byte(struct XTOpenTable *ot __attribute__((unused)),
 	return OK;
 }
 
-xtBool XTIndexLog::il_write_word4(struct XTOpenTable *ot __attribute__((unused)), xtWord4 value)
+xtBool XTIndexLog::il_write_word4(struct XTOpenTable *ot, xtWord4 value)
 {
 	xtWord1 *buffer;
 
@@ -3591,7 +4314,7 @@ xtBool XTIndexLog::il_write_word4(struct XTOpenTable *ot __attribute__((unused))
 	return OK;
 }
 
-xtBool XTIndexLog::il_write_block(struct XTOpenTable *ot __attribute__((unused)), XTIndBlockPtr block)
+xtBool XTIndexLog::il_write_block(struct XTOpenTable *ot, XTIndBlockPtr block)
 {
 	XTIndPageDataDPtr	page_data;
 	xtIndexNodeID		node_id;
@@ -3618,7 +4341,7 @@ xtBool XTIndexLog::il_write_block(struct XTOpenTable *ot __attribute__((unused))
 	return OK;
 }
 
-xtBool XTIndexLog::il_write_header(struct XTOpenTable *ot __attribute__((unused)), size_t head_size, xtWord1 *head_buf)
+xtBool XTIndexLog::il_write_header(struct XTOpenTable *ot, size_t head_size, xtWord1 *head_buf)
 {
 	XTIndHeadDataDPtr	head_data;
 
diff --git a/storage/pbxt/src/index_xt.h b/storage/pbxt/src/index_xt.h
index b0b2813f5e1..ed4b9cef6ae 100644
--- a/storage/pbxt/src/index_xt.h
+++ b/storage/pbxt/src/index_xt.h
@@ -24,6 +24,7 @@
 #define __xt_index_h__
 
 #ifdef DRIZZLED
+#include <drizzled/definitions.h>
 #include <mysys/my_bitmap.h>
 #else
 #include <mysql_version.h>
@@ -34,7 +35,6 @@
 #include "linklist_xt.h"
 #include "datalog_xt.h"
 #include "datadic_xt.h"
-//#include "cache_xt.h"
 
 #ifndef MYSQL_VERSION_ID
 #error MYSQL_VERSION_ID must be defined!
@@ -109,7 +109,7 @@ class Field;
 
 #define XT_MAX_RECORD_REF_SIZE		8
 
-#define XT_INDEX_PAGE_DATA_SIZE		XT_INDEX_PAGE_SIZE - 2			/* NOTE: 2 == offsetof(XTIdxBranchDRec, tb_data) */
+#define XT_INDEX_PAGE_DATA_SIZE		(XT_INDEX_PAGE_SIZE - 2)			/* NOTE: 2 == offsetof(XTIdxBranchDRec, tb_data) */
 
 #define XT_MAKE_LEAF_SIZE(x)		((x) + offsetof(XTIdxBranchDRec, tb_data))
 
@@ -218,7 +218,7 @@ typedef struct XTIndFreeList {
  * in 32 threads on smalltab: runTest(SMALL_INSERT_TEST, 32, dbUrl)
  */
 /*
- * XT_INDEX_USE_RW_MUTEX:
+ * XT_INDEX_USE_RWMUTEX:
  * But the RW mutex is a close second, if not just as fast.
  * If it is at least as fast, then it is better because read lock
  * overhead is then zero.
@@ -240,17 +240,24 @@ typedef struct XTIndFreeList {
  * Latest test show that RW mutex is slightly faster:
  * 127460 to 123574 payment transactions.
  */
-#define XT_INDEX_USE_RW_MUTEX
+
+#ifdef XT_NO_ATOMICS
+#define XT_INDEX_USE_PTHREAD_RW
+#else
+//#define XT_INDEX_USE_RWMUTEX
 //#define XT_INDEX_USE_PTHREAD_RW
+//#define XT_INDEX_SPINXSLOCK
+#define XT_TAB_ROW_USE_XSMUTEX
+#endif
 
-#ifdef XT_INDEX_USE_FASTWRLOCK
-#define XT_INDEX_LOCK_TYPE				XTFastRWLockRec
-#define XT_INDEX_INIT_LOCK(s, i)		xt_fastrwlock_init(s, &(i)->mi_rwlock)
-#define XT_INDEX_FREE_LOCK(s, i)		xt_fastrwlock_free(s, &(i)->mi_rwlock)	
-#define XT_INDEX_READ_LOCK(i, o)		xt_fastrwlock_slock(&(i)->mi_rwlock, (o)->ot_thread)
-#define XT_INDEX_WRITE_LOCK(i, o)		xt_fastrwlock_xlock(&(i)->mi_rwlock, (o)->ot_thread)
-#define XT_INDEX_UNLOCK(i, o)			xt_fastrwlock_unlock(&(i)->mi_rwlock, (o)->ot_thread)
-#define XT_INDEX_HAVE_XLOCK(i, o)		TRUE
+#ifdef XT_TAB_ROW_USE_XSMUTEX
+#define XT_INDEX_LOCK_TYPE				XTXSMutexRec
+#define XT_INDEX_INIT_LOCK(s, i)		xt_xsmutex_init_with_autoname(s, &(i)->mi_rwlock)
+#define XT_INDEX_FREE_LOCK(s, i)		xt_xsmutex_free(s, &(i)->mi_rwlock)	
+#define XT_INDEX_READ_LOCK(i, o)		xt_xsmutex_slock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_WRITE_LOCK(i, o)		xt_xsmutex_xlock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_UNLOCK(i, o)			xt_xsmutex_unlock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_HAVE_XLOCK(i, o)		((i)->sxs_xlocker == (o)->ot_thread->t_id)
 #elif defined(XT_INDEX_USE_PTHREAD_RW)
 #define XT_INDEX_LOCK_TYPE				xt_rwlock_type
 #define XT_INDEX_INIT_LOCK(s, i)		xt_init_rwlock_with_autoname(s, &(i)->mi_rwlock)
@@ -259,7 +266,15 @@ typedef struct XTIndFreeList {
 #define XT_INDEX_WRITE_LOCK(i, o)		xt_xlock_rwlock_ns(&(i)->mi_rwlock)
 #define XT_INDEX_UNLOCK(i, o)			xt_unlock_rwlock_ns(&(i)->mi_rwlock)
 #define XT_INDEX_HAVE_XLOCK(i, o)		TRUE
-#else // XT_INDEX_USE_RW_MUTEX
+#elif defined(XT_INDEX_SPINXSLOCK)
+#define XT_INDEX_LOCK_TYPE				XTSpinXSLockRec
+#define XT_INDEX_INIT_LOCK(s, i)		xt_spinxslock_init_with_autoname(s, &(i)->mi_rwlock)
+#define XT_INDEX_FREE_LOCK(s, i)		xt_spinxslock_free(s, &(i)->mi_rwlock)	
+#define XT_INDEX_READ_LOCK(i, o)		xt_spinxslock_slock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_WRITE_LOCK(i, o)		xt_spinxslock_xlock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_UNLOCK(i, o)			xt_spinxslock_unlock(&(i)->mi_rwlock, (o)->ot_thread->t_id)
+#define XT_INDEX_HAVE_XLOCK(i, o)		((i)->mi_rwlock.nrw_xlocker == (o)->ot_thread->t_id)
+#else // XT_INDEX_USE_RWMUTEX
 #define XT_INDEX_LOCK_TYPE				XTRWMutexRec
 #define XT_INDEX_INIT_LOCK(s, i)		xt_rwmutex_init_with_autoname(s, &(i)->mi_rwlock)
 #define XT_INDEX_FREE_LOCK(s, i)		xt_rwmutex_free(s, &(i)->mi_rwlock)	
@@ -289,22 +304,24 @@ typedef struct XTIndex {
 	XTIndFreeListPtr	mi_free_list;				/* List of free pages for this index. */
 	
 	/* Protected by the mi_dirty_lock: */
-	XTSpinLockRec		mi_dirty_lock;			/* Spin lock protecting the dirty & free lists. */
+	XTSpinLockRec		mi_dirty_lock;				/* Spin lock protecting the dirty & free lists. */
 	struct XTIndBlock	*mi_dirty_list;				/* List of dirty pages for this index. */
 	u_int				mi_dirty_blocks;			/* Count of the dirty blocks. */
 
 	/* Index contants: */
 	u_int				mi_flags;
 	u_int				mi_key_size;
+	u_int				mi_max_items;				/* The maximum number of items that can fit in a leaf node. */
 	xtBool				mi_low_byte_first;
 	xtBool				mi_fix_key;
+	xtBool				mi_lazy_delete;				/* TRUE if index entries are "lazy deleted". */
 	u_int				mi_single_type;				/* Used when the index contains a single field. */
 	u_int				mi_select_total;
 	XTScanBranchFunc	mi_scan_branch;
 	XTPrevItemFunc		mi_prev_item;
 	XTLastItemFunc		mi_last_item;
 	XTSimpleCompFunc	mi_simple_comp_key;
-	MY_BITMAP			mi_col_map;					/* Bit-map of columns in the index. */
+	MX_BITMAP			mi_col_map;					/* Bit-map of columns in the index. */
 	u_int				mi_subset_of;				/* Indicates if this index is a complete subset of someother index. */
 	u_int				mi_seg_count;
 	XTIndexSegRec		mi_seg[200];
@@ -344,6 +361,7 @@ typedef struct XTDictionary {
 	Field				**dic_blob_cols;
 
 	/* MySQL related information. NULL when no tables are open from MySQL side! */
+	xtBool				dic_no_lazy_delete;			/* FALSE if lazy delete is OK. */
 	u_int				dic_disable_index;			/* Non-zero if the index cannot be used. */
 	u_int				dic_index_ver;				/* The version of the index. */
 	u_int				dic_key_count;
@@ -462,6 +480,8 @@ xtBool	xt_idx_prev(register struct XTOpenTable *ot, register struct XTIndex *ind
 xtBool	xt_idx_read(struct XTOpenTable *ot, struct XTIndex *ind, xtWord1 *rec_buf);
 void	xt_ind_set_index_selectivity(XTThreadPtr self, struct XTOpenTable *ot);
 void	xt_check_indices(struct XTOpenTable *ot);
+void	xt_load_indices(XTThreadPtr self, struct XTOpenTable *ot);
+void	xt_ind_count_deleted_items(struct XTTable *ot, struct XTIndex *ind, struct XTIndBlock *block);
 xtBool	xt_flush_indices(struct XTOpenTable *ot, off_t *bytes_flushed, xtBool have_table_lock);
 void	xt_ind_track_dump_block(struct XTTable *tab, xtIndexNodeID address);
 
@@ -482,6 +502,7 @@ void	xt_prev_branch_item_var(struct XTTable *tab, XTIndexPtr ind, XTIdxBranchDPt
 
 void	xt_last_branch_item_fix(struct XTTable *tab, XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result);
 void	xt_last_branch_item_var(struct XTTable *tab, XTIndexPtr ind, XTIdxBranchDPtr branch, register XTIdxResultPtr result);
+xtBool	xt_idx_lazy_delete_on_leaf(XTIndexPtr ind, struct XTIndBlock *block, xtWord2 branch_size);
 
 //#define TRACK_ACTIVITY
 #ifdef TRACK_ACTIVITY
diff --git a/storage/pbxt/src/lock_xt.cc b/storage/pbxt/src/lock_xt.cc
index ec698bb81b2..6a2d7a5b0e3 100644
--- a/storage/pbxt/src/lock_xt.cc
+++ b/storage/pbxt/src/lock_xt.cc
@@ -25,6 +25,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <stdio.h>
 
 #include "lock_xt.h"
@@ -40,6 +44,16 @@
 #endif
 
 /*
+ * This function should never be called. It indicates a link
+ * error!
+ */
+xtPublic void xt_log_atomic_error_and_abort(c_char *func, c_char *file, u_int line)
+{
+	xt_logf(NULL, func, file, line, XT_LOG_ERROR, "%s", "Atomic operations not supported\n");
+	abort();
+}
+
+/*
  * -----------------------------------------------------------------------
  * ROW LOCKS, LIST BASED
  */
@@ -715,7 +729,7 @@ xtBool xt_init_row_locks(XTRowLocksPtr rl)
 	return OK;
 }
 
-void xt_exit_row_locks(XTRowLocksPtr rl __attribute__((unused)))
+void xt_exit_row_locks(XTRowLocksPtr rl)
 {
 	for (int i=0; i<XT_ROW_LOCK_GROUP_COUNT; i++) {
 		xt_spinlock_free(NULL, &rl->rl_groups[i].lg_lock);
@@ -982,7 +996,7 @@ xtBool old_xt_init_row_locks(XTRowLocksPtr rl)
 	return OK;
 }
 
-void old_xt_exit_row_locks(XTRowLocksPtr rl __attribute__((unused)))
+void old_xt_exit_row_locks(XTRowLocksPtr XT_UNUSED(rl))
 {
 }
 
@@ -1007,10 +1021,6 @@ xtPublic void xt_exit_row_lock_list(XTRowLockListPtr lock_list)
  * SPECIAL EXCLUSIVE/SHARED (XS) LOCK
  */
 
-#define XT_GET1(x)		*(x)
-#define XT_SET4(x, y)	xt_atomic_set4(x, y)
-#define XT_GET4(x)		xt_atomic_get4(x)
-
 #ifdef XT_THREAD_LOCK_INFO
 xtPublic void xt_rwmutex_init(struct XTThread *self, XTRWMutexPtr xsl, const char *n)
 #else
@@ -1023,7 +1033,7 @@ xtPublic void xt_rwmutex_init(XTThreadPtr self, XTRWMutexPtr xsl)
 #endif
 	xt_init_mutex_with_autoname(self, &xsl->xs_lock);
 	xt_init_cond(self, &xsl->xs_cond);
-	XT_SET4(&xsl->xs_state, 0);
+	xt_atomic_set4(&xsl->xs_state, 0);
 	xsl->xs_xlocker = 0;
 	/* Must be aligned! */
 	ASSERT(xt_thr_maximum_threads == xt_align_size(xt_thr_maximum_threads, XT_XS_LOCK_ALIGN));
@@ -1068,7 +1078,7 @@ xtPublic xtBool xt_rwmutex_xlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 	}
 
 	/* I am the locker (set state before locker!): */
-	XT_SET4(&xsl->xs_state, 0);
+	xt_atomic_set4(&xsl->xs_state, 0);
 	xsl->xs_xlocker = thd_id;
 
 	/* Wait for all the read lockers: */
@@ -1078,7 +1088,7 @@ xtPublic xtBool xt_rwmutex_xlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 			 * Just in case of this, we keep the wait time down!
 			 */
 			if (!xt_timed_wait_cond_ns(&xsl->xs_cond, &xsl->xs_lock, 10)) {
-				XT_SET4(&xsl->xs_state, 0);
+				xt_atomic_set4(&xsl->xs_state, 0);
 				xsl->xs_xlocker = 0;
 				xt_unlock_mutex_ns(&xsl->xs_lock);
 				return FAILED;
@@ -1087,11 +1097,11 @@ xtPublic xtBool xt_rwmutex_xlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 		/* State can be incremented in parallel by a reader
 		 * thread!
 		 */
-		XT_SET4(&xsl->xs_state, xsl->xs_state + 1);
+		xt_atomic_set4(&xsl->xs_state, xsl->xs_state + 1);
 	}
 
 	/* I have waited for all: */
-	XT_SET4(&xsl->xs_state, xt_thr_maximum_threads);
+	xt_atomic_set4(&xsl->xs_state, xt_thr_maximum_threads);
 
 #ifdef XT_THREAD_LOCK_INFO
 	xt_thread_lock_info_add_owner(&xsl->xs_lock_info);
@@ -1107,7 +1117,7 @@ xtPublic xtBool xt_rwmutex_slock(XTRWMutexPtr xsl, xtThreadID thd_id)
 #endif
 	ASSERT_NS(xt_get_self()->t_id == thd_id);
 
-	xt_flushed_inc1(&xsl->x.xs_rlock[thd_id]);
+	xt_atomic_inc1(&xsl->x.xs_rlock[thd_id]);
 
 	if (xsl->x.xs_rlock[thd_id] > 1)
 		return OK;
@@ -1158,7 +1168,7 @@ xtPublic xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 		/* I have an X lock. */
 		ASSERT_NS(xsl->x.xs_rlock[thd_id] == XT_NO_LOCK);
 		ASSERT_NS(xsl->xs_state == xt_thr_maximum_threads);
-		XT_SET4(&xsl->xs_state, 0);
+		xt_atomic_set4(&xsl->xs_state, 0);
 		xsl->xs_xlocker = 0;
 		xt_unlock_mutex_ns(&xsl->xs_lock);
 		/* Wake up any other X or shared lockers: */
@@ -1201,7 +1211,7 @@ xtPublic xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 						return FAILED;
 					}
 				}
-				xt_flushed_dec1(&xsl->x.xs_rlock[thd_id]);
+				xt_atomic_dec1(&xsl->x.xs_rlock[thd_id]);
 				xt_unlock_mutex_ns(&xsl->xs_lock);
 			}
 			else
@@ -1213,7 +1223,7 @@ xtPublic xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id)
 				 * try to get the lock xs_lock, I could hand for the duration
 				 * of the X lock.
 				 */
-				xt_flushed_dec1(&xsl->x.xs_rlock[thd_id]);
+				xt_atomic_dec1(&xsl->x.xs_rlock[thd_id]);
 		}
 	}
 #ifdef XT_THREAD_LOCK_INFO
@@ -1228,13 +1238,14 @@ xtPublic xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id)
  */
 
 #ifdef XT_THREAD_LOCK_INFO
-xtPublic void xt_spinlock_init(XTThreadPtr self __attribute__((unused)), XTSpinLockPtr spl, const char *n)
+xtPublic void xt_spinlock_init(XTThreadPtr self, XTSpinLockPtr spl, const char *n)
 #else
-xtPublic void xt_spinlock_init(XTThreadPtr self __attribute__((unused)), XTSpinLockPtr spl)
+xtPublic void xt_spinlock_init(XTThreadPtr self, XTSpinLockPtr spl)
 #endif
 {
+	(void) self;
 	spl->spl_lock = 0;
-#ifdef XT_SPL_DEFAULT
+#ifdef XT_NO_ATOMICS
 	xt_init_mutex(self, &spl->spl_mutex);
 #endif
 #ifdef DEBUG
@@ -1246,9 +1257,10 @@ xtPublic void xt_spinlock_init(XTThreadPtr self __attribute__((unused)), XTSpinL
 #endif
 }
 
-xtPublic void xt_spinlock_free(XTThreadPtr self __attribute__((unused)), XTSpinLockPtr spl __attribute__((unused)))
+xtPublic void xt_spinlock_free(XTThreadPtr XT_UNUSED(self), XTSpinLockPtr spl)
 {
-#ifdef XT_SPL_DEFAULT
+	(void) spl;
+#ifdef XT_NO_ATOMICS
 	xt_free_mutex(&spl->spl_mutex);
 #endif
 #ifdef XT_THREAD_LOCK_INFO
@@ -1266,7 +1278,7 @@ xtPublic xtBool xt_spinlock_spin(XTSpinLockPtr spl)
 			if (!*lck) {
 				/* Try to get the lock: */
 				if (!xt_spinlock_set(spl))
-					return OK;
+					goto done_ok;
 			}
 		}
 
@@ -1274,6 +1286,7 @@ xtPublic xtBool xt_spinlock_spin(XTSpinLockPtr spl)
 		xt_critical_wait();
 	}
 
+	done_ok:
 	return OK;
 }
 
@@ -1400,147 +1413,96 @@ xtPublic void xt_fastlock_wakeup(XTFastLockPtr fal)
 /*
  * -----------------------------------------------------------------------
  * READ/WRITE SPIN LOCK
+ *
+ * An extremely genius very fast read/write lock based on atomics!
  */
 
 #ifdef XT_THREAD_LOCK_INFO
-xtPublic void xt_spinrwlock_init(struct XTThread *self, XTSpinRWLockPtr srw, const char *name)
+xtPublic void xt_spinxslock_init(struct XTThread *XT_UNUSED(self), XTSpinXSLockPtr sxs, const char *name)
 #else
-xtPublic void xt_spinrwlock_init(struct XTThread *self, XTSpinRWLockPtr srw)
+xtPublic void xt_spinxslock_init(struct XTThread *XT_UNUSED(self), XTSpinXSLockPtr sxs)
 #endif
 {
-	xt_spinlock_init_with_autoname(self, &srw->srw_lock);
-	xt_spinlock_init_with_autoname(self, &srw->srw_state_lock);
-	srw->srw_state = 0;
-	srw->srw_xlocker = 0;
-	/* Must be aligned! */
-	ASSERT(xt_thr_maximum_threads == xt_align_size(xt_thr_maximum_threads, XT_XS_LOCK_ALIGN));
-	srw->x.srw_rlock = (xtWord1 *) xt_calloc(self, xt_thr_maximum_threads);
+	sxs->sxs_xlocked = 0;
+	sxs->sxs_rlock_count = 0;
+	sxs->sxs_wait_count = 0;
+#ifdef DEBUG
+	sxs->sxs_locker = 0;
+#endif
 #ifdef XT_THREAD_LOCK_INFO
-	srw->srw_name = name;
-	xt_thread_lock_info_init(&srw->srw_lock_info, srw);
+	sxs->sxs_name = name;
+	xt_thread_lock_info_init(&sxs->sxs_lock_info, sxs);
 #endif
 }
 
-xtPublic void xt_spinrwlock_free(struct XTThread *self, XTSpinRWLockPtr srw)
+xtPublic void xt_spinxslock_free(struct XTThread *XT_UNUSED(self), XTSpinXSLockPtr sxs)
 {
-	if (srw->x.srw_rlock)
-		xt_free(self, (void *) srw->x.srw_rlock);
-	xt_spinlock_free(self, &srw->srw_lock);
-	xt_spinlock_free(self, &srw->srw_state_lock);
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_free(&srw->srw_lock_info);
+	xt_thread_lock_info_free(&sxs->sxs_lock_info);
+#else
+	(void) sxs;
 #endif
 }
 
-xtPublic xtBool xt_spinrwlock_xlock(XTSpinRWLockPtr srw, xtThreadID thd_id)
+xtPublic xtBool xt_spinxslock_xlock(XTSpinXSLockPtr sxs, xtThreadID XT_NDEBUG_UNUSED(thd_id))
 {
-	xt_spinlock_lock(&srw->srw_lock);
-	ASSERT_NS(srw->x.srw_rlock[thd_id] == XT_NO_LOCK);
-	
-	xt_spinlock_lock(&srw->srw_state_lock);
-
-	/* Set the state before xlocker (dirty read!) */
-	srw->srw_state = 0;
-
-	/* I am the locker: */
-	srw->srw_xlocker = thd_id;
+	register xtWord2 set;
 
-	/* Wait for all the read lockers: */
-	while (srw->srw_state < xt_thr_current_max_threads) {
-		while (srw->x.srw_rlock[srw->srw_state]) {
-			xt_spinlock_unlock(&srw->srw_state_lock);
-			/* Wait for this reader, during this time, the reader
-			 * himself, may increment the state. */
-			xt_critical_wait();
-			xt_spinlock_lock(&srw->srw_state_lock);
-		}
-		/* State can be incremented in parallel by a reader
-		 * thread!
-		 */
-		srw->srw_state++;
+	/* Wait for exclusive locker: */
+	for (;;) {
+		set = xt_atomic_tas2(&sxs->sxs_xlocked, 1);
+		if (!set)
+			break;
+		xt_yield();
 	}
 
-	/* I have waited for all: */
-	srw->srw_state = xt_thr_maximum_threads;
+#ifdef DEBUG
+	sxs->sxs_locker = thd_id;
+#endif
 
-	xt_spinlock_unlock(&srw->srw_state_lock);
+	/* Wait for all the reader to wait! */
+	while (sxs->sxs_wait_count < sxs->sxs_rlock_count)
+		xt_yield();
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_add_owner(&srw->srw_lock_info);
+	xt_thread_lock_info_add_owner(&sxs->sxs_lock_info);
 #endif
-
 	return OK;
 }
 
-xtPublic xtBool xt_spinrwlock_slock(XTSpinRWLockPtr srw, xtThreadID thd_id)
+xtPublic xtBool xt_spinxslock_slock(XTSpinXSLockPtr sxs)
 {
-	ASSERT_NS(srw->x.srw_rlock[thd_id] == XT_NO_LOCK);
-	srw->x.srw_rlock[thd_id] = XT_WANT_LOCK;
+	xt_atomic_inc2(&sxs->sxs_rlock_count);
+
 	/* Check if there could be an X locker: */
-	if (srw->srw_xlocker) {
-		/* There is an X locker.
-		 * If srw_state < thd_id then the X locker will wait for me.
-		 * So I should not wait!
-		 */
-		if (srw->srw_state >= thd_id) {
-			/* If srw->srw_state >= thd_id, then the locker may have, or
-			 * has already checked me, and I will have to wait.
-			 *
-			 * Otherwise, srw_state <= thd_id, which means the
-			 * X locker has not checked me, and will still wait for me (or 
-			 * is already waiting for me). In this case, I will have to
-			 * take the mutex to make sure exactly how far he
-			 * is with the checking.
-			 */
-			xt_spinlock_lock(&srw->srw_state_lock);
-			while (srw->srw_state > thd_id && srw->srw_xlocker) {
-				xt_spinlock_unlock(&srw->srw_state_lock);
-				xt_critical_wait();
-				xt_spinlock_lock(&srw->srw_state_lock);
-			}
-			xt_spinlock_unlock(&srw->srw_state_lock);
-		}
+	if (sxs->sxs_xlocked) {
+		/* I am waiting... */
+		xt_atomic_inc2(&sxs->sxs_wait_count);
+		while (sxs->sxs_xlocked)
+			xt_yield();
+		xt_atomic_dec2(&sxs->sxs_wait_count);
 	}
-	/* There is no exclusive locker, so we have the read lock: */
-	srw->x.srw_rlock[thd_id] = XT_HAVE_LOCK;
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_add_owner(&srw->srw_lock_info);
+	xt_thread_lock_info_add_owner(&sxs->sxs_lock_info);
 #endif
-
 	return OK;
 }
 
-xtPublic xtBool xt_spinrwlock_unlock(XTSpinRWLockPtr srw, xtThreadID thd_id)
+xtPublic xtBool xt_spinxslock_unlock(XTSpinXSLockPtr sxs, xtBool xlocked)
 {
-	if (srw->srw_xlocker == thd_id) {
-		/* I have an X lock. */
-		ASSERT_NS(srw->srw_state == xt_thr_maximum_threads);
-		srw->srw_state = 0;
-		srw->srw_xlocker = 0;
-		xt_spinlock_unlock(&srw->srw_lock);
-	}
-	else {
-		/* I have a shared lock: */
-		ASSERT_NS(srw->x.srw_rlock[thd_id] == XT_HAVE_LOCK);
-		ASSERT_NS(srw->srw_state != xt_thr_maximum_threads);
-		srw->x.srw_rlock[thd_id] = XT_NO_LOCK;
-		if (srw->srw_xlocker && srw->srw_state == thd_id) {
-			xt_spinlock_lock(&srw->srw_state_lock);
-			if (srw->srw_xlocker && srw->srw_state == thd_id) {
-				/* If the X locker is waiting for me,
-				 * then allow him to continue. 
-				 */
-				srw->srw_state = thd_id+1;
-			}
-			xt_spinlock_unlock(&srw->srw_state_lock);
-		}
+	if (xlocked) {
+#ifdef DEBUG
+		sxs->sxs_locker = 0;
+#endif
+		sxs->sxs_xlocked = 0;
 	}
+	else
+		xt_atomic_dec2(&sxs->sxs_rlock_count);
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_release_owner(&srw->srw_lock_info);
+	xt_thread_lock_info_release_owner(&sxs->sxs_lock_info);
 #endif
-
 	return OK;
 }
 
@@ -1550,194 +1512,159 @@ xtPublic xtBool xt_spinrwlock_unlock(XTSpinRWLockPtr srw, xtThreadID thd_id)
  */
 
 #ifdef XT_THREAD_LOCK_INFO
-xtPublic void xt_fastrwlock_init(struct XTThread *self, XTFastRWLockPtr frw, const char *n)
+xtPublic void xt_xsmutex_init(struct XTThread *self, XTXSMutexLockPtr xsm, const char *name)
 #else
-xtPublic void xt_fastrwlock_init(struct XTThread *self, XTFastRWLockPtr frw)
+xtPublic void xt_xsmutex_init(struct XTThread *self, XTXSMutexLockPtr xsm)
 #endif
 {
-	xt_fastlock_init_with_autoname(self, &frw->frw_lock);
-	frw->frw_xlocker = NULL;
-	xt_spinlock_init_with_autoname(self, &frw->frw_state_lock);
-	frw->frw_state = 0;
-	frw->frw_read_waiters = 0;
-	/* Must be aligned! */
-	ASSERT(xt_thr_maximum_threads == xt_align_size(xt_thr_maximum_threads, XT_XS_LOCK_ALIGN));
-	frw->x.frw_rlock = (xtWord1 *) xt_calloc(self, xt_thr_maximum_threads);
+	xt_init_mutex_with_autoname(self, &xsm->xsm_lock);
+	xt_init_cond(self, &xsm->xsm_cond);
+	xt_init_cond(self, &xsm->xsm_cond_2);
+	xsm->xsm_xlocker = 0;
+	xsm->xsm_rlock_count = 0;
+	xsm->xsm_wait_count = 0;
+#ifdef DEBUG
+	xsm->xsm_locker = 0;
+#endif
 #ifdef XT_THREAD_LOCK_INFO
-	frw->frw_name = n;
-	xt_thread_lock_info_init(&frw->frw_lock_info, frw);
+	xsm->xsm_name = name;
+	xt_thread_lock_info_init(&xsm->xsm_lock_info, xsm);
 #endif
 }
 
-xtPublic void xt_fastrwlock_free(struct XTThread *self, XTFastRWLockPtr frw)
+xtPublic void xt_xsmutex_free(struct XTThread *XT_UNUSED(self), XTXSMutexLockPtr xsm)
 {
-	if (frw->x.frw_rlock)
-		xt_free(self, (void *) frw->x.frw_rlock);
-	xt_fastlock_free(self, &frw->frw_lock);
-	xt_spinlock_free(self, &frw->frw_state_lock);
+	xt_free_mutex(&xsm->xsm_lock);
+	xt_free_cond(&xsm->xsm_cond);
+	xt_free_cond(&xsm->xsm_cond_2);
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_free(&frw->frw_lock_info);
+	xt_thread_lock_info_free(&xsm->xsm_lock_info);
 #endif
 }
 
-xtPublic xtBool xt_fastrwlock_xlock(XTFastRWLockPtr frw, struct XTThread *thread)
+xtPublic xtBool xt_xsmutex_xlock(XTXSMutexLockPtr xsm, xtThreadID thd_id)
 {
-	xt_fastlock_lock(&frw->frw_lock, thread);
-	ASSERT_NS(frw->x.frw_rlock[thread->t_id] == XT_NO_LOCK);
-	
-	xt_spinlock_lock(&frw->frw_state_lock);
-
-	/* Set the state before xlocker (dirty read!) */
-	frw->frw_state = 0;
+	xt_lock_mutex_ns(&xsm->xsm_lock);
 
-	/* I am the locker: */
-	frw->frw_xlocker = thread;
-
-	/* Wait for all the read lockers: */
-	while (frw->frw_state < xt_thr_current_max_threads) {
-		while (frw->x.frw_rlock[frw->frw_state]) {
-			xt_lock_thread(thread);
-			xt_spinlock_unlock(&frw->frw_state_lock);
-			/* Wait for this reader. We rely on the reader to free
-			 * us from this wait! */
-			if (!xt_wait_thread(thread)) {
-				xt_unlock_thread(thread);
-				frw->frw_state = 0;
-				frw->frw_xlocker = NULL;
-				xt_fastlock_unlock(&frw->frw_lock, thread);
-				return FAILED;
-			}
-			xt_unlock_thread(thread);
-			xt_spinlock_lock(&frw->frw_state_lock);
+	/* Wait for exclusive locker: */
+	while (xsm->xsm_xlocker) {
+		if (!xt_timed_wait_cond_ns(&xsm->xsm_cond, &xsm->xsm_lock, 10000)) {
+			xt_unlock_mutex_ns(&xsm->xsm_lock);
+			return FAILED;
 		}
-		/* State can be incremented in parallel by a reader
-		 * thread!
-		 */
-		frw->frw_state++;
 	}
 
-	/* I have waited for all: */
-	frw->frw_state = xt_thr_maximum_threads;
-
-	xt_spinlock_unlock(&frw->frw_state_lock);
+	/* GOTCHA: You would think this is not necessary...
+	 * But is does not always work, if a normal insert is used.
+	 * The reason is, I guess, on MMP the assignment is not
+	 * always immediately visible to other processors, because they
+	 * have old versions of this variable in there cache.
+	 *
+	 * But this is required, because the locking mechanism is based
+	 * on:
+	 * Locker: sets xlocker, tests rlock_count
+	 * Reader: incs rlock_count, tests xlocker
+	 *
+	 * The test, in both cases, may not read stale values.
+	 * volatile does not help, because this just turns compiler
+	 * optimisations off.
+	 */
+	xt_atomic_set4(&xsm->xsm_xlocker, thd_id);
+
+	/* Wait for all the reader to wait! */
+	while (xsm->xsm_wait_count < xsm->xsm_rlock_count) {
+		/* {RACE-WR_MUTEX} Here as well: */
+		if (!xt_timed_wait_cond_ns(&xsm->xsm_cond, &xsm->xsm_lock, 100)) {
+			xsm->xsm_xlocker = 0;
+			xt_unlock_mutex_ns(&xsm->xsm_lock);
+			return FAILED;
+		}
+	}
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_add_owner(&frw->frw_lock_info);
+	xt_thread_lock_info_add_owner(&xsm->xsm_lock_info);
 #endif
-
 	return OK;
 }
 
-xtPublic xtBool xt_fastrwlock_slock(XTFastRWLockPtr frw, struct XTThread *thread)
+xtPublic xtBool xt_xsmutex_slock(XTXSMutexLockPtr xsm, xtThreadID XT_UNUSED(thd_id))
 {
-	xtThreadID thd_id = thread->t_id;
+	xt_atomic_inc2(&xsm->xsm_rlock_count);
 
-	ASSERT_NS(frw->x.frw_rlock[thd_id] == XT_NO_LOCK);
-	frw->x.frw_rlock[thd_id] = XT_WANT_LOCK;
 	/* Check if there could be an X locker: */
-	if (frw->frw_xlocker) {
-		/* There is an X locker.
-		 * If frw_state < thd_id then the X locker will wait for me.
-		 * So I should not wait!
-		 */
-		if (frw->frw_state >= thd_id) {
-			/* If frw->frw_state >= thd_id, then the locker may have, or
-			 * has already checked me, and I will have to wait.
-			 *
-			 * Otherwise, frw_state <= thd_id, which means the
-			 * X locker has not checked me, and will still wait for me (or 
-			 * is already waiting for me). In this case, I will have to
-			 * take the mutex to make sure exactly how far he
-			 * is with the checking.
-			 */
-			xt_spinlock_lock(&frw->frw_state_lock);
-			frw->frw_read_waiters++;
-			frw->x.frw_rlock[thd_id] = XT_WAITING;
-			while (frw->frw_state > thd_id && frw->frw_xlocker) {
-				xt_lock_thread(thread);
-				xt_spinlock_unlock(&frw->frw_state_lock);
-				if (!xt_wait_thread(thread)) {
-					xt_unlock_thread(thread);
-					xt_spinlock_lock(&frw->frw_state_lock);
-					frw->frw_read_waiters--;
-					frw->x.frw_rlock[thd_id] = XT_NO_LOCK;
-					xt_spinlock_unlock(&frw->frw_state_lock);
-					return FAILED;
-				}
-				xt_unlock_thread(thread);
-				xt_spinlock_lock(&frw->frw_state_lock);
+	if (xsm->xsm_xlocker) {
+		/* I am waiting... */
+		xt_lock_mutex_ns(&xsm->xsm_lock);
+		xsm->xsm_wait_count++;
+		/* Wake up the xlocker: */
+		if (xsm->xsm_xlocker && xsm->xsm_wait_count == xsm->xsm_rlock_count) {
+			if (!xt_broadcast_cond_ns(&xsm->xsm_cond)) {
+				xsm->xsm_wait_count--;
+				xt_unlock_mutex_ns(&xsm->xsm_lock);
+				return FAILED;
 			}
-			frw->x.frw_rlock[thd_id] = XT_HAVE_LOCK;
-			frw->frw_read_waiters--;
-			xt_spinlock_unlock(&frw->frw_state_lock);
-			return OK;
 		}
+		while (xsm->xsm_xlocker) {
+			if (!xt_timed_wait_cond_ns(&xsm->xsm_cond_2, &xsm->xsm_lock, 10000)) {
+				xsm->xsm_wait_count--;
+				xt_unlock_mutex_ns(&xsm->xsm_lock);
+				return FAILED;
+			}
+		}
+		xsm->xsm_wait_count--;
+		xt_unlock_mutex_ns(&xsm->xsm_lock);
 	}
-	/* There is no exclusive locker, so we have the read lock: */
-	frw->x.frw_rlock[thd_id] = XT_HAVE_LOCK;
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_add_owner(&frw->frw_lock_info);
+	xt_thread_lock_info_add_owner(&xsm->xsm_lock_info);
 #endif
-
 	return OK;
 }
 
-xtPublic xtBool xt_fastrwlock_unlock(XTFastRWLockPtr frw, struct XTThread *thread)
+xtPublic xtBool xt_xsmutex_unlock(XTXSMutexLockPtr xsm, xtThreadID thd_id)
 {
-	xtThreadID thd_id = thread->t_id;
-
-	if (frw->frw_xlocker == thread) {
-		/* I have an X lock. */
-		ASSERT_NS(frw->frw_state == xt_thr_maximum_threads);
-		frw->frw_state = 0;
-		frw->frw_xlocker = NULL;
-
-		/* Wake up all read waiters: */
-		if (frw->frw_read_waiters) {
-			xt_spinlock_lock(&frw->frw_state_lock);
-			if (frw->frw_read_waiters) {
-				XTThreadPtr	target;
-
-				for (u_int i=0; i<xt_thr_current_max_threads; i++) {
-					if (frw->x.frw_rlock[i] == XT_WAITING) {
-						if ((target = xt_thr_array[i])) {
-							xt_lock_thread(target);
-							xt_signal_thread(target);
-							xt_unlock_thread(target);
-						}
-					}
-				}
+	if (xsm->xsm_xlocker == thd_id) {
+		xsm->xsm_xlocker = 0;
+		if (xsm->xsm_wait_count) {
+			if (!xt_broadcast_cond_ns(&xsm->xsm_cond_2)) {
+				xt_unlock_mutex_ns(&xsm->xsm_lock);
+				return FAILED;
 			}
-			xt_spinlock_unlock(&frw->frw_state_lock);
 		}
-		xt_fastlock_unlock(&frw->frw_lock, thread);
+		else {
+			/* Wake up any other X or shared lockers: */
+			if (!xt_broadcast_cond_ns(&xsm->xsm_cond)) {
+				xt_unlock_mutex_ns(&xsm->xsm_lock);
+				return FAILED;
+			}
+		}
+		xt_unlock_mutex_ns(&xsm->xsm_lock);
 	}
 	else {
-		/* I have a shared lock: */
-		ASSERT_NS(frw->x.frw_rlock[thd_id] == XT_HAVE_LOCK);
-		ASSERT_NS(frw->frw_state != xt_thr_maximum_threads);
-		frw->x.frw_rlock[thd_id] = XT_NO_LOCK;
-		if (frw->frw_xlocker && frw->frw_state == thd_id) {
-			xt_spinlock_lock(&frw->frw_state_lock);
-			if (frw->frw_xlocker && frw->frw_state == thd_id) {
+		/* Taking the advice from {RACE-WR_MUTEX} I do the decrement
+		 * after I have a lock!
+		 */
+		if (xsm->xsm_xlocker) {
+			xt_lock_mutex_ns(&xsm->xsm_lock);
+			xt_atomic_dec2(&xsm->xsm_rlock_count);
+			if (xsm->xsm_xlocker && xsm->xsm_wait_count == xsm->xsm_rlock_count) {
 				/* If the X locker is waiting for me,
 				 * then allow him to continue. 
 				 */
-				frw->frw_state = thd_id+1;
-				/* Wake him up: */
-				xt_lock_thread(frw->frw_xlocker);
-				xt_signal_thread(frw->frw_xlocker);
-				xt_unlock_thread(frw->frw_xlocker);
+				if (!xt_broadcast_cond_ns(&xsm->xsm_cond)) {
+					xt_unlock_mutex_ns(&xsm->xsm_lock);
+					return FAILED;
+				}
 			}
-			xt_spinlock_unlock(&frw->frw_state_lock);
+			xt_unlock_mutex_ns(&xsm->xsm_lock);
 		}
+		else
+			xt_atomic_dec2(&xsm->xsm_rlock_count);
 	}
 
 #ifdef XT_THREAD_LOCK_INFO
-	xt_thread_lock_info_release_owner(&frw->frw_lock_info);
+	xt_thread_lock_info_release_owner(&xsm->xsm_lock_info);
 #endif
-
 	return OK;
 }
 
@@ -1747,9 +1674,9 @@ xtPublic xtBool xt_fastrwlock_unlock(XTFastRWLockPtr frw, struct XTThread *threa
  */
 
 #ifdef XT_THREAD_LOCK_INFO
-xtPublic void xt_atomicrwlock_init(struct XTThread XT_UNUSED(*self), XTAtomicRWLockPtr arw, const char *n)
+xtPublic void xt_atomicrwlock_init(struct XTThread *XT_UNUSED(self), XTAtomicRWLockPtr arw, const char *n)
 #else
-xtPublic void xt_atomicrwlock_init(struct XTThread XT_UNUSED(*self), XTAtomicRWLockPtr arw)
+xtPublic void xt_atomicrwlock_init(struct XTThread *XT_UNUSED(self), XTAtomicRWLockPtr arw)
 #endif
 {
 	arw->arw_reader_count = 0;
@@ -1760,14 +1687,18 @@ xtPublic void xt_atomicrwlock_init(struct XTThread XT_UNUSED(*self), XTAtomicRWL
 #endif
 }
 
+#ifdef XT_THREAD_LOCK_INFO
+xtPublic void xt_atomicrwlock_free(struct XTThread *, XTAtomicRWLockPtr arw)
+#else
 xtPublic void xt_atomicrwlock_free(struct XTThread *, XTAtomicRWLockPtr XT_UNUSED(arw))
+#endif
 {
 #ifdef XT_THREAD_LOCK_INFO
 	xt_thread_lock_info_free(&arw->arw_lock_info);
 #endif
 }
 
-xtPublic xtBool xt_atomicrwlock_xlock(XTAtomicRWLockPtr arw, xtThreadID XT_UNUSED(thr_id))
+xtPublic xtBool xt_atomicrwlock_xlock(XTAtomicRWLockPtr arw, xtThreadID XT_NDEBUG_UNUSED(thr_id))
 {
 	register xtWord2 set;
 
@@ -1819,16 +1750,118 @@ xtPublic xtBool xt_atomicrwlock_slock(XTAtomicRWLockPtr arw)
 
 xtPublic xtBool xt_atomicrwlock_unlock(XTAtomicRWLockPtr arw, xtBool xlocked)
 {
-	if (xlocked)
+	if (xlocked) {
+#ifdef DEBUG
+		arw->arw_locker = 0;
+#endif
 		arw->arw_xlock_set = 0;
+	}
 	else
 		xt_atomic_dec2(&arw->arw_reader_count);
 
 #ifdef XT_THREAD_LOCK_INFO
 	xt_thread_lock_info_release_owner(&arw->arw_lock_info);
 #endif
+
+	return OK;
+}
+
+/*
+ * -----------------------------------------------------------------------
+ * "SKEW" ATOMITC READ/WRITE LOCK (BASED ON ATOMIC OPERATIONS)
+ *
+ * This lock type favors writers. It only works if the proportion of readers
+ * to writer is high.
+ */
+
+#ifdef XT_THREAD_LOCK_INFO
+xtPublic void xt_skewrwlock_init(struct XTThread *XT_UNUSED(self), XTSkewRWLockPtr srw, const char *n)
+#else
+xtPublic void xt_skewrwlock_init(struct XTThread *XT_UNUSED(self), XTSkewRWLockPtr srw)
+#endif
+{
+	srw->srw_reader_count = 0;
+	srw->srw_xlock_set = 0;
+#ifdef XT_THREAD_LOCK_INFO
+	srw->srw_name = n;
+	xt_thread_lock_info_init(&srw->srw_lock_info, srw);
+#endif
+}
+
+#ifdef XT_THREAD_LOCK_INFO
+xtPublic void xt_skewrwlock_free(struct XTThread *, XTSkewRWLockPtr srw)
+#else
+xtPublic void xt_skewrwlock_free(struct XTThread *, XTSkewRWLockPtr XT_UNUSED(srw))
+#endif
+{
+#ifdef XT_THREAD_LOCK_INFO
+	xt_thread_lock_info_free(&srw->srw_lock_info);
+#endif
+}
+
+xtPublic xtBool xt_skewrwlock_xlock(XTSkewRWLockPtr srw, xtThreadID XT_NDEBUG_UNUSED(thr_id))
+{
+	register xtWord2 set;
+
+	/* First get an exclusive lock: */
+	for (;;) {
+		set = xt_atomic_tas2(&srw->srw_xlock_set, 1);
+		if (!set)
+			break;
+		xt_yield();
+	}
+
+	/* Wait for the remaining readers: */
+	while (srw->srw_reader_count)
+		xt_yield();
+
+#ifdef DEBUG
+	srw->srw_locker = thr_id;
+#endif
+
+#ifdef XT_THREAD_LOCK_INFO
+	xt_thread_lock_info_add_owner(&srw->srw_lock_info);
+#endif
+	return OK;
+}
+
+xtPublic xtBool xt_skewrwlock_slock(XTSkewRWLockPtr srw)
+{
+	/* Wait for an exclusive lock: */
+	retry:
+	for (;;) {
+		if (!srw->srw_xlock_set)
+			break;
+		xt_yield();
+	}
+
+	/* Add a reader: */
+	xt_atomic_inc2(&srw->srw_reader_count);
+
+	/* Check for xlock again: */
+	if (srw->srw_xlock_set) {
+		xt_atomic_dec2(&srw->srw_reader_count);
+		goto retry;
+	}
+
+#ifdef XT_THREAD_LOCK_INFO
+	xt_thread_lock_info_add_owner(&srw->srw_lock_info);
+#endif
+	return OK;
+}
+
+xtPublic xtBool xt_skewrwlock_unlock(XTSkewRWLockPtr srw, xtBool xlocked)
+{
+	if (xlocked)
+		srw->srw_xlock_set = 0;
+	else
+		xt_atomic_dec2(&srw->srw_reader_count);
+
+#ifdef XT_THREAD_LOCK_INFO
+	xt_thread_lock_info_release_owner(&srw->srw_lock_info);
+#endif
 #ifdef DEBUG
-	arw->arw_locker = 0;
+	srw->srw_locker = 0;
 #endif
 
 	return OK;
@@ -1844,15 +1877,17 @@ xtPublic xtBool xt_atomicrwlock_unlock(XTAtomicRWLockPtr arw, xtBool xlocked)
 #define JOB_PRINT			3
 #define JOB_INCREMENT		4
 #define JOB_SNOOZE			5
+#define JOB_DOUBLE_INC		6
 
 #define LOCK_PTHREAD_RW		1
 #define LOCK_PTHREAD_MUTEX	2
-#define LOCK_FASTRW			3
+#define LOCK_RWMUTEX		3
 #define LOCK_SPINLOCK		4
 #define LOCK_FASTLOCK		5
-#define LOCK_SPINRWLOCK		6
-#define LOCK_FASTRWLOCK		7
+#define LOCK_SPINXSLOCK		6
+#define LOCK_XSMUTEX		7
 #define LOCK_ATOMICRWLOCK	8
+#define LOCK_SKEWRWLOCK		9
 
 typedef struct XSLockTest {
 	u_int			xs_interations;
@@ -1864,18 +1899,19 @@ typedef struct XSLockTest {
 	XTSpinLockRec	xs_spinlock;
 	xt_mutex_type	xs_mutex;
 	XTFastLockRec	xs_fastlock;
-	XTSpinRWLockRec	xs_spinrwlock;
-	XTFastRWLockRec	xs_fastrwlock;
+	XTSpinXSLockRec	xs_spinrwlock;
+	XTXSMutexRec	xs_fastrwlock;
 	XTAtomicRWLockRec xs_atomicrwlock;
+	XTSkewRWLockRec xs_skewrwlock;
 	int				xs_progress;
 	xtWord4			xs_inc;
 } XSLockTestRec, *XSLockTestPtr;
 
-static void lck_free_thread_data(XTThreadPtr self __attribute__((unused)), void *data __attribute__((unused)))
+static void lck_free_thread_data(XTThreadPtr XT_UNUSED(self), void *XT_UNUSED(data))
 {
 }
 
-static void lck_do_job(XTThreadPtr self, int job, XSLockTestPtr data)
+static void lck_do_job(XTThreadPtr self, int job, XSLockTestPtr data, xtBool reader)
 {
 	char b1[2048], b2[2048];
 
@@ -1900,6 +1936,16 @@ static void lck_do_job(XTThreadPtr self, int job, XSLockTestPtr data)
 			xt_sleep_milli_second(10);
 			data->xs_inc++;
 			break;
+		case JOB_DOUBLE_INC:
+			if (reader) {
+				if ((data->xs_inc & 1) != 0)
+					printf("Noooo!\n");
+			}
+			else {
+				data->xs_inc++;
+				data->xs_inc++;
+			}
+			break;
 	}
 }
 
@@ -1929,29 +1975,34 @@ static void *lck_run_reader(XTThreadPtr self)
 			printf("- %s %d\n", self->t_name, i+1);
 		if (data->xs_which_lock == LOCK_PTHREAD_RW) {
 			xt_slock_rwlock_ns(&data->xs_plock);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
 			xt_unlock_rwlock_ns(&data->xs_plock);
 		}
-		else if (data->xs_which_lock == LOCK_FASTRW) {
+		else if (data->xs_which_lock == LOCK_RWMUTEX) {
 			xt_rwmutex_slock(&data->xs_lock, self->t_id);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
 			xt_rwmutex_unlock(&data->xs_lock, self->t_id);
 		}
-		else if (data->xs_which_lock == LOCK_SPINRWLOCK) {
-			xt_spinrwlock_slock(&data->xs_spinrwlock, self->t_id);
-			lck_do_job(self, data->xs_which_job, data);
-			xt_spinrwlock_unlock(&data->xs_spinrwlock, self->t_id);
+		else if (data->xs_which_lock == LOCK_SPINXSLOCK) {
+			xt_spinxslock_slock(&data->xs_spinrwlock);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
+			xt_spinxslock_unlock(&data->xs_spinrwlock, FALSE);
 		}
-		else if (data->xs_which_lock == LOCK_FASTRWLOCK) {
-			xt_fastrwlock_slock(&data->xs_fastrwlock, self);
-			lck_do_job(self, data->xs_which_job, data);
-			xt_fastrwlock_unlock(&data->xs_fastrwlock, self);
+		else if (data->xs_which_lock == LOCK_XSMUTEX) {
+			xt_xsmutex_slock(&data->xs_fastrwlock, self->t_id);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
+			xt_xsmutex_unlock(&data->xs_fastrwlock, self->t_id);
 		}
 		else if (data->xs_which_lock == LOCK_ATOMICRWLOCK) {
 			xt_atomicrwlock_slock(&data->xs_atomicrwlock);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
 			xt_atomicrwlock_unlock(&data->xs_atomicrwlock, FALSE);
 		}
+		else if (data->xs_which_lock == LOCK_SKEWRWLOCK) {
+			xt_skewrwlock_slock(&data->xs_skewrwlock);
+			lck_do_job(self, data->xs_which_job, data, TRUE);
+			xt_skewrwlock_unlock(&data->xs_skewrwlock, FALSE);
+		}
 		else
 			ASSERT(FALSE);
 	}
@@ -1971,29 +2022,34 @@ static void *lck_run_writer(XTThreadPtr self)
 			printf("- %s %d\n", self->t_name, i+1);
 		if (data->xs_which_lock == LOCK_PTHREAD_RW) {
 			xt_xlock_rwlock_ns(&data->xs_plock);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_unlock_rwlock_ns(&data->xs_plock);
 		}
-		else if (data->xs_which_lock == LOCK_FASTRW) {
+		else if (data->xs_which_lock == LOCK_RWMUTEX) {
 			xt_rwmutex_xlock(&data->xs_lock, self->t_id);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_rwmutex_unlock(&data->xs_lock, self->t_id);
 		}
-		else if (data->xs_which_lock == LOCK_SPINRWLOCK) {
-			xt_spinrwlock_xlock(&data->xs_spinrwlock, self->t_id);
-			lck_do_job(self, data->xs_which_job, data);
-			xt_spinrwlock_unlock(&data->xs_spinrwlock, self->t_id);
+		else if (data->xs_which_lock == LOCK_SPINXSLOCK) {
+			xt_spinxslock_xlock(&data->xs_spinrwlock, self->t_id);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
+			xt_spinxslock_unlock(&data->xs_spinrwlock, TRUE);
 		}
-		else if (data->xs_which_lock == LOCK_FASTRWLOCK) {
-			xt_fastrwlock_xlock(&data->xs_fastrwlock, self);
-			lck_do_job(self, data->xs_which_job, data);
-			xt_fastrwlock_unlock(&data->xs_fastrwlock, self);
+		else if (data->xs_which_lock == LOCK_XSMUTEX) {
+			xt_xsmutex_xlock(&data->xs_fastrwlock, self->t_id);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
+			xt_xsmutex_unlock(&data->xs_fastrwlock, self->t_id);
 		}
 		else if (data->xs_which_lock == LOCK_ATOMICRWLOCK) {
 			xt_atomicrwlock_xlock(&data->xs_atomicrwlock, self->t_id);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_atomicrwlock_unlock(&data->xs_atomicrwlock, TRUE);
 		}
+		else if (data->xs_which_lock == LOCK_SKEWRWLOCK) {
+			xt_skewrwlock_xlock(&data->xs_skewrwlock, self->t_id);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
+			xt_skewrwlock_unlock(&data->xs_skewrwlock, TRUE);
+		}
 		else
 			ASSERT(FALSE);
 	}
@@ -2011,7 +2067,7 @@ static void lck_print_test(XSLockTestRec *data)
 		case LOCK_PTHREAD_MUTEX:
 			printf("pthread mutex");
 			break;
-		case LOCK_FASTRW:
+		case LOCK_RWMUTEX:
 			printf("fast read/write mutex");
 			break;
 		case LOCK_SPINLOCK:
@@ -2020,15 +2076,18 @@ static void lck_print_test(XSLockTestRec *data)
 		case LOCK_FASTLOCK:
 			printf("fast mutex");
 			break;
-		case LOCK_SPINRWLOCK:
+		case LOCK_SPINXSLOCK:
 			printf("spin read/write lock");
 			break;
-		case LOCK_FASTRWLOCK:
-			printf("fast read/write lock");
+		case LOCK_XSMUTEX:
+			printf("fast x/s mutex");
 			break;
 		case LOCK_ATOMICRWLOCK:
 			printf("atomic read/write lock");
 			break;
+		case LOCK_SKEWRWLOCK:
+			printf("skew read/write lock");
+			break;
 	}
 
 	switch (data->xs_which_job) {
@@ -2063,17 +2122,17 @@ static void *lck_run_mutex_locker(XTThreadPtr self)
 			printf("- %s %d\n", self->t_name, i+1);
 		if (data->xs_which_lock == LOCK_PTHREAD_MUTEX) {
 			xt_lock_mutex_ns(&data->xs_mutex);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_unlock_mutex_ns(&data->xs_mutex);
 		}
 		else if (data->xs_which_lock == LOCK_SPINLOCK) {
 			xt_spinlock_lock(&data->xs_spinlock);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_spinlock_unlock(&data->xs_spinlock);
 		}
 		else if (data->xs_which_lock == LOCK_FASTLOCK) {
 			xt_fastlock_lock(&data->xs_fastlock, self);
-			lck_do_job(self, data->xs_which_job, data);
+			lck_do_job(self, data->xs_which_job, data, FALSE);
 			xt_fastlock_unlock(&data->xs_fastlock, self);
 		}
 		else
@@ -2164,15 +2223,19 @@ xtPublic void xt_unit_test_read_write_locks(XTThreadPtr self)
 	memset(&data, 0, sizeof(data));
 
 	printf("TEST: xt_unit_test_read_write_locks\n");
+	printf("size of XTXSMutexRec = %d\n", (int) sizeof(XTXSMutexRec));
+	printf("size of pthread_cond_t = %d\n", (int) sizeof(pthread_cond_t));
+	printf("size of pthread_mutex_t = %d\n", (int) sizeof(pthread_mutex_t));
 	xt_rwmutex_init_with_autoname(self, &data.xs_lock);
 	xt_init_rwlock_with_autoname(self, &data.xs_plock);
-	xt_spinrwlock_init_with_autoname(self, &data.xs_spinrwlock);
-	xt_fastrwlock_init_with_autoname(self, &data.xs_fastrwlock);
+	xt_spinxslock_init_with_autoname(self, &data.xs_spinrwlock);
+	xt_xsmutex_init_with_autoname(self, &data.xs_fastrwlock);
 	xt_atomicrwlock_init_with_autoname(self, &data.xs_atomicrwlock);
+	xt_skewrwlock_init_with_autoname(self, &data.xs_skewrwlock);
 
 	/**
 	data.xs_interations = 10;
-	data.xs_which_lock = LOCK_FASTRW; // LOCK_PTHREAD_RW, LOCK_FASTRW, LOCK_SPINRWLOCK, LOCK_FASTRWLOCK
+	data.xs_which_lock = LOCK_RWMUTEX; // LOCK_PTHREAD_RW, LOCK_RWMUTEX, LOCK_SPINXSLOCK, LOCK_XSMUTEX
 	data.xs_which_job = JOB_PRINT;
 	data.xs_debug_print = TRUE;
 	data.xs_progress = 0;
@@ -2184,7 +2247,7 @@ xtPublic void xt_unit_test_read_write_locks(XTThreadPtr self)
 
 	/**
 	data.xs_interations = 4000;
-	data.xs_which_lock = LOCK_FASTRW; // LOCK_PTHREAD_RW, LOCK_FASTRW, LOCK_SPINRWLOCK, LOCK_FASTRWLOCK
+	data.xs_which_lock = LOCK_RWMUTEX; // LOCK_PTHREAD_RW, LOCK_RWMUTEX, LOCK_SPINXSLOCK, LOCK_XSMUTEX
 	data.xs_which_job = JOB_SLEEP;
 	data.xs_debug_print = TRUE;
 	data.xs_progress = 200;
@@ -2194,37 +2257,52 @@ xtPublic void xt_unit_test_read_write_locks(XTThreadPtr self)
 	lck_reader_writer_test(self, &data, 4, 2);
 	**/
 
+	// LOCK_PTHREAD_RW, LOCK_RWMUTEX, LOCK_SPINXSLOCK, LOCK_XSMUTEX, LOCK_ATOMICRWLOCK, LOCK_SKEWRWLOCK
 	/**/
-	data.xs_interations = 1000000;
-	data.xs_which_lock = LOCK_FASTRW; // LOCK_PTHREAD_RW, LOCK_FASTRW, LOCK_SPINRWLOCK, LOCK_FASTRWLOCK, LOCK_ATOMICRWLOCK
-	data.xs_which_job = JOB_INCREMENT;
+	data.xs_interations = 100000;
+	data.xs_which_lock = LOCK_XSMUTEX;
+	data.xs_which_job = JOB_DOUBLE_INC; // JOB_INCREMENT, JOB_DOUBLE_INC
 	data.xs_debug_print = FALSE;
 	data.xs_progress = 0;
 	lck_reader_writer_test(self, &data, 10, 0);
+	data.xs_which_lock = LOCK_XSMUTEX;
+	lck_reader_writer_test(self, &data, 10, 0);
+	//lck_reader_writer_test(self, &data, 0, 5);
+	//lck_reader_writer_test(self, &data, 10, 0);
+	//lck_reader_writer_test(self, &data, 10, 5);
 	/**/
 
-	/**
+	/**/
 	data.xs_interations = 10000;
-	data.xs_which_lock = LOCK_FASTRW; // LOCK_PTHREAD_RW, LOCK_FASTRW, LOCK_SPINRWLOCK, LOCK_FASTRWLOCK
+	data.xs_which_lock = LOCK_XSMUTEX;
 	data.xs_which_job = JOB_MEMCPY;
 	data.xs_debug_print = FALSE;
 	data.xs_progress = 0;
-	lck_reader_writer_test(self, &data, 10, 5);
-	**/
+	lck_reader_writer_test(self, &data, 10, 0);
+	data.xs_which_lock = LOCK_XSMUTEX;
+	lck_reader_writer_test(self, &data, 10, 0);
+	//lck_reader_writer_test(self, &data, 0, 5);
+	//lck_reader_writer_test(self, &data, 10, 0);
+	//lck_reader_writer_test(self, &data, 10, 5);
+	/**/
 
-	/**
+	/**/
 	data.xs_interations = 1000;
-	data.xs_which_lock = LOCK_FASTRW; // LOCK_PTHREAD_RW, LOCK_FASTRW, LOCK_SPINRWLOCK, LOCK_FASTRWLOCK
-	data.xs_which_job = JOB_SLEEP;
+	data.xs_which_lock = LOCK_XSMUTEX;
+	data.xs_which_job = JOB_SLEEP; // JOB_SLEEP, JOB_SNOOZE
 	data.xs_debug_print = FALSE;
 	data.xs_progress = 0;
-	lck_reader_writer_test(self, &data, 10, 5);
-	**/
+	lck_reader_writer_test(self, &data, 10, 0);
+	data.xs_which_lock = LOCK_XSMUTEX;
+	lck_reader_writer_test(self, &data, 10, 0);
+	/**/
 
 	xt_rwmutex_free(self, &data.xs_lock);
 	xt_free_rwlock(&data.xs_plock);
-	xt_spinrwlock_free(self, &data.xs_spinrwlock);
-	xt_fastrwlock_free(self, &data.xs_fastrwlock);
+	xt_spinxslock_free(self, &data.xs_spinrwlock);
+	xt_xsmutex_free(self, &data.xs_fastrwlock);
+	xt_atomicrwlock_free(self, &data.xs_atomicrwlock);
+	xt_skewrwlock_free(self, &data.xs_skewrwlock);
 }
 
 xtPublic void xt_unit_test_mutex_locks(XTThreadPtr self)
diff --git a/storage/pbxt/src/lock_xt.h b/storage/pbxt/src/lock_xt.h
index 214b00a849e..05ba9af244e 100644
--- a/storage/pbxt/src/lock_xt.h
+++ b/storage/pbxt/src/lock_xt.h
@@ -36,96 +36,16 @@ struct XTOpenTable;
 struct XTXactData;
 struct XTTable;
 
-/* Possibilities are 2 = align 4 or 2 = align 8 */
-#define XT_XS_LOCK_SHIFT		2
-#define XT_XS_LOCK_ALIGN		(1 << XT_XS_LOCK_SHIFT)
-
-/* This lock is fast for reads but slow for writes.
- * Use this lock in situations where you have 99% reads,
- * and then some potentially long writes.
- */
-typedef struct XTRWMutex {
-#ifdef DEBUG
-	struct XTThread				*xs_lock_thread;
-	u_int						xs_inited;
-#endif
-#ifdef XT_THREAD_LOCK_INFO
-	XTThreadLockInfoRec			xs_lock_info;
-	const char				    *xs_name;
-#endif
-	xt_mutex_type				xs_lock;
-	xt_cond_type				xs_cond;
-	volatile xtWord4			xs_state;
-	volatile xtThreadID			xs_xlocker;
-	union {
-#if XT_XS_LOCK_ALIGN == 4
-		volatile xtWord4		*xs_rlock_align;
-#else
-		volatile  xtWord8		*xs_rlock_align;
-#endif
-		volatile  xtWord1		*xs_rlock;
-	}							x;
-} XTRWMutexRec, *XTRWMutexPtr;
-
-#ifdef XT_THREAD_LOCK_INFO
-#define xt_rwmutex_init_with_autoname(a,b) xt_rwmutex_init(a,b,LOCKLIST_ARG_SUFFIX(b))
-void xt_rwmutex_init(struct XTThread *self, XTRWMutexPtr xsl, const char *name);
-#else
-#define xt_rwmutex_init_with_autoname(a,b) xt_rwmutex_init(a,b)
-void xt_rwmutex_init(struct XTThread *self, XTRWMutexPtr xsl);
-#endif
-void xt_rwmutex_free(struct XTThread *self, XTRWMutexPtr xsl);
-xtBool xt_rwmutex_xlock(XTRWMutexPtr xsl, xtThreadID thd_id);
-xtBool xt_rwmutex_slock(XTRWMutexPtr xsl, xtThreadID thd_id);
-xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id);
-
-#ifdef XT_WIN
-#define XT_SPL_WIN32_ASM
-#else
-#if defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
-#define XT_SPL_GNUC_X86
-#else
-#define XT_SPL_DEFAULT
-#endif
-#endif
-
-#ifdef XT_SOLARIS
-/* Use Sun atomic operations library
- * http://docs.sun.com/app/docs/doc/816-5168/atomic-ops-3c?a=view
- */
-#define XT_SPL_SOLARIS_LIB
-#endif
-
-#ifdef XT_SPL_SOLARIS_LIB
+#ifdef XT_ATOMIC_SOLARIS_LIB
 #include <atomic.h>
 #endif
 
-typedef struct XTSpinLock {
-	volatile xtWord4			spl_lock;
-#ifdef XT_SPL_DEFAULT
-	xt_mutex_type				spl_mutex;
-#endif
-#ifdef DEBUG
-	struct XTThread				*spl_locker;
-#endif
-#ifdef XT_THREAD_LOCK_INFO
-	XTThreadLockInfoRec			spl_lock_info;
-	const char				    *spl_name;
-#endif
-} XTSpinLockRec, *XTSpinLockPtr;
+void xt_log_atomic_error_and_abort(c_char *func, c_char *file, u_int line);
 
-#ifdef XT_THREAD_LOCK_INFO
-#define xt_spinlock_init_with_autoname(a,b) xt_spinlock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
-void	xt_spinlock_init(struct XTThread *self, XTSpinLockPtr sp, const char *name);
-#else
-#define xt_spinlock_init_with_autoname(a,b) xt_spinlock_init(a,b)
-void	xt_spinlock_init(struct XTThread *self, XTSpinLockPtr sp);
-#endif
-void	xt_spinlock_free(struct XTThread *self, XTSpinLockPtr sp);
-xtBool	xt_spinlock_spin(XTSpinLockPtr spl);
-#ifdef DEBUG
-void	xt_spinlock_set_thread(XTSpinLockPtr spl);
-#endif
+/*
+ * -----------------------------------------------------------------------
+ * ATOMIC OPERATIONS
+ */
 
 /*
  * This macro is to remind me where it was safe
@@ -137,37 +57,38 @@ void	xt_spinlock_set_thread(XTSpinLockPtr spl);
  * is written atomically.
  * But the operations themselves are not atomic!
  */
-inline void xt_flushed_inc1(volatile xtWord1 *mptr)
+inline void xt_atomic_inc1(volatile xtWord1 *mptr)
 {
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, mptr
 	__asm MOV  DL, BYTE PTR [ECX]
 	__asm INC  DL
 	__asm XCHG DL, BYTE PTR [ECX]
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	xtWord1 val;
 
 	asm volatile ("movb %1,%0" : "=r" (val) : "m" (*mptr) : "memory");
 	val++;
 	asm volatile ("xchgb %1,%0" : "=r" (val) : "m" (*mptr), "0" (val) : "memory");
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	atomic_inc_8(mptr);
 #else
 	*mptr++;
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 }
 
-inline xtWord1 xt_flushed_dec1(volatile xtWord1 *mptr)
+inline xtWord1 xt_atomic_dec1(volatile xtWord1 *mptr)
 {
 	xtWord1 val;
 
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, mptr
 	__asm MOV  DL, BYTE PTR [ECX]
 	__asm DEC  DL
 	__asm MOV  val, DL
 	__asm XCHG DL, BYTE PTR [ECX]
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	xtWord1 val2;
 
 	asm volatile ("movb %1, %0" : "=r" (val) : "m" (*mptr) : "memory");
@@ -176,55 +97,58 @@ inline xtWord1 xt_flushed_dec1(volatile xtWord1 *mptr)
 	/* Should work, but compiler makes a mistake?
 	 * asm volatile ("xchgb %1, %0" : : "r" (val), "m" (*mptr) : "memory");
 	 */
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	val = atomic_dec_8_nv(mptr);
 #else
 	val = --(*mptr);
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 	return val;
 }
 
 inline void xt_atomic_inc2(volatile xtWord2 *mptr)
 {
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm LOCK INC	WORD PTR mptr
-#elif defined(XT_SPL_GNUC_X86)
-        asm volatile ("lock; incw %0" : : "m" (*mptr) : "memory");
-#elif defined(__GNUC__)
+#elif defined(XT_ATOMIC_GNUC_X86)
+	asm volatile ("lock; incw %0" : : "m" (*mptr) : "memory");
+#elif defined(XT_ATOMIC_GCC_OPS)
 	__sync_fetch_and_add(mptr, 1);
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	atomic_inc_16_nv(mptr);
 #else
 	(*mptr)++;
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 }
 
 inline void xt_atomic_dec2(volatile xtWord2 *mptr)
 {
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm LOCK DEC	WORD PTR mptr
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	asm volatile ("lock; decw %0" : : "m" (*mptr) : "memory");
-#elif defined(__GNUC__)
+#elif defined(XT_ATOMIC_GCC_OPS)
 	__sync_fetch_and_sub(mptr, 1);
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	atomic_dec_16_nv(mptr);
 #else
 	--(*mptr);
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 }
 
 /* Atomic test and set 2 byte word! */
 inline xtWord2 xt_atomic_tas2(volatile xtWord2 *mptr, xtWord2 val)
 {
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, mptr
 	__asm MOV  DX, val
 	__asm XCHG DX, WORD PTR [ECX]
 	__asm MOV  val, DX
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	asm volatile ("xchgw %1,%0" : "=r" (val) : "m" (*mptr), "0" (val) : "memory");
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	val = atomic_swap_16(mptr, val);
 #else
 	/* Yikes! */
@@ -232,43 +156,80 @@ inline xtWord2 xt_atomic_tas2(volatile xtWord2 *mptr, xtWord2 val)
 
 	val = *mptr;
 	*mptr = nval;
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 	return val;
 }
 
 inline void xt_atomic_set4(volatile xtWord4 *mptr, xtWord4 val)
 {
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, mptr
 	__asm MOV  EDX, val
 	__asm XCHG EDX, DWORD PTR [ECX]
 	//__asm MOV  DWORD PTR [ECX], EDX
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	asm volatile ("xchgl %1,%0" : "=r" (val) : "m" (*mptr), "0" (val) : "memory");
 	//asm volatile ("movl %0,%1" : "=r" (val) : "m" (*mptr) : "memory");
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	atomic_swap_32(mptr, val);
 #else
 	*mptr = val;
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 }
 
-inline xtWord4 xt_atomic_get4(volatile xtWord4 *mptr)
-{
-	xtWord4 val;
-
-#ifdef XT_SPL_WIN32_ASM
-	__asm MOV ECX, mptr
-	__asm MOV EDX, DWORD PTR [ECX]
-	__asm MOV val, EDX
-#elif defined(XT_SPL_GNUC_X86)
-	asm volatile ("movl %1,%0" : "=r" (val) : "m" (*mptr) : "memory");
+inline xtWord4 xt_atomic_tas4(volatile xtWord4 *mptr, xtWord4 val)
+{				
+#ifdef XT_ATOMIC_WIN32_X86
+	__asm MOV  ECX, mptr
+	__asm MOV  EDX, val
+	__asm XCHG EDX, DWORD PTR [ECX]
+	__asm MOV  val, EDX
+#elif defined(XT_ATOMIC_GNUC_X86)
+	val = val;
+	asm volatile ("xchgl %1,%0" : "=r" (val) : "m" (*mptr), "0" (val) : "memory");
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
+	val = atomic_swap_32(mptr, val);
 #else
-	val = *mptr;
+	*mptr = val;
+	xt_log_atomic_error_and_abort(__FUNC__, __FILE__, __LINE__);
 #endif
 	return val;
 }
 
+/*
+ * -----------------------------------------------------------------------
+ * DIFFERENT TYPES OF LOCKS
+ */
+
+typedef struct XTSpinLock {
+	volatile xtWord4			spl_lock;
+#ifdef XT_NO_ATOMICS
+	xt_mutex_type				spl_mutex;
+#endif
+#ifdef DEBUG
+	struct XTThread				*spl_locker;
+#endif
+#ifdef XT_THREAD_LOCK_INFO
+	XTThreadLockInfoRec			spl_lock_info;
+	const char				    *spl_name;
+#endif
+} XTSpinLockRec, *XTSpinLockPtr;
+
+#ifdef XT_THREAD_LOCK_INFO
+#define xt_spinlock_init_with_autoname(a,b) xt_spinlock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
+void	xt_spinlock_init(struct XTThread *self, XTSpinLockPtr sp, const char *name);
+#else
+#define xt_spinlock_init_with_autoname(a,b) xt_spinlock_init(a,b)
+void	xt_spinlock_init(struct XTThread *self, XTSpinLockPtr sp);
+#endif
+void	xt_spinlock_free(struct XTThread *self, XTSpinLockPtr sp);
+xtBool	xt_spinlock_spin(XTSpinLockPtr spl);
+#ifdef DEBUG
+void	xt_spinlock_set_thread(XTSpinLockPtr spl);
+#endif
+
 /* Code for test and set is derived from code by Larry Zhou and
  * Google: http://code.google.com/p/google-perftools
  */
@@ -278,15 +239,15 @@ inline xtWord4 xt_spinlock_set(XTSpinLockPtr spl)
 	volatile xtWord4	*lck;
 				
 	lck = &spl->spl_lock;
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, lck
 	__asm MOV  EDX, 1
 	__asm XCHG EDX, DWORD PTR [ECX]
 	__asm MOV  prv, EDX
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	prv = 1;
 	asm volatile ("xchgl %1,%0" : "=r" (prv) : "m" (*lck), "0" (prv) : "memory");
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	prv = atomic_swap_32(lck, 1);
 #else
 	/* The default implementation just uses a mutex, and
@@ -312,15 +273,15 @@ inline xtWord4 xt_spinlock_reset(XTSpinLockPtr spl)
 	spl->spl_locker = NULL;
 #endif
 	lck = &spl->spl_lock;
-#ifdef XT_SPL_WIN32_ASM
+#ifdef XT_ATOMIC_WIN32_X86
 	__asm MOV  ECX, lck
 	__asm MOV  EDX, 0
 	__asm XCHG EDX, DWORD PTR [ECX]
 	__asm MOV  prv, EDX
-#elif defined(XT_SPL_GNUC_X86)
+#elif defined(XT_ATOMIC_GNUC_X86)
 	prv = 0;
 	asm volatile ("xchgl %1,%0" : "=r" (prv) : "m" (*lck), "0" (prv) : "memory");
-#elif defined(XT_SPL_SOLARIS_LIB)
+#elif defined(XT_ATOMIC_SOLARIS_LIB)
 	prv = atomic_swap_32(lck, 0);
 #else
 	*lck = 0;
@@ -359,9 +320,48 @@ inline void xt_spinlock_unlock(XTSpinLockPtr spl)
 #endif
 }
 
-void xt_unit_test_read_write_locks(struct XTThread *self);
-void xt_unit_test_mutex_locks(struct XTThread *self);
-void xt_unit_test_create_threads(struct XTThread *self);
+/* Possibilities are 2 = align 4 or 2 = align 8 */
+#define XT_XS_LOCK_SHIFT		2
+#define XT_XS_LOCK_ALIGN		(1 << XT_XS_LOCK_SHIFT)
+
+/* This lock is fast for reads but slow for writes.
+ * Use this lock in situations where you have 99% reads,
+ * and then some potentially long writes.
+ */
+typedef struct XTRWMutex {
+#ifdef DEBUG
+	struct XTThread				*xs_lock_thread;
+	u_int						xs_inited;
+#endif
+#ifdef XT_THREAD_LOCK_INFO
+	XTThreadLockInfoRec			xs_lock_info;
+	const char				    *xs_name;
+#endif
+	xt_mutex_type				xs_lock;
+	xt_cond_type				xs_cond;
+	volatile xtWord4			xs_state;
+	volatile xtThreadID			xs_xlocker;
+	union {
+#if XT_XS_LOCK_ALIGN == 4
+		volatile xtWord4		*xs_rlock_align;
+#else
+		volatile  xtWord8		*xs_rlock_align;
+#endif
+		volatile  xtWord1		*xs_rlock;
+	}							x;
+} XTRWMutexRec, *XTRWMutexPtr;
+
+#ifdef XT_THREAD_LOCK_INFO
+#define xt_rwmutex_init_with_autoname(a,b) xt_rwmutex_init(a,b,LOCKLIST_ARG_SUFFIX(b))
+void xt_rwmutex_init(struct XTThread *self, XTRWMutexPtr xsl, const char *name);
+#else
+#define xt_rwmutex_init_with_autoname(a,b) xt_rwmutex_init(a,b)
+void xt_rwmutex_init(struct XTThread *self, XTRWMutexPtr xsl);
+#endif
+void xt_rwmutex_free(struct XTThread *self, XTRWMutexPtr xsl);
+xtBool xt_rwmutex_xlock(XTRWMutexPtr xsl, xtThreadID thd_id);
+xtBool xt_rwmutex_slock(XTRWMutexPtr xsl, xtThreadID thd_id);
+xtBool xt_rwmutex_unlock(XTRWMutexPtr xsl, xtThreadID thd_id);
 
 #define XT_FAST_LOCK_MAX_WAIT	100
 
@@ -410,7 +410,7 @@ inline xtBool xt_fastlock_lock(XTFastLockPtr fal, struct XTThread *thread)
 #endif
 }
 
-inline void xt_fastlock_unlock(XTFastLockPtr fal, struct XTThread *thread __attribute__((unused)))
+inline void xt_fastlock_unlock(XTFastLockPtr fal, struct XTThread *XT_UNUSED(thread))
 {
 	if (fal->fal_wait_count)
 		xt_fastlock_wakeup(fal);
@@ -423,73 +423,61 @@ inline void xt_fastlock_unlock(XTFastLockPtr fal, struct XTThread *thread __attr
 #endif
 }
 
-typedef struct XTSpinRWLock {
-	XTSpinLockRec				srw_lock;
-	volatile xtThreadID			srw_xlocker;
-	XTSpinLockRec				srw_state_lock;
-	volatile u_int				srw_state;
-	union {
-#if XT_XS_LOCK_ALIGN == 4
-		volatile xtWord4		*srw_rlock_align;
-#else
-		volatile  xtWord8		*srw_rlock_align;
-#endif
-		volatile  xtWord1		*srw_rlock;
-	} x;
+#define XT_SXS_SLOCK_COUNT		2
 
+typedef struct XTSpinXSLock {
+	volatile xtWord2			sxs_xlocked;
+	volatile xtWord2			sxs_rlock_count;
+	volatile xtWord2			sxs_wait_count;			/* The number of readers waiting for the xlocker. */
+#ifdef DEBUG
+	xtThreadID					sxs_locker;
+#endif
 #ifdef XT_THREAD_LOCK_INFO
-	XTThreadLockInfoRec			srw_lock_info;
-	const char				    *srw_name;
+	XTThreadLockInfoRec			sxs_lock_info;
+	const char				    *sxs_name;
 #endif
-
-} XTSpinRWLockRec, *XTSpinRWLockPtr;
+} XTSpinXSLockRec, *XTSpinXSLockPtr;
 
 #ifdef XT_THREAD_LOCK_INFO
-#define xt_spinrwlock_init_with_autoname(a,b) xt_spinrwlock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
-void xt_spinrwlock_init(struct XTThread *self, XTSpinRWLockPtr xsl, const char *name);
-#else
-#define xt_spinrwlock_init_with_autoname(a,b) xt_spinrwlock_init(a,b)
-void xt_spinrwlock_init(struct XTThread *self, XTSpinRWLockPtr xsl);
-#endif
-void xt_spinrwlock_free(struct XTThread *self, XTSpinRWLockPtr xsl);
-xtBool xt_spinrwlock_xlock(XTSpinRWLockPtr xsl, xtThreadID thd_id);
-xtBool xt_spinrwlock_slock(XTSpinRWLockPtr xsl, xtThreadID thd_id);
-xtBool xt_spinrwlock_unlock(XTSpinRWLockPtr xsl, xtThreadID thd_id);
-
-typedef struct XTFastRWLock {
-	XTFastLockRec				frw_lock;
-	struct XTThread				*frw_xlocker;
-	XTSpinLockRec				frw_state_lock;
-	volatile u_int				frw_state;
-	u_int						frw_read_waiters;
-	union {
-#if XT_XS_LOCK_ALIGN == 4
-		volatile xtWord4		*frw_rlock_align;
+#define xt_spinxslock_init_with_autoname(a,b) xt_spinxslock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
+void xt_spinxslock_init(struct XTThread *self, XTSpinXSLockPtr sxs, const char *name);
 #else
-		volatile  xtWord8		*frw_rlock_align;
+#define xt_spinxslock_init_with_autoname(a,b) xt_spinxslock_init(a,b)
+void xt_spinxslock_init(struct XTThread *self, XTSpinXSLockPtr sxs);
+#endif
+void xt_spinxslock_free(struct XTThread *self, XTSpinXSLockPtr sxs);
+xtBool xt_spinxslock_xlock(XTSpinXSLockPtr sxs, xtThreadID thd_id);
+xtBool xt_spinxslock_slock(XTSpinXSLockPtr sxs);
+xtBool xt_spinxslock_unlock(XTSpinXSLockPtr sxs, xtBool xlocked);
+
+typedef struct XTXSMutexLock {
+	xt_mutex_type				xsm_lock;
+	xt_cond_type				xsm_cond;
+	xt_cond_type				xsm_cond_2;
+	volatile xtThreadID			xsm_xlocker;
+	volatile xtWord2			xsm_rlock_count;
+	volatile xtWord2			xsm_wait_count;			/* The number of readers waiting for the xlocker. */
+#ifdef DEBUG
+	xtThreadID					xsm_locker;
 #endif
-		volatile  xtWord1		*frw_rlock;
-	} x;
-
 #ifdef XT_THREAD_LOCK_INFO
-	XTThreadLockInfoRec			frw_lock_info;
-	const char				    *frw_name;
+	XTThreadLockInfoRec			xsm_lock_info;
+	const char				    *xsm_name;
 #endif
-
-} XTFastRWLockRec, *XTFastRWLockPtr;
+} XTXSMutexRec, *XTXSMutexLockPtr;
 
 #ifdef XT_THREAD_LOCK_INFO
-#define xt_fastrwlock_init_with_autoname(a,b) xt_fastrwlock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
-void xt_fastrwlock_init(struct XTThread *self, XTFastRWLockPtr frw, const char *name);
+#define xt_xsmutex_init_with_autoname(a,b) xt_xsmutex_init(a,b,LOCKLIST_ARG_SUFFIX(b))
+void xt_xsmutex_init(struct XTThread *self, XTXSMutexLockPtr xsm, const char *name);
 #else
-#define xt_fastrwlock_init_with_autoname(a,b) xt_fastrwlock_init(a,b)
-void xt_fastrwlock_init(struct XTThread *self, XTFastRWLockPtr frw);
+#define xt_xsmutex_init_with_autoname(a,b) xt_xsmutex_init(a,b)
+void xt_xsmutex_init(struct XTThread *self, XTXSMutexLockPtr xsm);
 #endif
 
-void xt_fastrwlock_free(struct XTThread *self, XTFastRWLockPtr frw);
-xtBool xt_fastrwlock_xlock(XTFastRWLockPtr frw, struct XTThread *thread);
-xtBool xt_fastrwlock_slock(XTFastRWLockPtr frw, struct XTThread *thread);
-xtBool xt_fastrwlock_unlock(XTFastRWLockPtr frw, struct XTThread *thread);
+void xt_xsmutex_free(struct XTThread *self, XTXSMutexLockPtr xsm);
+xtBool xt_xsmutex_xlock(XTXSMutexLockPtr xsm, xtThreadID thd_id);
+xtBool xt_xsmutex_slock(XTXSMutexLockPtr xsm, xtThreadID thd_id);
+xtBool xt_xsmutex_unlock(XTXSMutexLockPtr xsm, xtThreadID thd_id);
 
 typedef struct XTAtomicRWLock {
 	volatile xtWord2			arw_reader_count;
@@ -516,6 +504,35 @@ xtBool xt_atomicrwlock_xlock(XTAtomicRWLockPtr xsl, xtThreadID thr_id);
 xtBool xt_atomicrwlock_slock(XTAtomicRWLockPtr xsl);
 xtBool xt_atomicrwlock_unlock(XTAtomicRWLockPtr xsl, xtBool xlocked);
 
+typedef struct XTSkewRWLock {
+	volatile xtWord2			srw_reader_count;
+	volatile xtWord2			srw_xlock_set;
+
+#ifdef XT_THREAD_LOCK_INFO
+	XTThreadLockInfoRec			srw_lock_info;
+	const char				    *srw_name;
+#endif
+#ifdef DEBUG
+	xtThreadID					srw_locker;
+#endif
+} XTSkewRWLockRec, *XTSkewRWLockPtr;
+
+#ifdef XT_THREAD_LOCK_INFO
+#define xt_skewrwlock_init_with_autoname(a,b) xt_skewrwlock_init(a,b,LOCKLIST_ARG_SUFFIX(b))
+void xt_skewrwlock_init(struct XTThread *self, XTSkewRWLockPtr xsl, const char *name);
+#else
+#define xt_skewrwlock_init_with_autoname(a,b) xt_skewrwlock_init(a,b)
+void xt_skewrwlock_init(struct XTThread *self, XTSkewRWLockPtr xsl);
+#endif
+void xt_skewrwlock_free(struct XTThread *self, XTSkewRWLockPtr xsl);
+xtBool xt_skewrwlock_xlock(XTSkewRWLockPtr xsl, xtThreadID thr_id);
+xtBool xt_skewrwlock_slock(XTSkewRWLockPtr xsl);
+xtBool xt_skewrwlock_unlock(XTSkewRWLockPtr xsl, xtBool xlocked);
+
+void xt_unit_test_read_write_locks(struct XTThread *self);
+void xt_unit_test_mutex_locks(struct XTThread *self);
+void xt_unit_test_create_threads(struct XTThread *self);
+
 /*
  * -----------------------------------------------------------------------
  * ROW LOCKS
diff --git a/storage/pbxt/src/locklist_xt.cc b/storage/pbxt/src/locklist_xt.cc
index 0a3df584ba6..9a4aeb8f501 100644
--- a/storage/pbxt/src/locklist_xt.cc
+++ b/storage/pbxt/src/locklist_xt.cc
@@ -59,13 +59,13 @@ void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, xt_rwlock_struct *lock)
 	ptr->li_lock_type = XTThreadLockInfo::RW_LOCK;
 }
 
-void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTFastRWLock *lock)
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTXSMutexLock *lock)
 {
 	ptr->li_fast_rwlock = lock;
 	ptr->li_lock_type   = XTThreadLockInfo::FAST_RW_LOCK;
 }
 
-void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSpinRWLock *lock)
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSpinXSLock *lock)
 {
 	ptr->li_spin_rwlock = lock;
 	ptr->li_lock_type   = XTThreadLockInfo::SPIN_RW_LOCK;
@@ -77,6 +77,12 @@ void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTAtomicRWLock *lock)
 	ptr->li_lock_type   = XTThreadLockInfo::ATOMIC_RW_LOCK;
 }
 
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSkewRWLock *lock)
+{
+	ptr->li_skew_rwlock = lock;
+	ptr->li_lock_type   = XTThreadLockInfo::SKEW_RW_LOCK;
+}
+
 void xt_thread_lock_info_free(XTThreadLockInfoPtr ptr)
 {
 	/* TODO: check to see if it's present in a thread's list */
@@ -163,12 +169,12 @@ void xt_trace_thread_locks(XTThread *self)
 				lock_name = li->li_fast_lock->fal_name;
 				break;
 			case XTThreadLockInfo::FAST_RW_LOCK:
-				lock_type = "XTFastRWLock";
-				lock_name = li->li_fast_rwlock->frw_name;
+				lock_type = "XTXSMutexLock";
+				lock_name = li->li_fast_rwlock->xsm_name;
 				break;
 			case XTThreadLockInfo::SPIN_RW_LOCK:
 				lock_type = "XTSpinRWLock";
-				lock_name = li->li_spin_rwlock->srw_name;
+				lock_name = li->li_spin_rwlock->sxs_name;
 				break;
 			case XTThreadLockInfo::ATOMIC_RW_LOCK:
 				lock_type = "XTAtomicRWLock";
diff --git a/storage/pbxt/src/locklist_xt.h b/storage/pbxt/src/locklist_xt.h
index f0f16009ca1..4170e220ae1 100644
--- a/storage/pbxt/src/locklist_xt.h
+++ b/storage/pbxt/src/locklist_xt.h
@@ -25,7 +25,7 @@
 #define __xt_locklist_h__
 
 #ifdef DEBUG
-//#define XT_THREAD_LOCK_INFO
+#define XT_THREAD_LOCK_INFO
 #ifndef XT_WIN
 /* We need DEBUG_LOCKING in order to enable pthread function wrappers */
 #define DEBUG_LOCKING
@@ -40,9 +40,10 @@ struct XTRWMutex;
 struct xt_mutex_struct;
 struct xt_rwlock_struct;
 struct XTFastLock;
-struct XTFastRWLock;
-struct XTSpinRWLock;
+struct XTXSMutexLock;
+struct XTSpinXSLock;
 struct XTAtomicRWLock;
+struct XTSkewRWLock;
 
 #ifdef XT_THREAD_LOCK_INFO
 
@@ -61,7 +62,7 @@ struct XTAtomicRWLock;
  */
 typedef struct XTThreadLockInfo {
 
-	enum LockType { SPIN_LOCK, RW_MUTEX, MUTEX, RW_LOCK, FAST_LOCK, FAST_RW_LOCK, SPIN_RW_LOCK, ATOMIC_RW_LOCK };
+	enum LockType { SPIN_LOCK, RW_MUTEX, MUTEX, RW_LOCK, FAST_LOCK, FAST_RW_LOCK, SPIN_RW_LOCK, ATOMIC_RW_LOCK, SKEW_RW_LOCK };
 
 	LockType		  li_lock_type;
 
@@ -69,11 +70,12 @@ typedef struct XTThreadLockInfo {
 		XTSpinLock       *li_spin_lock;	  // SPIN_LOCK
 		XTRWMutex        *li_rw_mutex;	  // RW_MUTEX
 		XTFastLock		 *li_fast_lock;   // FAST_LOCK
-		XTFastRWLock	 *li_fast_rwlock; // FAST_RW_LOCK
-		XTSpinRWLock	 *li_spin_rwlock; // SPIN_RW_LOCK
+		XTXSMutexLock	 *li_fast_rwlock; // FAST_RW_LOCK
+		XTSpinXSLock	 *li_spin_rwlock; // SPIN_RW_LOCK
 		XTAtomicRWLock	 *li_atomic_rwlock; // ATOMIC_RW_LOCK
 		xt_mutex_struct  *li_mutex;		  // MUTEX
 		xt_rwlock_struct *li_rwlock;	  // RW_LOCK
+		XTSkewRWLock	 *li_skew_rwlock;	// SKEW_RW_LOCK
 	};
 } 
 XTThreadLockInfoRec, *XTThreadLockInfoPtr;
@@ -81,11 +83,12 @@ XTThreadLockInfoRec, *XTThreadLockInfoPtr;
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSpinLock *lock);
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTRWMutex *lock);
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTFastLock *lock);
-void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTFastRWLock *lock);
-void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSpinRWLock *lock);
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTXSMutexLock *lock);
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSpinXSLock *lock);
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTAtomicRWLock *lock);
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, xt_mutex_struct *lock);
 void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, xt_rwlock_struct *lock);
+void xt_thread_lock_info_init(XTThreadLockInfoPtr ptr, XTSkewRWLock *lock);
 void xt_thread_lock_info_free(XTThreadLockInfoPtr ptr);
 
 void xt_thread_lock_info_add_owner (XTThreadLockInfoPtr ptr);
diff --git a/storage/pbxt/src/memory_xt.cc b/storage/pbxt/src/memory_xt.cc
index 6da8673cfcc..6fb39396098 100644
--- a/storage/pbxt/src/memory_xt.cc
+++ b/storage/pbxt/src/memory_xt.cc
@@ -117,7 +117,7 @@ xtPublic xtBool	xt_realloc(XTThreadPtr self, void **ptr, size_t size)
 	return OK;
 }
 
-xtPublic void xt_free(XTThreadPtr self __attribute__((unused)), void *ptr)
+xtPublic void xt_free(XTThreadPtr XT_UNUSED(self), void *ptr)
 {
 	free(ptr);
 }
@@ -186,7 +186,7 @@ xtPublic void xt_free_ns(void *ptr)
 	free(ptr);
 }
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 
 /*
  * -----------------------------------------------------------------------
@@ -678,7 +678,7 @@ void xt_mm_memset(void *block, void *dest, int value, size_t size)
 	memset(dest, value, size);
 }
 
-void *xt_mm_malloc(XTThreadPtr self, size_t size, u_int line __attribute__((unused)), c_char *file __attribute__((unused)))
+void *xt_mm_malloc(XTThreadPtr self, size_t size, u_int line, c_char *file)
 {
 	unsigned char *p;
 
@@ -695,6 +695,8 @@ void *xt_mm_malloc(XTThreadPtr self, size_t size, u_int line __attribute__((unus
 	*(p + size + MEM_DEBUG_HDR_SIZE) = MEM_TRAILER_BYTE;
 	*(p + size + MEM_DEBUG_HDR_SIZE + 1L) = MEM_TRAILER_BYTE;
 
+	(void) line;
+	(void) file;
 #ifdef RECORD_MM
 	xt_lock_mutex(self, &mm_mutex);
 	mm_add_core_ptr(self, p + MEM_DEBUG_HDR_SIZE, 0, line, file);
@@ -704,7 +706,7 @@ void *xt_mm_malloc(XTThreadPtr self, size_t size, u_int line __attribute__((unus
 	return p + MEM_DEBUG_HDR_SIZE;
 }
 
-void *xt_mm_calloc(XTThreadPtr self, size_t size, u_int line __attribute__((unused)), c_char *file __attribute__((unused)))
+void *xt_mm_calloc(XTThreadPtr self, size_t size, u_int line, c_char *file)
 {
 	unsigned char *p;
 	
@@ -719,6 +721,8 @@ void *xt_mm_calloc(XTThreadPtr self, size_t size, u_int line __attribute__((unus
 	*(p + size + MEM_DEBUG_HDR_SIZE) = MEM_TRAILER_BYTE;
 	*(p + size + MEM_DEBUG_HDR_SIZE + 1L) = MEM_TRAILER_BYTE;
 
+	(void) line;
+	(void) file;
 #ifdef RECORD_MM
 	xt_lock_mutex(self, &mm_mutex);
 	mm_add_core_ptr(self, p + MEM_DEBUG_HDR_SIZE, 0, line, file);
@@ -849,7 +853,7 @@ void xt_mm_check_ptr(XTThreadPtr self, void *ptr)
 
 xtPublic xtBool xt_init_memory(void)
 {
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	XTThreadPtr	self = NULL;
 
 	if (!xt_init_mutex_with_autoname(NULL, &mm_mutex))
@@ -875,7 +879,7 @@ xtPublic void debug_ik_sum(void);
 
 xtPublic void xt_exit_memory(void)
 {
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	long	mm;
 	int		i;
 
@@ -919,7 +923,7 @@ xtPublic void xt_exit_memory(void)
  * MEMORY ALLOCATION UTILITIES
  */
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 char	*xt_mm_dup_string(XTThreadPtr self, c_char *str, u_int line, c_char *file)
 #else
 char	*xt_dup_string(XTThreadPtr self, c_char *str)
@@ -931,7 +935,7 @@ char	*xt_dup_string(XTThreadPtr self, c_char *str)
 	if (!str)
 		return NULL;
 	len = strlen(str);
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 	new_str = (char *) xt_mm_malloc(self, len + 1, line, file);
 #else
 	new_str = (char *) xt_malloc(self, len + 1);
@@ -1020,7 +1024,7 @@ xtPublic xtBool	xt_realloc(XTThreadPtr self, void **ptr, size_t size)
 	return *ptr != NULL;
 }
 
-xtPublic void xt_free(XTThreadPtr self __attribute__((unused)), void *ptr)
+xtPublic void xt_free(XTThreadPtr XT_UNUSED(self), void *ptr)
 {
 	char	*old_ptr;
 	xtWord4 size;
diff --git a/storage/pbxt/src/memory_xt.h b/storage/pbxt/src/memory_xt.h
index 3b4150df185..1785cd0bd51 100644
--- a/storage/pbxt/src/memory_xt.h
+++ b/storage/pbxt/src/memory_xt.h
@@ -30,6 +30,10 @@
 struct XTThread;
 
 #ifdef DEBUG
+#define DEBUG_MEMORY
+#endif
+
+#ifdef DEBUG_MEMORY
 
 #define XT_MM_STACK_TRACE	200
 #define XT_MM_TRACE_DEPTH	4
@@ -109,7 +113,7 @@ void	xt_free_ns(void *ptr);
 
 #endif
 
-#ifdef DEBUG
+#ifdef DEBUG_MEMORY
 #define xt_dup_string(t, s)		xt_mm_dup_string(t, s, __LINE__, __FILE__)
 
 char	*xt_mm_dup_string(struct XTThread *self, const char *path, u_int line, const char *file);
diff --git a/storage/pbxt/src/myxt_xt.cc b/storage/pbxt/src/myxt_xt.cc
index ba9a72c87a3..aa47f77a8ac 100644
--- a/storage/pbxt/src/myxt_xt.cc
+++ b/storage/pbxt/src/myxt_xt.cc
@@ -58,6 +58,9 @@ extern pthread_key_t THR_Session;
 #include "cache_xt.h"
 #include "datalog_xt.h"
 
+static void		myxt_bitmap_init(XTThreadPtr self, MX_BITMAP *map, u_int n_bits);
+static void		myxt_bitmap_free(XTThreadPtr self, MX_BITMAP *map);
+
 #ifdef DRIZZLED
 #define swap_variables(TYPE, a, b) \
   do {                             \
@@ -143,7 +146,7 @@ static void my_store_blob_length(byte *pos,uint pack_length,uint length)
 
 static int my_compare_text(MX_CONST_CHARSET_INFO *charset_info, uchar *a, uint a_length,
 				uchar *b, uint b_length, my_bool part_key,
-				my_bool skip_end_space __attribute__((unused)))
+				my_bool XT_UNUSED(skip_end_space))
 {
 	if (!part_key)
 		/* The last parameter is diff_if_only_endspace_difference, which means
@@ -632,7 +635,6 @@ static char *mx_get_length_and_data(Field *field, char *dest, xtWord4 *len)
 		case DRIZZLE_TYPE_DATE:
 		case DRIZZLE_TYPE_NEWDECIMAL:
 		case DRIZZLE_TYPE_ENUM:
-		case DRIZZLE_TYPE_VIRTUAL:
 #endif
 			break;
 	}
@@ -751,7 +753,6 @@ static void mx_set_length_and_data(Field *field, char *dest, xtWord4 len, char *
 		case DRIZZLE_TYPE_DATE:
 		case DRIZZLE_TYPE_NEWDECIMAL:
 		case DRIZZLE_TYPE_ENUM:
-		case DRIZZLE_TYPE_VIRTUAL:
 #endif
 			break;
 	}
@@ -764,7 +765,7 @@ static void mx_set_length_and_data(Field *field, char *dest, xtWord4 len, char *
 		bzero(from, field->pack_length());
 }
 
-xtPublic void myxt_set_null_row_from_key(XTOpenTablePtr ot __attribute__((unused)), XTIndexPtr ind, xtWord1 *record)
+xtPublic void myxt_set_null_row_from_key(XTOpenTablePtr XT_UNUSED(ot), XTIndexPtr ind, xtWord1 *record)
 {
 	register XTIndexSegRec *keyseg = ind->mi_seg;
 
@@ -800,7 +801,7 @@ xtPublic void myxt_set_default_row_from_key(XTOpenTablePtr ot, XTIndexPtr ind, x
 }
 
 /* Derived from _mi_put_key_in_record */
-xtPublic xtBool myxt_create_row_from_key(XTOpenTablePtr ot __attribute__((unused)), XTIndexPtr ind, xtWord1 *b_value, u_int key_len, xtWord1 *dest_buff)
+xtPublic xtBool myxt_create_row_from_key(XTOpenTablePtr XT_UNUSED(ot), XTIndexPtr ind, xtWord1 *b_value, u_int key_len, xtWord1 *dest_buff)
 {
 	byte					*record = (byte *) dest_buff;
 	register byte			*key;
@@ -935,8 +936,8 @@ xtPublic xtBool myxt_create_row_from_key(XTOpenTablePtr ot __attribute__((unused
 
 #ifdef CHECK_KEYS
 	err:
-#endif
 	return FAILED;				/* Crashed row */
+#endif
 }
 
 /*
@@ -1715,7 +1716,7 @@ xtPublic void myxt_get_column_as_string(XTOpenTablePtr ot, char *buffer, u_int c
 
 		/* Required by store() - or an assertion will fail: */
 		if (table->read_set)
-			bitmap_set_bit(table->read_set, col_idx);
+			MX_BIT_SET(table->read_set, col_idx);
 
 		save = field->ptr;
 		xt_lock_mutex(self, &tab->tab_dic_field_lock);
@@ -1743,7 +1744,7 @@ xtPublic xtBool myxt_set_column(XTOpenTablePtr ot, char *buffer, u_int col_idx,
 
 	/* Required by store() - or an assertion will fail: */
 	if (table->write_set)
-		bitmap_set_bit(table->write_set, col_idx);
+		MX_BIT_SET(table->write_set, col_idx);
 
 	mx_set_notnull_in_record(field, buffer);
 
@@ -1875,7 +1876,12 @@ xtPublic void myxt_print_key(XTIndexPtr ind, xtWord1 *key_value)
 
 static void my_close_table(TABLE *table)
 {
-#ifndef DRIZZLED
+#ifdef DRIZZLED
+	TABLE_SHARE	*share;
+
+	share = (TABLE_SHARE *) ((char *) table + sizeof(TABLE));
+	share->free_table_share();
+#else
 	closefrm(table, 1);  // TODO: Q, why did Stewart remove this?
 #endif
 	xt_free_ns(table);
@@ -1885,7 +1891,7 @@ static void my_close_table(TABLE *table)
  * This function returns NULL if the table cannot be opened 
  * because this is not a MySQL thread.
  */ 
-static TABLE *my_open_table(XTThreadPtr self, XTDatabaseHPtr db __attribute__((unused)), XTPathStrPtr tab_path)
+static TABLE *my_open_table(XTThreadPtr self, XTDatabaseHPtr XT_UNUSED(db), XTPathStrPtr tab_path)
 {
 	THD			*thd = current_thd;
 	char		path_buffer[PATH_MAX];
@@ -1946,6 +1952,18 @@ static TABLE *my_open_table(XTThreadPtr self, XTDatabaseHPtr db __attribute__((u
 	new_lex.current_select= NULL;
 	lex_start(thd);
 
+#ifdef DRIZZLED
+	share->init(db_name, 0, name, path);
+	if ((error = open_table_def(thd, share)) ||
+		(error = open_table_from_share(thd, share, "", 0, (uint32_t) READ_ALL, 0, table, OTM_OPEN)))
+	{
+		xt_free(self, table);
+		lex_end(&new_lex);
+		thd->lex = old_lex;
+		xt_throw_ulxterr(XT_CONTEXT, XT_ERR_LOADING_MYSQL_DIC, (u_long) error);
+		return NULL;
+	}
+#else
 #if MYSQL_VERSION_ID < 60000
 #if MYSQL_VERSION_ID < 50123
 	init_tmp_table_share(share, db_name, 0, name, path);
@@ -1980,6 +1998,7 @@ static TABLE *my_open_table(XTThreadPtr self, XTDatabaseHPtr db __attribute__((u
 		xt_throw_ulxterr(XT_CONTEXT, XT_ERR_LOADING_MYSQL_DIC, (u_long) error);
 		return NULL;
 	}
+#endif
 
 	lex_end(&new_lex);
 	thd->lex = old_lex;
@@ -1989,8 +2008,10 @@ static TABLE *my_open_table(XTThreadPtr self, XTDatabaseHPtr db __attribute__((u
 	 * plugin_shutdown() and reap_plugins() in sql_plugin.cc
 	 * from doing their job on shutdown!
 	 */
+#ifndef DRIZZLED
 	plugin_unlock(NULL, table->s->db_plugin);
 	table->s->db_plugin = NULL;
+#endif
 	return table;
 }
 
@@ -2069,6 +2090,11 @@ static xtBool my_is_not_null_int4(XTIndexSegPtr seg)
 	return (seg->type == HA_KEYTYPE_LONG_INT && !(seg->flag & HA_NULL_PART));
 }
 
+/* MY_BITMAP definition in Drizzle does not like if
+ * I use a NULL pointer to calculate the offset!?
+ */
+#define MX_OFFSETOF(x, y)		((size_t)(&((x *) 8)->y) - 8)
+
 /* Derived from ha_myisam::create and mi_create */
 static XTIndexPtr my_create_index(XTThreadPtr self, TABLE *table_arg, u_int idx, KEY *index)
 {
@@ -2084,7 +2110,7 @@ static XTIndexPtr my_create_index(XTThreadPtr self, TABLE *table_arg, u_int idx,
 
 	enter_();
 
-	pushsr_(ind, my_deref_index_data, (XTIndexPtr) xt_calloc(self, offsetof(XTIndexRec, mi_seg) + sizeof(XTIndexSegRec) * index->key_parts));
+	pushsr_(ind, my_deref_index_data, (XTIndexPtr) xt_calloc(self, MX_OFFSETOF(XTIndexRec, mi_seg) + sizeof(XTIndexSegRec) * index->key_parts));
 
 	XT_INDEX_INIT_LOCK(self, ind);
 	xt_init_mutex_with_autoname(self, &ind->mi_flush_lock);
@@ -2235,7 +2261,7 @@ static XTIndexPtr my_create_index(XTThreadPtr self, TABLE *table_arg, u_int idx,
 
 		/* NOTE: do not set if the field is only partially in the index!!! */
 		if (!partial_field)
-			bitmap_fast_test_and_set(&ind->mi_col_map, field->field_index);
+			MX_BIT_FAST_TEST_AND_SET(&ind->mi_col_map, field->field_index);
 	}
 
 	if (key_length > XT_INDEX_MAX_KEY_SIZE)
@@ -2243,6 +2269,7 @@ static XTIndexPtr my_create_index(XTThreadPtr self, TABLE *table_arg, u_int idx,
 
 	/* This is the maximum size of the index on disk: */
 	ind->mi_key_size = key_length;
+	ind->mi_max_items = (XT_INDEX_PAGE_SIZE-2) / (key_length+XT_RECORD_REF_SIZE);
 
 	if (ind->mi_fix_key) {
 		/* Special case for not-NULL 4 byte int value: */
@@ -2281,6 +2308,7 @@ static XTIndexPtr my_create_index(XTThreadPtr self, TABLE *table_arg, u_int idx,
 		ind->mi_prev_item = xt_prev_branch_item_var;
 		ind->mi_last_item = xt_last_branch_item_var;
 	}
+	ind->mi_lazy_delete = ind->mi_fix_key && ind->mi_max_items >= 4;
 
 	XT_NODE_ID(ind->mi_root) = 0;
 
@@ -2344,6 +2372,10 @@ xtPublic void myxt_setup_dictionary(XTThreadPtr self, XTDictionaryPtr dic)
 	KEY_PART_INFO	*key_part;
 	KEY_PART_INFO	*key_part_end;
 
+#ifndef XT_USE_LAZY_DELETE
+	dic->dic_no_lazy_delete = TRUE;
+#endif
+
 	dic->dic_ind_cols_req = 0;
 	for (uint i=0; i<TS(my_tab)->keys; i++) {
 		index = &my_tab->key_info[i];
@@ -2602,7 +2634,7 @@ xtPublic void myxt_setup_dictionary(XTThreadPtr self, XTDictionaryPtr dic)
 	dic->dic_mysql_rec_size = TS(my_tab)->reclength;
 }
 
-static u_int my_get_best_superset(XTThreadPtr self __attribute__((unused)), XTDictionaryPtr dic, XTIndexPtr ind)
+static u_int my_get_best_superset(XTThreadPtr XT_UNUSED(self), XTDictionaryPtr dic, XTIndexPtr ind)
 {
 	XTIndexPtr	super_ind;
 	u_int		super = 0;
@@ -2762,7 +2794,7 @@ static void ha_create_dd_index(XTThreadPtr self, XTDDIndex *ind, KEY *key)
 	}
 }
 
-static char *my_type_to_string(XTThreadPtr self, Field *field, TABLE *my_tab __attribute__((unused)))
+static char *my_type_to_string(XTThreadPtr self, Field *field, TABLE *XT_UNUSED(my_tab))
 {
 	char		buffer[MAX_FIELD_WIDTH + 400], *ptr;
 	String		type((char *) buffer, sizeof(buffer), system_charset_info);
@@ -2834,7 +2866,7 @@ xtPublic XTDDTable *myxt_create_table_from_table(XTThreadPtr self, TABLE *my_tab
  * MySQL CHARACTER UTILITIES
  */
 
-xtPublic void myxt_static_convert_identifier(XTThreadPtr self __attribute__((unused)), MX_CHARSET_INFO *cs, char *from, char *to, size_t to_len)
+xtPublic void myxt_static_convert_identifier(XTThreadPtr XT_UNUSED(self), MX_CHARSET_INFO *cs, char *from, char *to, size_t to_len)
 {
 	uint errors;
 
@@ -2877,7 +2909,7 @@ xtPublic char *myxt_convert_table_name(XTThreadPtr self, char *from)
 	return to;
 }
 
-xtPublic void myxt_static_convert_table_name(XTThreadPtr self __attribute__((unused)), char *from, char *to, size_t to_len)
+xtPublic void myxt_static_convert_table_name(XTThreadPtr XT_UNUSED(self), char *from, char *to, size_t to_len)
 {
 	tablename_to_filename(from, to, to_len);
 }
@@ -2948,12 +2980,12 @@ xtPublic xtBool myxt_use_blobs(XTOpenTablePtr ot, void **ret_pbms_table, xtWord1
 	return OK;
 }
 
-xtPublic void myxt_unuse_blobs(XTOpenTablePtr ot __attribute__((unused)), void *pbms_table)
+xtPublic void myxt_unuse_blobs(XTOpenTablePtr XT_UNUSED(ot), void *pbms_table)
 {
 	xt_pbms_close_table(pbms_table);
 }
 
-xtPublic xtBool myxt_retain_blobs(XTOpenTablePtr ot __attribute__((unused)), void *pbms_table, xtRecordID rec_id)
+xtPublic xtBool myxt_retain_blobs(XTOpenTablePtr XT_UNUSED(ot), void *pbms_table, xtRecordID rec_id)
 {
 	xtBool				ok;
 	PBMSEngineRefRec	eng_ref;
@@ -2997,6 +3029,9 @@ xtPublic void myxt_release_blobs(XTOpenTablePtr ot, xtWord1 *rec_buf, xtRecordID
 
 xtPublic void *myxt_create_thread()
 {
+#ifdef DRIZZLED
+	return (void *) 1;
+#else
 	THD *new_thd;
 
 	if (my_thread_init()) {
@@ -3015,8 +3050,14 @@ xtPublic void *myxt_create_thread()
 	lex_start(new_thd);
 
 	return (void *) new_thd;
+#endif
 }
 
+#ifdef DRIZZLED
+xtPublic void myxt_destroy_thread(void *, xtBool)
+{
+}
+#else
 xtPublic void myxt_destroy_thread(void *thread, xtBool end_threads)
 {
 	THD *thd = (THD *) thread;
@@ -3043,6 +3084,7 @@ xtPublic void myxt_destroy_thread(void *thread, xtBool end_threads)
 	if (end_threads)
 		my_thread_end();
 }
+#endif
 
 xtPublic XTThreadPtr myxt_get_self()
 {
@@ -3182,7 +3224,7 @@ xtPublic void myxt_get_status(XTThreadPtr self, XTStringBufferPtr strbuf)
  * MySQL Bit Maps
  */
 
-xtPublic void myxt_bitmap_init(XTThreadPtr self, MY_BITMAP *map, u_int n_bits)
+static void myxt_bitmap_init(XTThreadPtr self, MX_BITMAP *map, u_int n_bits)
 {
 	my_bitmap_map	*buf;
     uint			size_in_bytes = (((n_bits) + 31) / 32) * 4;
@@ -3194,7 +3236,7 @@ xtPublic void myxt_bitmap_init(XTThreadPtr self, MY_BITMAP *map, u_int n_bits)
 	bitmap_clear_all(map);
 }
 
-xtPublic void myxt_bitmap_free(XTThreadPtr self, MY_BITMAP *map)
+static void myxt_bitmap_free(XTThreadPtr self, MX_BITMAP *map)
 {
 	if (map->bitmap) {
 		xt_free(self, map->bitmap);
diff --git a/storage/pbxt/src/myxt_xt.h b/storage/pbxt/src/myxt_xt.h
index 4d33431088e..9f4d2d364e5 100644
--- a/storage/pbxt/src/myxt_xt.h
+++ b/storage/pbxt/src/myxt_xt.h
@@ -92,9 +92,6 @@ XTThreadPtr	myxt_get_self();
 int			myxt_statistics_fill_table(XTThreadPtr self, void *th, void *ta, void *co, MX_CONST void *ch);
 void		myxt_get_status(XTThreadPtr self, XTStringBufferPtr strbuf);
 
-void		myxt_bitmap_init(XTThreadPtr self, MY_BITMAP *map, u_int n_bits);
-void		myxt_bitmap_free(XTThreadPtr self, MY_BITMAP *map);
-
 class XTDDColumnFactory
 {
 public:
diff --git a/storage/pbxt/src/pbms.h b/storage/pbxt/src/pbms.h
index bf29c63651e..d0de34fa814 100644
--- a/storage/pbxt/src/pbms.h
+++ b/storage/pbxt/src/pbms.h
@@ -782,19 +782,20 @@ private:
 
 	void deleteTempFiles()
 	{
-		struct dirent *entry;
+		struct dirent	*entry;
 		struct dirent	*result;
 		DIR				*odir;
 		int				err;
-		size_t				sz;
+		size_t			sz;
 		char			temp_file[100];
 
-#ifdef XT_SOLARIS
+#ifdef __sun
 		sz = sizeof(struct dirent) + pathconf("/tmp/", _PC_NAME_MAX); // Solaris, see readdir(3C)
 #else
 		sz = sizeof(struct dirent);
 #endif
-		entry = (struct dirent*)malloc(sz);
+		if (!(entry = (struct dirent *) malloc(sz)))
+			return;
 		if (!(odir = opendir("/tmp/")))
 			return;
 		err = readdir_r(odir, entry, &result);
diff --git a/storage/pbxt/src/restart_xt.cc b/storage/pbxt/src/restart_xt.cc
index 50390bfb727..ca7e2c7e942 100644
--- a/storage/pbxt/src/restart_xt.cc
+++ b/storage/pbxt/src/restart_xt.cc
@@ -410,7 +410,7 @@ typedef struct XTOperation {
 	xtLogOffset				or_log_offset;
 } XTOperationRec, *XTOperationPtr;
 
-static int xres_cmp_op_seq(struct XTThread *self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int xres_cmp_op_seq(struct XTThread *XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtOpSeqNo		lf_op_seq = *((xtOpSeqNo *) a);
 	XTOperationPtr	lf_ptr = (XTOperationPtr) b;
@@ -480,19 +480,6 @@ static xtBool xres_add_index_entries(XTOpenTablePtr ot, xtRowID row_id, xtRecord
 		return OK;
 
 	for (idx_cnt=0, ind=tab->tab_dic.dic_keys; idx_cnt<tab->tab_dic.dic_key_count; idx_cnt++, ind++) {
-		/*
-		key.sk_on_key = FALSE;
-		key.sk_key_value.sv_flags = XT_SEARCH_WHOLE_KEY;
-		key.sk_key_value.sv_rec_id = rec_offset;
-		key.sk_key_value.sv_key = key.sk_key_buf;
-		key.sk_key_value.sv_length = myxt_create_key_from_row(*ind, key.sk_key_buf, rec_data, NULL);
-		if (!xt_idx_search(ot, *ind, &key)) {
-			ot->ot_err_index_no = (*ind)->mi_index_no;
-			return FAILED;
-		}
-		if (!key.sk_on_key) {
-		}
-		*/
 		if (!xt_idx_insert(ot, *ind, row_id, rec_id, rec_data, NULL, TRUE)) {
 			/* Check the error, certain errors are recoverable! */
 			XTThreadPtr self = xt_get_self();
@@ -651,6 +638,15 @@ static void xres_apply_change(XTThreadPtr self, XTOpenTablePtr ot, XTXactLogBuff
 		case XT_LOG_ENT_INSERT_BG:
 		case XT_LOG_ENT_DELETE_BG:
 			rec_id = XT_GET_DISK_4(record->xu.xu_rec_id_4);
+
+			/* This should be done before we apply change to table, as otherwise we lose
+			 * the key value that we need to remove from index
+			 */
+			if (check_index && ot->ot_table->tab_dic.dic_key_count && record->xl.xl_status_1 == XT_LOG_ENT_REC_MODIFIED) {
+				if ((rec_data = xres_load_record(self, ot, rec_id, NULL, 0, rec_buf, tab->tab_dic.dic_ind_cols_req)))
+					xres_remove_index_entries(ot, rec_id, rec_data);			
+			}
+
 			len = (size_t) XT_GET_DISK_2(record->xu.xu_size_2);
 			if (!XT_PWRITE_RR_FILE(ot->ot_rec_file, xt_rec_id_to_rec_offset(tab, rec_id), len, (xtWord1 *) &record->xu.xu_rec_type_1, &ot->ot_thread->st_statistics.st_rec, ot->ot_thread))
 				xt_throw(self);
@@ -661,10 +657,6 @@ static void xres_apply_change(XTThreadPtr self, XTOpenTablePtr ot, XTXactLogBuff
 					case XT_LOG_ENT_DELETE:
 					case XT_LOG_ENT_DELETE_BG:
 						break;
-					case XT_LOG_ENT_REC_MODIFIED:
-						if ((rec_data = xres_load_record(self, ot, rec_id, NULL, 0, rec_buf, tab->tab_dic.dic_ind_cols_req)))
-							xres_remove_index_entries(ot, rec_id, rec_data);
-						/* No break required: */
 					default:
 						if ((rec_data = xres_load_record(self, ot, rec_id, &record->xu.xu_rec_type_1, len, rec_buf, tab->tab_dic.dic_ind_cols_req))) {
 							row_id = XT_GET_DISK_4(record->xu.xu_row_id_4);
@@ -1560,7 +1552,7 @@ static xtBool xres_delete_data_log(XTDatabaseHPtr db, xtLogID log_id)
 	return OK;
 }
 
-static int xres_comp_flush_tabs(XTThreadPtr self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int xres_comp_flush_tabs(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtTableID				tab_id = *((xtTableID *) a);
 	XTCheckPointTablePtr	cp_tab = (XTCheckPointTablePtr) b;
@@ -1868,7 +1860,7 @@ void XTXactRestart::xres_init(XTThreadPtr self, XTDatabaseHPtr db, xtLogID *log_
 	exit_();
 }
 
-void XTXactRestart::xres_exit(XTThreadPtr self __attribute__((unused)))
+void XTXactRestart::xres_exit(XTThreadPtr XT_UNUSED(self))
 {
 }
 
@@ -2700,7 +2692,7 @@ xtPublic xtBool xt_begin_checkpoint(XTDatabaseHPtr db, xtBool have_table_lock, X
 		XTXactSegPtr 	seg;
 
 		seg = &db->db_xn_idx[i];
-		XT_XACT_WRITE_LOCK(&seg->xs_tab_lock, self);
+		XT_XACT_READ_LOCK(&seg->xs_tab_lock, self);
 		for (u_int j=0; j<XT_XN_HASH_TABLE_SIZE; j++) {
 			XTXactDataPtr	xact;
 			
@@ -2716,7 +2708,7 @@ xtPublic xtBool xt_begin_checkpoint(XTDatabaseHPtr db, xtBool have_table_lock, X
 				xact = xact->xd_next_xact;
 			}
 		}
-		XT_XACT_UNLOCK(&seg->xs_tab_lock, self);
+		XT_XACT_UNLOCK(&seg->xs_tab_lock, self, FALSE);
 	}
 
 #ifdef TRACE_CHECKPOINT
@@ -3201,3 +3193,36 @@ xtPublic void xt_dump_xlogs(XTDatabaseHPtr db, xtLogID start_log)
 	done:
 	db->db_xlog.xlog_seq_exit(&seq);
 }
+
+/* ----------------------------------------------------------------------
+ * D A T A B A S E   R E C O V E R Y   T H R E A D
+ */
+
+extern XTDatabaseHPtr pbxt_database;
+
+static void *xn_xres_run_recovery_thread(XTThreadPtr self)
+{
+	THD *mysql_thread;
+
+	mysql_thread = (THD *)myxt_create_thread();
+
+	while(!ha_resolve_by_legacy_type(mysql_thread, DB_TYPE_PBXT))
+		xt_sleep_milli_second(1);
+
+	xt_open_database(self, mysql_real_data_home, TRUE);
+	pbxt_database = self->st_database;
+	xt_heap_reference(self, pbxt_database);
+	myxt_destroy_thread(mysql_thread, TRUE);
+
+	return NULL;
+}
+
+xtPublic void xt_xres_start_database_recovery(XTThreadPtr self)
+{
+	char name[PATH_MAX];
+
+	sprintf(name, "DB-RECOVERY-%s", xt_last_directory_of_path(mysql_real_data_home));
+	xt_remove_dir_char(name);
+	XTThreadPtr thread = xt_create_daemon(self, name);
+	xt_run_thread(self, thread, xn_xres_run_recovery_thread);
+}
diff --git a/storage/pbxt/src/restart_xt.h b/storage/pbxt/src/restart_xt.h
index 259b3cbda90..af185223ccb 100644
--- a/storage/pbxt/src/restart_xt.h
+++ b/storage/pbxt/src/restart_xt.h
@@ -131,4 +131,6 @@ xtWord8	xt_bytes_since_last_checkpoint(struct XTDatabase *db, xtLogID curr_log_i
 void xt_print_log_record(xtLogID log, off_t offset, XTXactLogBufferDPtr record);
 void xt_dump_xlogs(struct XTDatabase *db, xtLogID start_log);
 
+xtPublic void xt_xres_start_database_recovery(XTThreadPtr self, const char *path);
+
 #endif
diff --git a/storage/pbxt/src/sortedlist_xt.cc b/storage/pbxt/src/sortedlist_xt.cc
index b4c525dbb22..f1742b64330 100644
--- a/storage/pbxt/src/sortedlist_xt.cc
+++ b/storage/pbxt/src/sortedlist_xt.cc
@@ -234,7 +234,7 @@ xtPublic void xt_sl_delete_item_at(struct XTThread *self, XTSortedListPtr sl, si
 	XT_MEMMOVE(sl->sl_data, &sl->sl_data[idx * sl->sl_item_size], &sl->sl_data[(idx+1) * sl->sl_item_size], (sl->sl_usage_count-idx) * sl->sl_item_size);
 }
 
-xtPublic void xt_sl_remove_from_front(struct XTThread *self __attribute__((unused)), XTSortedListPtr sl, size_t items)
+xtPublic void xt_sl_remove_from_front(struct XTThread *XT_UNUSED(self), XTSortedListPtr sl, size_t items)
 {
 	if (sl->sl_usage_count <= items)
 		xt_sl_set_size(sl, 0);
diff --git a/storage/pbxt/src/streaming_xt.cc b/storage/pbxt/src/streaming_xt.cc
index 710b3256d60..8ac8e25c607 100755
--- a/storage/pbxt/src/streaming_xt.cc
+++ b/storage/pbxt/src/streaming_xt.cc
@@ -274,7 +274,7 @@ static int pbxt_lock_table(void *thread, int *xact, void *open_table, int lock_t
 	return err;
 }
 
-static int pbxt_unlock_table(void *thread, int xact, void *open_table __attribute__((unused)), PBMSResultPtr result)
+static int pbxt_unlock_table(void *thread, int xact, void *XT_UNUSED(open_table), PBMSResultPtr result)
 {
 	THD				*thd = (THD *) thread;
 	XTThreadPtr		self = xt_ha_thd_to_self(thd);
@@ -418,6 +418,10 @@ static int pbxt_send_blob(void *thread, void *open_table, const char *blob_colum
 			err = MS_ERR_ENGINE;
 	}
 	cont_(a);
+	if (ot->ot_ind_rhandle) {
+		xt_ind_release_handle(ot->ot_ind_rhandle, FALSE, self);
+		ot->ot_ind_rhandle = NULL;
+	}
 	xt_sb_set_size(NULL, &value, 0);
 	return err;
 }
diff --git a/storage/pbxt/src/strutil_xt.cc b/storage/pbxt/src/strutil_xt.cc
index 60e45c455d1..a639d5cbf35 100644
--- a/storage/pbxt/src/strutil_xt.cc
+++ b/storage/pbxt/src/strutil_xt.cc
@@ -365,6 +365,7 @@ xtPublic void xt_int8_to_byte_size(xtInt8 value, char *string)
 	sprintf(string, "%s %s (%"PRId64" bytes)", val_str, unit, value);
 }
 
+/* Version number must also be set in configure.in! */
 xtPublic c_char *xt_get_version(void)
 {
 	return "1.0.08 RC";
diff --git a/storage/pbxt/src/systab_xt.cc b/storage/pbxt/src/systab_xt.cc
index 73ecc7a07cb..19d6c1d0ccb 100644
--- a/storage/pbxt/src/systab_xt.cc
+++ b/storage/pbxt/src/systab_xt.cc
@@ -217,7 +217,7 @@ bool XTLocationTable::seqScanNext(char *buf, bool *eof)
 	uint32		len;
 	Field		*curr_field;
 	byte		*save;
-	MY_BITMAP	*save_write_set;
+	MX_BITMAP	*save_write_set;
 
 	last_access = CS_GET_DISK_4(blob->rb_last_access_4);
 	last_ref = CS_GET_DISK_4(blob->rb_last_ref_4);
@@ -336,7 +336,6 @@ bool XTLocationTable::seqScanNext(char *buf, bool *eof)
 	table->write_set = save_write_set;
 	return true;
 #endif
-	return false;
 }
 
 void XTLocationTable::loadRow(char *buf, xtWord4 row_id)
@@ -345,7 +344,7 @@ void XTLocationTable::loadRow(char *buf, xtWord4 row_id)
 	Field			*curr_field;
 	XTTablePathPtr	tp_ptr;
 	byte			*save;
-	MY_BITMAP		*save_write_set;
+	MX_BITMAP		*save_write_set;
 
 	/* ASSERT_COLUMN_MARKED_FOR_WRITE is failing when
 	 * I use store()!??
@@ -386,7 +385,7 @@ void XTLocationTable::loadRow(char *buf, xtWord4 row_id)
 	table->write_set = save_write_set;
 }
 
-xtWord4 XTLocationTable::seqScanPos(xtWord1 *buf __attribute__((unused)))
+xtWord4 XTLocationTable::seqScanPos(xtWord1 *XT_UNUSED(buf))
 {
 	return lt_index-1;
 }
@@ -451,7 +450,7 @@ bool XTStatisticsTable::seqScanNext(char *buf, bool *eof)
 void XTStatisticsTable::loadRow(char *buf, xtWord4 rec_id)
 {
 	TABLE			*table = ost_my_table;
-	MY_BITMAP		*save_write_set;
+	MX_BITMAP		*save_write_set;
 	Field			*curr_field;
 	byte			*save;
 	const char		*stat_name;
@@ -503,7 +502,7 @@ void XTStatisticsTable::loadRow(char *buf, xtWord4 rec_id)
 	table->write_set = save_write_set;
 }
 
-xtWord4 XTStatisticsTable::seqScanPos(xtWord1 *buf __attribute__((unused)))
+xtWord4 XTStatisticsTable::seqScanPos(xtWord1 *XT_UNUSED(buf))
 {
 	return tt_index-1;
 }
@@ -531,14 +530,14 @@ void st_path_to_table_name(size_t size, char *buffer, const char *path)
 		*str = '.';
 }
 
-void XTSystemTableShare::startUp(XTThreadPtr self __attribute__((unused)))
+void XTSystemTableShare::startUp(XTThreadPtr XT_UNUSED(self))
 {
 	thr_lock_init(&sys_location_lock);
 	thr_lock_init(&sys_statistics_lock);
 	sys_lock_inited = TRUE;
 }
 
-void XTSystemTableShare::shutDown(XTThreadPtr self __attribute__((unused)))
+void XTSystemTableShare::shutDown(XTThreadPtr XT_UNUSED(self))
 {
 	if (sys_lock_inited) {
 		thr_lock_delete(&sys_location_lock);
@@ -588,7 +587,7 @@ bool XTSystemTableShare::doesSystemTableExist()
 	return false;
 }
 
-void XTSystemTableShare::createSystemTables(XTThreadPtr self __attribute__((unused)), XTDatabaseHPtr db __attribute__((unused)))
+void XTSystemTableShare::createSystemTables(XTThreadPtr XT_UNUSED(self), XTDatabaseHPtr XT_UNUSED(db))
 {
 	int		i = 0;
 
diff --git a/storage/pbxt/src/systab_xt.h b/storage/pbxt/src/systab_xt.h
index e64bcc816f4..408a8749dd0 100644
--- a/storage/pbxt/src/systab_xt.h
+++ b/storage/pbxt/src/systab_xt.h
@@ -85,15 +85,15 @@ public:
 	virtual bool use() { return true; }
 	virtual bool unuse() { return true; }
 	virtual bool seqScanInit() { return true; }
-	virtual bool seqScanNext(char *buf __attribute__((unused)), bool *eof) {
+	virtual bool seqScanNext(char *XT_UNUSED(buf), bool *eof) {
 		*eof = true;
 		return false;
 	}
 	virtual int	getRefLen() { return 4; }
-	virtual xtWord4 seqScanPos(xtWord1 *buf __attribute__((unused))) {
+	virtual xtWord4 seqScanPos(xtWord1 *XT_UNUSED(buf)) {
 		return 0;
 	}
-	virtual bool seqScanRead(xtWord4 rec_id __attribute__((unused)), char *buf __attribute__((unused))) {
+	virtual bool seqScanRead(xtWord4 XT_UNUSED(rec_id), char *XT_UNUSED(buf)) {
 		return true;
 	}
 
diff --git a/storage/pbxt/src/tabcache_xt.cc b/storage/pbxt/src/tabcache_xt.cc
index b9f9ccd37e1..cba89a39cc2 100644
--- a/storage/pbxt/src/tabcache_xt.cc
+++ b/storage/pbxt/src/tabcache_xt.cc
@@ -26,6 +26,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <signal.h>
 
 #include "pthread_xt.h"
@@ -63,7 +67,7 @@ xtPublic void xt_tc_init(XTThreadPtr self, size_t cache_size)
 		for (u_int i=0; i<XT_TC_SEGMENT_COUNT; i++) {
 			xt_tab_cache.tcm_segment[i].tcs_cache_in_use = 0;
 			xt_tab_cache.tcm_segment[i].tcs_hash_table = (XTTabCachePagePtr *) xt_calloc(self, xt_tab_cache.tcm_hash_size * sizeof(XTTabCachePagePtr));
-			xt_rwmutex_init_with_autoname(self, &xt_tab_cache.tcm_segment[i].tcs_lock);
+			TAB_CAC_INIT_LOCK(self, &xt_tab_cache.tcm_segment[i].tcs_lock);
 		}
 
 		xt_init_mutex_with_autoname(self, &xt_tab_cache.tcm_lock);
@@ -97,7 +101,7 @@ xtPublic void xt_tc_exit(XTThreadPtr self)
 
 			xt_free(self, xt_tab_cache.tcm_segment[i].tcs_hash_table);
 			xt_tab_cache.tcm_segment[i].tcs_hash_table = NULL;
-			xt_rwmutex_free(self, &xt_tab_cache.tcm_segment[i].tcs_lock);
+			TAB_CAC_FREE_LOCK(self, &xt_tab_cache.tcm_segment[i].tcs_lock);
 		}
 	}
 
@@ -213,7 +217,7 @@ xtBool XTTabCache::xt_tc_write(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, size_t
 	page->tcp_dirty = TRUE;
 	ASSERT_NS(page->tcp_db_id == tci_table->tab_db->db_id && page->tcp_tab_id == tci_table->tab_id);
 	*op_seq = tci_table->tab_seq.ts_set_op_seq(page);
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	return OK;
 }
 
@@ -269,21 +273,36 @@ xtBool XTTabCache::xt_tc_write_cond(XTThreadPtr self, XT_ROW_REC_FILE_PTR file,
 	page->tcp_dirty = TRUE;
 	ASSERT(page->tcp_db_id == tci_table->tab_db->db_id && page->tcp_tab_id == tci_table->tab_id);
 	*op_seq = tci_table->tab_seq.ts_set_op_seq(page);
-	xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 	return TRUE;
 
 	no_change:
-	xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 	return FALSE;
 }
 
 xtBool XTTabCache::xt_tc_read(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, size_t size, xtWord1 *data, XTThreadPtr thread)
 {
+#ifdef XT_USE_ROW_REC_MMAP_FILES
 	return tc_read_direct(file, ref_id, size, data, thread);
+#else
+	size_t				offset;
+	XTTabCachePagePtr	page;
+	XTTabCacheSegPtr	seg;
+
+	if (!tc_fetch(file, ref_id, &seg, &page, &offset, TRUE, thread))
+		return FAILED;
+	/* A read must be completely on a page: */
+	ASSERT_NS(offset + size <= tci_page_size);
+	memcpy(data, page->tcp_data + offset, size);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
+	return OK;
+#endif
 }
 
 xtBool XTTabCache::xt_tc_read_4(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord4 *value, XTThreadPtr thread)
 {
+#ifdef XT_USE_ROW_REC_MMAP_FILES
 	register u_int				page_idx;
 	register XTTabCachePagePtr	page;
 	register XTTabCacheSegPtr	seg;
@@ -300,7 +319,7 @@ xtBool XTTabCache::xt_tc_read_4(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord
 	seg = &dcg->tcm_segment[hash_idx & XT_TC_SEGMENT_MASK];
 	hash_idx = (hash_idx >> XT_TC_SEGMENT_SHIFTS) % dcg->tcm_hash_size;
 
-	xt_rwmutex_slock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_READ_LOCK(&seg->tcs_lock, thread->t_id);
 	page = seg->tcs_hash_table[hash_idx];
 	while (page) {
 		if (page->tcp_page_idx == page_idx && page->tcp_file_id == file->fr_id) {
@@ -311,53 +330,60 @@ xtBool XTTabCache::xt_tc_read_4(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord
 			ASSERT_NS(offset + 4 <= this->tci_page_size);
 			buffer = page->tcp_data + offset;
 			*value = XT_GET_DISK_4(buffer);
-			xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+			TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 			return OK;
 		}
 		page = page->tcp_next;
 	}
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 
-#ifdef XT_USE_ROW_REC_MMAP_FILES
 	return xt_pread_fmap_4(file, address, value, &thread->st_statistics.st_rec, thread);
 #else
-	xtWord1 data[4];
+	size_t				offset;
+	XTTabCachePagePtr	page;
+	XTTabCacheSegPtr	seg;
+	xtWord1				*data;
 
-	if (!XT_PREAD_RR_FILE(file, address, 4, 4, data, NULL, &thread->st_statistics.st_rec, thread))
+	if (!tc_fetch(file, ref_id, &seg, &page, &offset, TRUE, thread))
 		return FAILED;
+	/* A read must be completely on a page: */
+	ASSERT_NS(offset + 4 <= tci_page_size);
+	data = page->tcp_data + offset;
 	*value = XT_GET_DISK_4(data);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	return OK;
 #endif
 }
 
-xtBool XTTabCache::xt_tc_get_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCachePagePtr *ret_page, size_t *offset, XTThreadPtr thread)
+xtBool XTTabCache::xt_tc_get_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtBool load, XTTabCachePagePtr *ret_page, size_t *offset, XTThreadPtr thread)
 {
 	XTTabCachePagePtr	page;
 	XTTabCacheSegPtr	seg;
 
-#ifdef XT_SEQ_SCAN_FROM_MEMORY
-	if (!tc_fetch_direct(file, ref_id, &seg, &page, offset, thread))
-		return FAILED;
-	if (!seg) {
-		*ret_page = NULL;
-		return OK;
+	if (load) {
+		if (!tc_fetch(file, ref_id, &seg, &page, offset, TRUE, thread))
+			return FAILED;
+	}
+	else {
+		if (!tc_fetch_direct(file, ref_id, &seg, &page, offset, thread))
+			return FAILED;
+		if (!seg) {
+			*ret_page = NULL;
+			return OK;
+		}
 	}
-#else
-	if (!tc_fetch(file, ref_id, &seg, &page, offset, TRUE, thread))
-		return FAILED;
-#endif
 	page->tcp_lock_count++;
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	*ret_page = page;
 	return OK;
 }
 
-void XTTabCache::xt_tc_release_page(XT_ROW_REC_FILE_PTR file __attribute__((unused)), XTTabCachePagePtr page, XTThreadPtr thread)
+void XTTabCache::xt_tc_release_page(XT_ROW_REC_FILE_PTR XT_UNUSED(file), XTTabCachePagePtr page, XTThreadPtr thread)
 {
 	XTTabCacheSegPtr	seg;
 
 	seg = &xt_tab_cache.tcm_segment[page->tcp_seg];
-	xt_rwmutex_xlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_WRITE_LOCK(&seg->tcs_lock, thread->t_id);
 
 #ifdef DEBUG
 	XTTabCachePagePtr lpage, ppage;
@@ -379,7 +405,7 @@ void XTTabCache::xt_tc_release_page(XT_ROW_REC_FILE_PTR file __attribute__((unus
 	if (page->tcp_lock_count > 0)
 		page->tcp_lock_count--;
 
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 }
 
 xtBool XTTabCache::xt_tc_read_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord1 *data, XTThreadPtr thread)
@@ -412,7 +438,7 @@ xtBool XTTabCache::tc_read_direct(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, size
 	seg = &dcg->tcm_segment[hash_idx & XT_TC_SEGMENT_MASK];
 	hash_idx = (hash_idx >> XT_TC_SEGMENT_SHIFTS) % dcg->tcm_hash_size;
 
-	xt_rwmutex_slock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_READ_LOCK(&seg->tcs_lock, thread->t_id);
 	page = seg->tcs_hash_table[hash_idx];
 	while (page) {
 		if (page->tcp_page_idx == page_idx && page->tcp_file_id == file->fr_id) {
@@ -421,12 +447,12 @@ xtBool XTTabCache::tc_read_direct(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, size
 			offset = (ref_id % this->tci_rows_per_page) * this->tci_rec_size;
 			ASSERT_NS(offset + size <= this->tci_page_size);
 			memcpy(data, page->tcp_data + offset, size);
-			xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+			TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 			return OK;
 		}
 		page = page->tcp_next;
 	}
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	if (!XT_PREAD_RR_FILE(file, address, size, 0, data, &red_size, &thread->st_statistics.st_rec, thread))
 		return FAILED;
 	memset(data + red_size, 0, size - red_size);
@@ -450,7 +476,7 @@ xtBool XTTabCache::tc_fetch_direct(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTT
 	seg = &dcg->tcm_segment[hash_idx & XT_TC_SEGMENT_MASK];
 	hash_idx = (hash_idx >> XT_TC_SEGMENT_SHIFTS) % dcg->tcm_hash_size;
 
-	xt_rwmutex_xlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_WRITE_LOCK(&seg->tcs_lock, thread->t_id);
 	page = seg->tcs_hash_table[hash_idx];
 	while (page) {
 		if (page->tcp_page_idx == page_idx && page->tcp_file_id == file->fr_id) {
@@ -460,7 +486,7 @@ xtBool XTTabCache::tc_fetch_direct(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTT
 		}
 		page = page->tcp_next;
 	}
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	*ret_seg = NULL;
 	*ret_page = NULL;
 	return OK;
@@ -492,7 +518,7 @@ xtBool XTTabCache::tc_fetch(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCache
 	seg = &dcg->tcm_segment[hash_idx & XT_TC_SEGMENT_MASK];
 	hash_idx = (hash_idx >> XT_TC_SEGMENT_SHIFTS) % dcg->tcm_hash_size;
 
-	xt_rwmutex_slock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_READ_LOCK(&seg->tcs_lock, thread->t_id);
 	page = seg->tcs_hash_table[hash_idx];
 	while (page) {
 		if (page->tcp_page_idx == page_idx && page->tcp_file_id == file->fr_id) {
@@ -528,7 +554,7 @@ xtBool XTTabCache::tc_fetch(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCache
 		}
 		page = page->tcp_next;
 	}
-	xt_rwmutex_unlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, thread->t_id);
 	
 	/* Page not found, allocate a new page: */
 	size_t page_size = offsetof(XTTabCachePageRec, tcp_data) + this->tci_page_size;
@@ -674,7 +700,7 @@ xtBool XTTabCache::tc_fetch(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCache
 #endif
 
 	/* Add the page to the cache! */
-	xt_rwmutex_xlock(&seg->tcs_lock, thread->t_id);
+	TAB_CAC_WRITE_LOCK(&seg->tcs_lock, thread->t_id);
 	page = seg->tcs_hash_table[hash_idx];
 	while (page) {
 		if (page->tcp_page_idx == page_idx && page->tcp_file_id == file->fr_id) {
@@ -898,11 +924,11 @@ static size_t tabc_free_page(XTThreadPtr self, TCResourcePtr tc)
 	}
 
 	seg = &dcg->tcm_segment[page->tcp_seg];
-	xt_rwmutex_xlock(&seg->tcs_lock, self->t_id);
+	TAB_CAC_WRITE_LOCK(&seg->tcs_lock, self->t_id);
 
 	if (page->tcp_dirty) {
 		if (!was_dirty) {
-			xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+			TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 			goto retry_2;
 		}
 
@@ -923,7 +949,7 @@ static size_t tabc_free_page(XTThreadPtr self, TCResourcePtr tc)
 				XTDatabaseHPtr db = tab->tab_db;
 
 				rewait:
-				xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+				TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 
 				/* Flush the log, in case this is holding up the
 				 * writer!
@@ -963,7 +989,7 @@ static size_t tabc_free_page(XTThreadPtr self, TCResourcePtr tc)
 				db->db_wr_freeer_waiting = FALSE;
 				freer_(); // xt_unlock_mutex(&db->db_wr_lock)
 
-				xt_rwmutex_xlock(&seg->tcs_lock, self->t_id);
+				TAB_CAC_WRITE_LOCK(&seg->tcs_lock, self->t_id);
 				if (XTTableSeq::xt_op_is_before(tab->tab_head_op_seq, page->tcp_op_seq))
 					goto rewait;
 			}
@@ -988,11 +1014,11 @@ static size_t tabc_free_page(XTThreadPtr self, TCResourcePtr tc)
 			 */
 			if ((page = page->tcp_mr_used)) {
 				page_cnt++;
-				xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+				TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 				goto retry_2;
 			}
 		}
-		xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+		TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 		dcg->tcm_free_try_count++;				
 
 		/* Starting to spin, free the threads: */
@@ -1047,7 +1073,7 @@ static size_t tabc_free_page(XTThreadPtr self, TCResourcePtr tc)
 	seg->tcs_cache_in_use -= freed_space;
 	xt_free_ns(page);
 
-	xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+	TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 	self->st_statistics.st_rec_cache_frees++;
 	dcg->tcm_free_try_count = 0;
 	return freed_space;
@@ -1156,7 +1182,7 @@ static void *tabc_fr_run_thread(XTThreadPtr self)
 	return NULL;
 }
 
-static void tabc_fr_free_thread(XTThreadPtr self, void *data __attribute__((unused)))
+static void tabc_fr_free_thread(XTThreadPtr self, void *XT_UNUSED(data))
 {
 	if (xt_tab_cache.tcm_freeer_thread) {
 		xt_lock_mutex(self, &xt_tab_cache.tcm_freeer_lock);
@@ -1238,7 +1264,7 @@ xtPublic void xt_load_pages(XTThreadPtr self, XTOpenTablePtr ot)
 	while (rec_id<tab->tab_row_eof_id) {
 		if (!tab->tab_rows.tc_fetch(ot->ot_row_file, rec_id, &seg, &page, &poffset, TRUE, self))
 			xt_throw(self);
-		xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+		TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 		rec_id += tab->tab_rows.tci_rows_per_page;
 	}
 
@@ -1246,7 +1272,7 @@ xtPublic void xt_load_pages(XTThreadPtr self, XTOpenTablePtr ot)
 	while (rec_id<tab->tab_rec_eof_id) {
 		if (!tab->tab_recs.tc_fetch(ot->ot_rec_file, rec_id, &seg, &page, &poffset, TRUE, self))
 			xt_throw(self);
-		xt_rwmutex_unlock(&seg->tcs_lock, self->t_id);
+		TAB_CAC_UNLOCK(&seg->tcs_lock, self->t_id);
 		rec_id += tab->tab_recs.tci_rows_per_page;
 	}
 }
diff --git a/storage/pbxt/src/tabcache_xt.h b/storage/pbxt/src/tabcache_xt.h
index 694244835b4..46973037b49 100644
--- a/storage/pbxt/src/tabcache_xt.h
+++ b/storage/pbxt/src/tabcache_xt.h
@@ -125,11 +125,11 @@ typedef struct XTTableSeq {
 		xt_init_mutex_with_autoname(self, &ts_ns_lock);
 	}
 
-	void xt_op_seq_set(XTThreadPtr self __attribute__((unused)), xtOpSeqNo n) {
+	void xt_op_seq_set(XTThreadPtr XT_UNUSED(self), xtOpSeqNo n) {
 		ts_next_seq = n;
 	}
 
-	void xt_op_seq_exit(XTThreadPtr self __attribute__((unused))) {
+	void xt_op_seq_exit(XTThreadPtr XT_UNUSED(self)) {
 		xt_free_mutex(&ts_ns_lock);
 	}
 
@@ -150,12 +150,50 @@ typedef struct XTTableSeq {
 #endif
 } XTTableSeqRec, *XTTableSeqPtr;
 
+#ifdef XT_NO_ATOMICS
+#define TAB_CAC_USE_PTHREAD_RW
+#else
+//#define TAB_CAC_USE_RWMUTEX
+//#define TAB_CAC_USE_PTHREAD_RW
+//#define IDX_USE_SPINXSLOCK
+#define TAB_CAC_USE_XSMUTEX
+#endif
+
+#ifdef TAB_CAC_USE_XSMUTEX
+#define TAB_CAC_LOCK_TYPE				XTXSMutexRec
+#define TAB_CAC_INIT_LOCK(s, i)			xt_xsmutex_init_with_autoname(s, i)
+#define TAB_CAC_FREE_LOCK(s, i)			xt_xsmutex_free(s, i)	
+#define TAB_CAC_READ_LOCK(i, o)			xt_xsmutex_slock(i, o)
+#define TAB_CAC_WRITE_LOCK(i, o)		xt_xsmutex_xlock(i, o)
+#define TAB_CAC_UNLOCK(i, o)			xt_xsmutex_unlock(i, o)
+#elif defined(TAB_CAC_USE_PTHREAD_RW)
+#define TAB_CAC_LOCK_TYPE				xt_rwlock_type
+#define TAB_CAC_INIT_LOCK(s, i)			xt_init_rwlock(s, i)
+#define TAB_CAC_FREE_LOCK(s, i)			xt_free_rwlock(i)	
+#define TAB_CAC_READ_LOCK(i, o)			xt_slock_rwlock_ns(i)
+#define TAB_CAC_WRITE_LOCK(i, o)		xt_xlock_rwlock_ns(i)
+#define TAB_CAC_UNLOCK(i, o)			xt_unlock_rwlock_ns(i)
+#elif defined(TAB_CAC_USE_RWMUTEX)
+#define TAB_CAC_LOCK_TYPE				XTRWMutexRec
+#define TAB_CAC_INIT_LOCK(s, i)			xt_rwmutex_init_with_autoname(s, i)
+#define TAB_CAC_FREE_LOCK(s, i)			xt_rwmutex_free(s, i)	
+#define TAB_CAC_READ_LOCK(i, o)			xt_rwmutex_slock(i, o)
+#define TAB_CAC_WRITE_LOCK(i, o)		xt_rwmutex_xlock(i, o)
+#define TAB_CAC_UNLOCK(i, o)			xt_rwmutex_unlock(i, o)
+#elif defined(TAB_CAC_USE_SPINXSLOCK)
+#define TAB_CAC_LOCK_TYPE				XTSpinXSLockRec
+#define TAB_CAC_INIT_LOCK(s, i)			xt_spinxslock_init_with_autoname(s, i)
+#define TAB_CAC_FREE_LOCK(s, i)			xt_spinxslock_free(s, i)	
+#define TAB_CAC_READ_LOCK(i, o)			xt_spinxslock_slock(i, o)
+#define TAB_CAC_WRITE_LOCK(i, o)		xt_spinxslock_xlock(i, o)
+#define TAB_CAC_UNLOCK(i, o)			xt_spinxslock_unlock(i, o)
+#endif
+
 /* A disk cache segment. The cache is divided into a number of segments
  * to improve concurrency.
  */
 typedef struct XTTabCacheSeg {
-	XTRWMutexRec				tcs_lock;						/* The cache segment read/write lock. */
-	//xt_cond_type			tcs_cond;
+	TAB_CAC_LOCK_TYPE		tcs_lock;						/* The cache segment read/write lock. */
 	XTTabCachePagePtr		*tcs_hash_table;
 	size_t					tcs_cache_in_use;
 } XTTabCacheSegRec, *XTTabCacheSegPtr;
@@ -220,7 +258,7 @@ public:
 	xtBool					xt_tc_read(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, size_t size, xtWord1 *data, XTThreadPtr thread);
 	xtBool					xt_tc_read_4(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord4 *data, XTThreadPtr thread);
 	xtBool					xt_tc_read_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtWord1 *data, XTThreadPtr thread);
-	xtBool					xt_tc_get_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCachePagePtr *page, size_t *offset, XTThreadPtr thread);
+	xtBool					xt_tc_get_page(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, xtBool load, XTTabCachePagePtr *page, size_t *offset, XTThreadPtr thread);
 	void					xt_tc_release_page(XT_ROW_REC_FILE_PTR file, XTTabCachePagePtr page, XTThreadPtr thread);
 	xtBool					tc_fetch(XT_ROW_REC_FILE_PTR file, xtRefID ref_id, XTTabCacheSegPtr *ret_seg, XTTabCachePagePtr *ret_page, size_t *offset, xtBool read, XTThreadPtr thread);
 
diff --git a/storage/pbxt/src/table_xt.cc b/storage/pbxt/src/table_xt.cc
index fc9ae6156cb..5ab0bd45eb4 100644
--- a/storage/pbxt/src/table_xt.cc
+++ b/storage/pbxt/src/table_xt.cc
@@ -35,7 +35,7 @@
 #include <drizzled/common.h>
 #include <mysys/thr_lock.h>
 #include <drizzled/dtcollation.h>
-#include <drizzled/handlerton.h>
+#include <drizzled/plugin/storage_engine.h>
 #else
 #include "mysql_priv.h"
 #endif
@@ -293,17 +293,17 @@ static void tab_get_row_file_name(char *table_name, char *name, xtTableID tab_id
 	sprintf(table_name, "%s-%lu.xtr", name, (u_long) tab_id);
 }
 
-static void tab_get_data_file_name(char *table_name, char *name, xtTableID tab_id __attribute__((unused)))
+static void tab_get_data_file_name(char *table_name, char *name, xtTableID XT_UNUSED(tab_id))
 {
 	sprintf(table_name, "%s.xtd", name);
 }
 
-static void tab_get_index_file_name(char *table_name, char *name, xtTableID tab_id __attribute__((unused)))
+static void tab_get_index_file_name(char *table_name, char *name, xtTableID XT_UNUSED(tab_id))
 {
 	sprintf(table_name, "%s.xti", name);
 }
 
-static void tab_free_by_id(XTThreadPtr self __attribute__((unused)), void *thunk __attribute__((unused)), void *item)
+static void tab_free_by_id(XTThreadPtr self, void *XT_UNUSED(thunk), void *item)
 {
 	XTTableEntryPtr	te_ptr = (XTTableEntryPtr) item;
 
@@ -315,7 +315,7 @@ static void tab_free_by_id(XTThreadPtr self __attribute__((unused)), void *thunk
 	te_ptr->te_table = NULL;
 }
 
-static int tab_comp_by_id(XTThreadPtr self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int tab_comp_by_id(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtTableID		te_id = *((xtTableID *) a);
 	XTTableEntryPtr	te_ptr = (XTTableEntryPtr) b;
@@ -327,14 +327,14 @@ static int tab_comp_by_id(XTThreadPtr self __attribute__((unused)), register con
 	return 1;
 }
 
-static void tab_free_path(XTThreadPtr self __attribute__((unused)), void *thunk __attribute__((unused)), void *item)
+static void tab_free_path(XTThreadPtr self, void *XT_UNUSED(thunk), void *item)
 {
 	XTTablePathPtr	tp_ptr = *((XTTablePathPtr *) item);
 
 	xt_free(self, tp_ptr);
 }
 
-static int tab_comp_path(XTThreadPtr self __attribute__((unused)), register const void *thunk __attribute__((unused)), register const void *a, register const void *b)
+static int tab_comp_path(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	char			*path = (char *) a;
 	XTTablePathPtr	tp_ptr = *((XTTablePathPtr *) b);
@@ -398,7 +398,7 @@ xtPublic xtBool xt_describe_tables_next(XTThreadPtr self, XTTableDescPtr td)
 	return_(TRUE);
 }
 
-xtPublic void xt_describe_tables_exit(XTThreadPtr self __attribute__((unused)), XTTableDescPtr td)
+xtPublic void xt_describe_tables_exit(XTThreadPtr XT_UNUSED(self), XTTableDescPtr td)
 {
 	if (td->td_open_dir)
 		xt_dir_close(NULL, td->td_open_dir);
@@ -411,6 +411,7 @@ xtPublic void xt_tab_init_db(XTThreadPtr self, XTDatabaseHPtr db)
 	XTTableDescRec		desc;
 	XTTableEntryRec		te_tab;
 	XTTablePathPtr		db_path;
+	char			pbuf[PATH_MAX];
 	int					len;
 
 	enter_();
@@ -425,7 +426,6 @@ xtPublic void xt_tab_init_db(XTThreadPtr self, XTDatabaseHPtr db)
 	if (db->db_multi_path) {
 		XTOpenFilePtr	of;
 		char			*buffer, *ptr, *path;
-		char			pbuf[PATH_MAX];
 
 		xt_strcpy(PATH_MAX, pbuf, db->db_main_path);
 		xt_add_location_file(PATH_MAX, pbuf);
@@ -487,6 +487,11 @@ xtPublic void xt_tab_init_db(XTThreadPtr self, XTDatabaseHPtr db)
 		desc.td_tab_path->tp_tab_count++;
 		te_tab.te_table = NULL;
 		xt_sl_insert(self, db->db_table_by_id, &desc.td_tab_id, &te_tab);
+
+		xt_strcpy(PATH_MAX, pbuf, desc.td_tab_path->tp_path);
+		xt_add_dir_char(PATH_MAX, pbuf);
+		xt_strcat(PATH_MAX, pbuf, desc.td_tab_name);
+		xt_heap_release(self, xt_use_table_no_lock(self, db, (XTPathStrPtr)pbuf, FALSE, FALSE, NULL, NULL));
 	}
 	freer_(); // xt_describe_tables_exit(&desc)
 
@@ -605,8 +610,9 @@ xtPublic void xt_tab_exit_db(XTThreadPtr self, XTDatabaseHPtr db)
 	}
 }
 
-static void tab_check_table(XTThreadPtr self __attribute__((unused)), XTTableHPtr tab __attribute__((unused)))
+static void tab_check_table(XTThreadPtr self, XTTableHPtr XT_UNUSED(tab))
 {
+	(void) self;
 	enter_();
 	exit_();
 }
@@ -661,7 +667,7 @@ xtPublic void xt_enum_tables_init(u_int *edx)
 	*edx = 0;
 }
 
-xtPublic XTTableEntryPtr xt_enum_tables_next(XTThreadPtr self __attribute__((unused)), XTDatabaseHPtr db, u_int *edx)
+xtPublic XTTableEntryPtr xt_enum_tables_next(XTThreadPtr XT_UNUSED(self), XTDatabaseHPtr db, u_int *edx)
 {
 	XTTableEntryPtr en_ptr;
 
@@ -803,15 +809,32 @@ static void tab_load_index_header(XTThreadPtr self, XTTableHPtr tab, XTOpenFileP
 			tab->tab_index_page_size = XT_GET_DISK_4(index_fmt->if_page_size_4);
 		}	
 
+#ifdef XT_USE_LAZY_DELETE
+		if (tab->tab_dic.dic_index_ver <= XT_IND_NO_LAZY_DELETE)
+			tab->tab_dic.dic_no_lazy_delete = TRUE;
+		else
+			tab->tab_dic.dic_no_lazy_delete = FALSE;
+#else
+		tab->tab_dic.dic_no_lazy_delete = TRUE;
+#endif
+
 		/* Incorrect version of index is handled by allowing a sequential scan, but no index access.
 		 * Recovery with the wrong index type will not recover the indexes, a REPAIR TABLE
 		 * will be required!
 		 */
 		if (tab->tab_dic.dic_index_ver != XT_IND_CURRENT_VERSION) {
-			if (tab->tab_dic.dic_index_ver != XT_IND_CURRENT_VERSION)
-				tab->tab_dic.dic_disable_index = XT_INDEX_TOO_OLD;
-			else
-				tab->tab_dic.dic_disable_index = XT_INDEX_TOO_NEW;
+			switch (tab->tab_dic.dic_index_ver) {
+				case XT_IND_NO_LAZY_DELETE:
+				case XT_IND_LAZY_DELETE_OK:
+					/* I can handle this type of index. */
+					break;
+				default:
+					if (tab->tab_dic.dic_index_ver < XT_IND_CURRENT_VERSION)
+						tab->tab_dic.dic_disable_index = XT_INDEX_TOO_OLD;
+					else
+						tab->tab_dic.dic_disable_index = XT_INDEX_TOO_NEW;
+					break;
+			}
 		}
 		else if (tab->tab_index_page_size != XT_INDEX_PAGE_SIZE)
 			tab->tab_dic.dic_disable_index = XT_INDEX_BAD_BLOCK;
@@ -1572,7 +1595,7 @@ xtPublic void xt_create_table(XTThreadPtr self, XTPathStrPtr name, XTDictionaryP
 	exit_();
 }
 
-xtPublic void xt_drop_table(XTThreadPtr self, XTPathStrPtr tab_name)
+xtPublic void xt_drop_table(XTThreadPtr self, XTPathStrPtr tab_name, xtBool drop_db)
 {
 	XTDatabaseHPtr		db = self->st_database;
 	XTOpenTablePoolPtr	table_pool;
@@ -1596,8 +1619,16 @@ xtPublic void xt_drop_table(XTThreadPtr self, XTPathStrPtr tab_name)
 		tab_id = tab->tab_id;	/* tab is not null if returned table_pool is not null */
 		/* check if other tables refer this */
 		if (!self->st_ignore_fkeys) 
-			can_drop = tab->tab_dic.dic_table->checkCanDrop();
+			can_drop = tab->tab_dic.dic_table->checkCanDrop(drop_db);
+	}
+#ifdef DRIZZLED 
+	/* See the comment in ha_pbxt::delete_table regarding different implmentation of DROP TABLE
+         * in MySQL and Drizzle
+         */
+	else {
+		xt_throw_xterr(XT_CONTEXT, XT_ERR_TABLE_NOT_FOUND);
 	}
+#endif
 
 	if (can_drop) {
 		if (tab_id) {
@@ -2306,6 +2337,9 @@ xtPublic XTOpenTablePtr tab_open_table(XTTableHPtr tab)
 		return NULL;
 	memset(ot, 0, offsetof(XTOpenTableRec, ot_ind_wbuf));
 
+	ot->ot_seq_page = NULL;
+	ot->ot_seq_data = NULL;
+
 	self = xt_get_self();
 	try_(a) {
 		xt_heap_reference(self, tab);
@@ -3353,6 +3387,16 @@ xtPublic int xt_tab_dirty_read_record(register XTOpenTablePtr ot, xtWord1 *buffe
 	return OK;
 }
 
+#ifdef XT_USE_ROW_REC_MMAP_FILES
+/* Loading into cache is not required,
+ * Instead we copy the memory map to load the
+ * data.
+ */
+#define TAB_ROW_LOAD_CACHE		FALSE
+#else
+#define TAB_ROW_LOAD_CACHE		TRUE
+#endif
+
 /*
  * Pull the entire row pointer file into memory.
  */
@@ -3376,7 +3420,7 @@ xtPublic void xt_tab_load_row_pointers(XTThreadPtr self, XTOpenTablePtr ot)
 		end_offset = xt_row_id_to_row_offset(tab, eof_rec_id);
 		rec_id = 1;
 		while (rec_id < eof_rec_id) {
-			if (!tab->tab_rows.xt_tc_get_page(ot->ot_row_file, rec_id, &page, &poffset, self))
+			if (!tab->tab_rows.xt_tc_get_page(ot->ot_row_file, rec_id, TAB_ROW_LOAD_CACHE, &page, &poffset, self))
 				xt_throw(self);
 			if (page)
 				tab->tab_rows.xt_tc_release_page(ot->ot_row_file, page, self);
@@ -3392,7 +3436,7 @@ xtPublic void xt_tab_load_row_pointers(XTThreadPtr self, XTOpenTablePtr ot)
 				XT_LOCK_MEMORY_PTR(buff_ptr, ot->ot_row_file, offset, tfer, &self->st_statistics.st_rec, self);
 				if (buff_ptr) {
 					memcpy(buffer, buff_ptr, tfer);
-					XT_UNLOCK_MEMORY_PTR(ot->ot_row_file, self);
+					XT_UNLOCK_MEMORY_PTR(ot->ot_row_file, buff_ptr, TRUE, self);
 				}
 			}
 			rec_id += tab->tab_rows.tci_rows_per_page;
@@ -3521,7 +3565,7 @@ static void tab_restore_exception(XTExceptionPtr e)
  * FALSE if the record has already been freed. 
  *
  */
-xtPublic int xt_tab_remove_record(XTOpenTablePtr ot, xtRecordID rec_id, xtWord1 *rec_data, xtRecordID *prev_var_id, xtBool clean_delete, xtRowID row_id, xtXactID xn_id __attribute__((unused)))
+xtPublic int xt_tab_remove_record(XTOpenTablePtr ot, xtRecordID rec_id, xtWord1 *rec_data, xtRecordID *prev_var_id, xtBool clean_delete, xtRowID row_id, xtXactID XT_UNUSED(xn_id))
 {
 	register XTTableHPtr	tab = ot->ot_table;
 	size_t					rec_size;
@@ -3810,7 +3854,7 @@ static xtRowID tab_new_row(XTOpenTablePtr ot, XTTableHPtr tab)
 				xt_unlock_mutex_ns(&tab->tab_row_lock);
 				return 0;
 			}
-			xt_rwmutex_unlock(&seg->tcs_lock, ot->ot_thread->t_id);
+			TAB_CAC_UNLOCK(&seg->tcs_lock, ot->ot_thread->t_id);
 		}
 		tab->tab_row_eof_id++;
 	}
@@ -4906,6 +4950,7 @@ xtPublic xtBool xt_tab_seq_init(XTOpenTablePtr ot)
 	register XTTableHPtr tab = ot->ot_table;
 	
 	ot->ot_seq_page = NULL;
+	ot->ot_seq_data = NULL;
 	ot->ot_on_page = FALSE;
 	ot->ot_seq_offset = 0;
 
@@ -4958,6 +5003,7 @@ xtPublic void xt_tab_seq_reset(XTOpenTablePtr ot)
 	ot->ot_seq_rec_id = 0;
 	ot->ot_seq_eof_id = 0;
 	ot->ot_seq_page = NULL;
+	ot->ot_seq_data = NULL;
 	ot->ot_on_page = FALSE;
 	ot->ot_seq_offset = 0;
 }
@@ -4970,23 +5016,40 @@ xtPublic void xt_tab_seq_exit(XTOpenTablePtr ot)
 		tab->tab_recs.xt_tc_release_page(ot->ot_rec_file, ot->ot_seq_page, ot->ot_thread);
 		ot->ot_seq_page = NULL;
 	}
+	if (ot->ot_seq_data)
+		XT_UNLOCK_MEMORY_PTR(ot->ot_rec_file, ot->ot_seq_data, TRUE, ot->ot_thread);
 	ot->ot_on_page = FALSE;
 }
 
+#ifdef XT_USE_ROW_REC_MMAP_FILES
+#define TAB_SEQ_LOAD_CACHE		FALSE
+#else
+#ifdef XT_SEQ_SCAN_LOADS_CACHE
+#define TAB_SEQ_LOAD_CACHE		TRUE
+#else
+#define TAB_SEQ_LOAD_CACHE		FALSE
+#endif
+#endif
+
 xtPublic xtBool xt_tab_seq_next(XTOpenTablePtr ot, xtWord1 *buffer, xtBool *eof)
 {
 	register XTTableHPtr	tab = ot->ot_table;
 	register size_t			rec_size = tab->tab_dic.dic_rec_size;
 	xtWord1					*buff_ptr;
 	xtRecordID				new_rec_id;
-	xtBool					ptr_locked;
 	xtRecordID				invalid_rec = 0;
-	XTTabRecHeadDRec		rec_head;
 
 	next_page:
 	if (!ot->ot_on_page) {
-		if (!(ot->ot_on_page = tab->tab_recs.xt_tc_get_page(ot->ot_rec_file, ot->ot_seq_rec_id, &ot->ot_seq_page, &ot->ot_seq_offset, ot->ot_thread)))
+		if (!(ot->ot_on_page = tab->tab_recs.xt_tc_get_page(ot->ot_rec_file, ot->ot_seq_rec_id, TAB_SEQ_LOAD_CACHE, &ot->ot_seq_page, &ot->ot_seq_offset, ot->ot_thread)))
 			return FAILED;
+		if (!ot->ot_seq_page) {
+			XT_LOCK_MEMORY_PTR(ot->ot_seq_data, ot->ot_rec_file, xt_rec_id_to_rec_offset(tab, ot->ot_seq_rec_id), tab->tab_rows.tci_page_size, &ot->ot_thread->st_statistics.st_rec, ot->ot_thread);
+			if (!ot->ot_seq_data)
+				return FAILED;
+			ot->ot_on_page = TRUE;
+			ot->ot_seq_offset = 0;
+		}
 	}
 
 	next_record:
@@ -5001,22 +5064,19 @@ xtPublic xtBool xt_tab_seq_next(XTOpenTablePtr ot, xtWord1 *buffer, xtBool *eof)
 			tab->tab_recs.xt_tc_release_page(ot->ot_rec_file, ot->ot_seq_page, ot->ot_thread);
 			ot->ot_seq_page = NULL;
 		}
+		if (ot->ot_seq_data)
+			/* NULL here means that in the case of non-memory mapped
+			 * files we "keep" the lock.
+			 */
+			XT_UNLOCK_MEMORY_PTR(ot->ot_rec_file, ot->ot_seq_data, FALSE, ot->ot_thread);
 		ot->ot_on_page = FALSE;
 		goto next_page;
 	}
 
-	if (ot->ot_seq_page) {
-		ptr_locked = FALSE;
+	if (ot->ot_seq_page)
 		buff_ptr = ot->ot_seq_page->tcp_data + ot->ot_seq_offset;
-	}
-	else {
-		size_t red_size;
-
-		ptr_locked = TRUE;
-		if (!xt_pread_fmap(ot->ot_rec_file, xt_rec_id_to_rec_offset(tab, ot->ot_seq_rec_id), sizeof(XTTabRecHeadDRec), sizeof(XTTabRecHeadDRec), &rec_head, &red_size, &ot->ot_thread->st_statistics.st_rec, ot->ot_thread))
-			return FAILED;
-		buff_ptr = (xtWord1 *) &rec_head;
-	}
+	else
+		buff_ptr = ot->ot_seq_data + ot->ot_seq_offset;
 
 	/* This is the current record: */
 	ot->ot_curr_rec_id = ot->ot_seq_rec_id;
@@ -5033,7 +5093,6 @@ xtPublic xtBool xt_tab_seq_next(XTOpenTablePtr ot, xtWord1 *buffer, xtBool *eof)
 		case XT_ERR:
 			goto failed;
 		case XT_NEW:
-			ptr_locked = FALSE;
 			buff_ptr = ot->ot_row_rbuffer;
 			if (!xt_tab_get_rec_data(ot, new_rec_id, rec_size, ot->ot_row_rbuffer))
 				return XT_ERR;
@@ -5066,8 +5125,6 @@ xtPublic xtBool xt_tab_seq_next(XTOpenTablePtr ot, xtWord1 *buffer, xtBool *eof)
 			invalid_rec = 0;
 			goto next_record;
 		default:
-			if (ptr_locked)
-				XT_LOCK_MEMORY_PTR(buff_ptr, ot->ot_rec_file, xt_rec_id_to_rec_offset(tab, ot->ot_curr_rec_id), tab->tab_rows.tci_page_size, &ot->ot_thread->st_statistics.st_rec, ot->ot_thread);
 			break;
 	}
 
@@ -5099,15 +5156,11 @@ xtPublic xtBool xt_tab_seq_next(XTOpenTablePtr ot, xtWord1 *buffer, xtBool *eof)
 			break;
 		}
 	}
-	if (ptr_locked)
-		XT_UNLOCK_MEMORY_PTR(ot->ot_rec_file, ot->ot_thread);
 
 	*eof = FALSE;
 	return OK;
 
 	failed_1:
-	if (ptr_locked)
-		XT_UNLOCK_MEMORY_PTR(ot->ot_rec_file, ot->ot_thread);
 
 	failed:
 	return FAILED;
diff --git a/storage/pbxt/src/table_xt.h b/storage/pbxt/src/table_xt.h
index 6ba3eadd2ae..2382349a9c5 100644
--- a/storage/pbxt/src/table_xt.h
+++ b/storage/pbxt/src/table_xt.h
@@ -45,7 +45,17 @@ struct XTTablePath;
 #define XT_TAB_INCOMPATIBLE_VERSION	4
 #define XT_TAB_CURRENT_VERSION		5
 
-#define XT_IND_CURRENT_VERSION		3
+/* This version of the index does not have lazy
+ * delete. The new version is compatible with
+ * this and maintains the old format.
+ */
+#define XT_IND_NO_LAZY_DELETE		3
+#define XT_IND_LAZY_DELETE_OK		4
+#ifdef XT_USE_LAZY_DELETE
+#define XT_IND_CURRENT_VERSION		XT_IND_LAZY_DELETE_OK
+#else
+#define XT_IND_CURRENT_VERSION		XT_IND_NO_LAZY_DELETE
+#endif
 
 #define XT_HEAD_BUFFER_SIZE			1024
 
@@ -100,15 +110,21 @@ struct XTTablePath;
 #define XT_TAB_POOL_CLOSED			3				/* Cannot open table at the moment, the pool is closed. */
 #define XT_TAB_FAILED				4
 
-#define XT_TAB_ROW_USE_RW_MUTEX
+#ifdef XT_NO_ATOMICS
+#define XT_TAB_ROW_USE_PTHREAD_RW
+#else
+//#define XT_TAB_ROW_USE_RWMUTEX
+//#define XT_TAB_ROW_USE_SPINXSLOCK
+#define XT_TAB_ROW_USE_XSMUTEX
+#endif
 
-#ifdef XT_TAB_ROW_USE_FASTWRLOCK
-#define XT_TAB_ROW_LOCK_TYPE			XTFastRWLockRec
-#define XT_TAB_ROW_INIT_LOCK(s, i)		xt_fastrwlock_init(s, i)
-#define XT_TAB_ROW_FREE_LOCK(s, i)		xt_fastrwlock_free(s, i)	
-#define XT_TAB_ROW_READ_LOCK(i, s)		xt_fastrwlock_slock(i, s)
-#define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_fastrwlock_xlock(i, s)
-#define XT_TAB_ROW_UNLOCK(i, s)			xt_fastrwlock_unlock(i, s)
+#ifdef XT_TAB_ROW_USE_XSMUTEX
+#define XT_TAB_ROW_LOCK_TYPE			XTXSMutexRec
+#define XT_TAB_ROW_INIT_LOCK(s, i)		xt_xsmutex_init_with_autoname(s, i)
+#define XT_TAB_ROW_FREE_LOCK(s, i)		xt_xsmutex_free(s, i)	
+#define XT_TAB_ROW_READ_LOCK(i, s)		xt_xsmutex_slock(i, (s)->t_id)
+#define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_xsmutex_xlock(i, (s)->t_id)
+#define XT_TAB_ROW_UNLOCK(i, s)			xt_xsmutex_unlock(i, (s)->t_id)
 #elif defined(XT_TAB_ROW_USE_PTHREAD_RW)
 #define XT_TAB_ROW_LOCK_TYPE			xt_rwlock_type
 #define XT_TAB_ROW_INIT_LOCK(s, i)		xt_init_rwlock(s, i)
@@ -116,16 +132,23 @@ struct XTTablePath;
 #define XT_TAB_ROW_READ_LOCK(i, s)		xt_slock_rwlock_ns(i)
 #define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_xlock_rwlock_ns(i)
 #define XT_TAB_ROW_UNLOCK(i, s)			xt_unlock_rwlock_ns(i)
-#elif defined(XT_TAB_ROW_USE_RW_MUTEX)
+#elif defined(XT_TAB_ROW_USE_RWMUTEX)
 #define XT_TAB_ROW_LOCK_TYPE			XTRWMutexRec
 #define XT_TAB_ROW_INIT_LOCK(s, i)		xt_rwmutex_init_with_autoname(s, i)
 #define XT_TAB_ROW_FREE_LOCK(s, i)		xt_rwmutex_free(s, i)	
 #define XT_TAB_ROW_READ_LOCK(i, s)		xt_rwmutex_slock(i, (s)->t_id)
 #define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_rwmutex_xlock(i, (s)->t_id)
 #define XT_TAB_ROW_UNLOCK(i, s)			xt_rwmutex_unlock(i, (s)->t_id)
+#elif defined(XT_TAB_ROW_USE_SPINXSLOCK)
+#define XT_TAB_ROW_LOCK_TYPE			XTSpinXSLockRec
+#define XT_TAB_ROW_INIT_LOCK(s, i)		xt_spinxslock_init_with_autoname(s, i)
+#define XT_TAB_ROW_FREE_LOCK(s, i)		xt_spinxslock_free(s, i)	
+#define XT_TAB_ROW_READ_LOCK(i, s)		xt_spinxslock_slock(i, (s)->t_id)
+#define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_spinxslock_xlock(i, (s)->t_id)
+#define XT_TAB_ROW_UNLOCK(i, s)			xt_spinxslock_unlock(i, (s)->t_id)
 #else
 #define XT_TAB_ROW_LOCK_TYPE			XTSpinLockRec
-#define XT_TAB_ROW_INIT_LOCK(s, i)		xt_spinlock_init(s, i)
+#define XT_TAB_ROW_INIT_LOCK(s, i)		xt_spinlock_init_with_autoname(s, i)
 #define XT_TAB_ROW_FREE_LOCK(s, i)		xt_spinlock_free(s, i)	
 #define XT_TAB_ROW_READ_LOCK(i, s)		xt_spinlock_lock(i)
 #define XT_TAB_ROW_WRITE_LOCK(i, s)		xt_spinlock_lock(i)
@@ -441,6 +464,9 @@ typedef struct XTOpenTable {
 	xtRecordID				ot_seq_rec_id;							/* Current position of a sequential scan. */
 	xtRecordID				ot_seq_eof_id;							/* The EOF at the start of the sequential scan. */
 	XTTabCachePagePtr		ot_seq_page;							/* If ot_seq_buffer is non-NULL, then a page has been locked! */
+	xtWord1					*ot_seq_data;							/* Non-NULL if the data references memory mapped memory, or if it was
+																	 * allocated if no memory mapping is being used.
+																	 */
 	xtBool					ot_on_page;
 	size_t					ot_seq_offset;							/* Offset on the current page. */
 } XTOpenTableRec, *XTOpenTablePtr;
@@ -488,7 +514,7 @@ XTTableHPtr			xt_use_table_no_lock(XTThreadPtr self, struct XTDatabase *db, XTPa
 int					xt_use_table_by_id(struct XTThread *self, XTTableHPtr *tab, struct XTDatabase *db, xtTableID tab_id);
 XTOpenTablePtr		xt_open_table(XTTableHPtr tab);
 void				xt_close_table(XTOpenTablePtr ot, xtBool flush, xtBool have_table_lock);
-void				xt_drop_table(struct XTThread *self, XTPathStrPtr name);
+void				xt_drop_table(struct XTThread *self, XTPathStrPtr name, xtBool drop_db);
 void				xt_check_table(XTThreadPtr self, XTOpenTablePtr tab);
 void				xt_rename_table(struct XTThread *self, XTPathStrPtr old_name, XTPathStrPtr new_name);
 
diff --git a/storage/pbxt/src/thread_xt.cc b/storage/pbxt/src/thread_xt.cc
index bd7dca31cb5..a15b600a431 100644
--- a/storage/pbxt/src/thread_xt.cc
+++ b/storage/pbxt/src/thread_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #ifndef XT_WIN
 #include <unistd.h>
 #include <sys/time.h>
@@ -177,7 +181,7 @@ static void thr_log_newline(XTThreadPtr self, c_char *func, c_char *file, u_int
 #endif
 #endif
 
-void xt_log_flush(XTThreadPtr self __attribute__((unused)))
+void xt_log_flush(XTThreadPtr XT_UNUSED(self))
 {
 	fflush(log_file);
 }
@@ -466,7 +470,7 @@ static void thr_free_resources(XTThreadPtr self, XTResourcePtr top)
 	}
 }
 
-xtPublic void xt_bug(XTThreadPtr self __attribute__((unused)))
+xtPublic void xt_bug(XTThreadPtr XT_UNUSED(self))
 {
 	static int *bug_ptr = NULL;
 	
@@ -532,7 +536,11 @@ xtPublic void xt_throw_error(XTThreadPtr self, c_char *func, c_char *file, u_int
 
 #define XT_SYS_ERR_SIZE		300
 
-static c_char *thr_get_sys_error(int err, char *err_msg __attribute__((unused)))
+#ifdef XT_WIN
+static c_char *thr_get_sys_error(int err, char *err_msg)
+#else
+static c_char *thr_get_sys_error(int err, char *XT_UNUSED(err_msg))
+#endif
 {
 #ifdef XT_WIN
 	char *ptr;
@@ -638,7 +646,7 @@ static c_char *thr_get_err_string(int xt_err)
 		case XT_ERR_INDEX_CORRUPTED:		str = "Table `%s` index is corrupted, REPAIR TABLE required"; break;
 		case XT_ERR_NO_INDEX_CACHE:			str = "Not enough index cache memory to handle concurrent updates"; break;
 		case XT_ERR_INDEX_LOG_CORRUPT:		str = "Index log corrupt: '%s'"; break;
-		case XT_ERR_TOO_MANY_THREADS:		str = "Too many threads: %s, increase max_connections"; break;
+		case XT_ERR_TOO_MANY_THREADS:		str = "Too many threads: %s, increase pbxt_max_threads"; break;
 		case XT_ERR_TOO_MANY_WAITERS:		str = "Too many waiting threads: %s"; break;
 		case XT_ERR_INDEX_OLD_VERSION:		str = "Table `%s` index created by an older version, REPAIR TABLE required"; break;
 		case XT_ERR_PBXT_TABLE_EXISTS:		str = "System table cannot be dropped because PBXT table still exists"; break;
@@ -869,13 +877,18 @@ xtPublic void xt_log_errno(XTThreadPtr self, c_char *func, c_char *file, u_int l
  * -----------------------------------------------------------------------
  * Assertions and failures (one breakpoints for all failures)
  */
+//#define CRASH_ON_ASSERT
 
-xtPublic xtBool xt_assert(XTThreadPtr self __attribute__((unused)), c_char *expr, c_char *func, c_char *file, u_int line)
+xtPublic xtBool xt_assert(XTThreadPtr self, c_char *expr, c_char *func, c_char *file, u_int line)
 {
+	(void) self;
 #ifdef DEBUG
 	//xt_set_fflush(TRUE);
 	//xt_dump_trace();
 	printf("%s(%s:%d) %s\n", func, file, (int) line, expr);
+#ifdef CRASH_ON_ASSERT
+	abort();
+#endif
 #ifdef XT_WIN
 	FatalAppExit(0, "Assertion Failed!");
 #endif
@@ -981,11 +994,13 @@ static xtBool thr_setup_signals(void)
 }
 #endif
 
-static void *thr_main(void *data)
+typedef void *(*ThreadMainFunc)(XTThreadPtr self);
+
+extern "C" void *thr_main(void *data)
 {
 	ThreadDataPtr	td = (ThreadDataPtr) data;
 	XTThreadPtr		self = td->td_thr;
-	void			*(*start_routine)(XTThreadPtr);
+	ThreadMainFunc		start_routine;
 	void			*return_data;
 
 	enter_();
@@ -1857,7 +1872,7 @@ xtPublic void xt_signal_thread(XTThreadPtr target)
 	xt_broadcast_cond_ns(&target->t_cond);
 }
 
-xtPublic void xt_terminate_thread(XTThreadPtr self __attribute__((unused)), XTThreadPtr target)
+xtPublic void xt_terminate_thread(XTThreadPtr XT_UNUSED(self), XTThreadPtr target)
 {
 	target->t_quit = TRUE;
 	target->t_delayed_signal = SIGTERM;
diff --git a/storage/pbxt/src/thread_xt.h b/storage/pbxt/src/thread_xt.h
index 4344c5335b9..69f828758a1 100644
--- a/storage/pbxt/src/thread_xt.h
+++ b/storage/pbxt/src/thread_xt.h
@@ -44,14 +44,6 @@
  * Macros and defines
  */
 
-#ifdef XT_WIN
-#define __FUNC__						__FUNCTION__
-#elif defined(XT_SOLARIS)
-#define __FUNC__						"__func__"
-#else
-#define __FUNC__						__PRETTY_FUNCTION__
-#endif
-
 #define XT_ERR_MSG_SIZE					(PATH_MAX + 200)
 
 #ifdef DEBUG
@@ -291,6 +283,12 @@ typedef struct XTThread {
 	xtBool					st_xact_long_running;			/* TRUE if this is a long running writer transaction. */
 	xtWord4					st_visible_time;				/* Transactions committed before this time are visible. */
 	XTDataLogBufferRec		st_dlog_buf;
+	
+	/* A list of the last 10 transactions run by this connection: */
+#ifdef XT_WAIT_FOR_CLEANUP
+	u_int					st_last_xact;
+	xtXactID				st_prev_xact[XT_MAX_XACT_BEHIND];
+#endif
 
 	int						st_xact_mode;					/* The transaction mode. */
 	xtBool					st_ignore_fkeys;				/* TRUE if we must ignore foreign keys. */
diff --git a/storage/pbxt/src/trace_xt.cc b/storage/pbxt/src/trace_xt.cc
index e5881cb6d12..a913e2ae385 100644
--- a/storage/pbxt/src/trace_xt.cc
+++ b/storage/pbxt/src/trace_xt.cc
@@ -72,6 +72,12 @@ xtPublic xtBool xt_init_trace(void)
 	trace_log_offset = 0;
 	trace_log_end = 0;
 	trace_stat_count = 0;
+
+#ifdef XT_TRACK_CONNECTIONS
+	for (int i=0; i<XT_TRACK_MAX_CONNS; i++)
+		xt_track_conn_info[i].cu_t_id = i;
+#endif
+
 	return TRUE;
 }
 
@@ -343,3 +349,45 @@ xtPublic void xt_ftracef(char *fmt, ...)
 	va_end(ap);
 }
 
+/*
+ * -----------------------------------------------------------------------
+ * CONNECTION TRACKING
+ */
+
+#ifdef XT_TRACK_CONNECTIONS
+XTConnInfoRec	xt_track_conn_info[XT_TRACK_MAX_CONNS];
+
+static int trace_comp_conn_info(const void *a, const void *b)
+{
+	XTConnInfoPtr	ci_a = (XTConnInfoPtr) a, ci_b = (XTConnInfoPtr) b;
+
+	if (ci_a->ci_curr_xact_id > ci_b->ci_curr_xact_id)
+		return 1;
+	if (ci_a->ci_curr_xact_id < ci_b->ci_curr_xact_id)
+		return -1;
+	return 0;
+}
+
+xtPublic void xt_dump_conn_tracking(void)
+{
+	XTConnInfoRec	conn_info[XT_TRACK_MAX_CONNS];
+	XTConnInfoPtr	ptr;
+
+	memcpy(conn_info, xt_track_conn_info, sizeof(xt_track_conn_info));
+	qsort(conn_info, XT_TRACK_MAX_CONNS, sizeof(XTConnInfoRec), trace_comp_conn_info);
+
+	ptr = conn_info;
+	for (int i=0; i<XT_TRACK_MAX_CONNS; i++) {
+		if (ptr->ci_curr_xact_id || ptr->ci_prev_xact_id) {
+			printf("%3d curr=%d prev=%d prev-time=%ld\n", (int) ptr->cu_t_id, (int) ptr->ci_curr_xact_id, (int) ptr->ci_prev_xact_id, (long) ptr->ci_prev_xact_time);
+			if (i+1<XT_TRACK_MAX_CONNS) {
+				printf("    diff=%d\n", (int) (ptr+1)->ci_curr_xact_id - (int) ptr->ci_curr_xact_id);
+			}
+		}
+		ptr++;
+	}
+}
+
+#endif
+
+
diff --git a/storage/pbxt/src/trace_xt.h b/storage/pbxt/src/trace_xt.h
index 44cfa9945f1..9b00a5a05c7 100644
--- a/storage/pbxt/src/trace_xt.h
+++ b/storage/pbxt/src/trace_xt.h
@@ -46,4 +46,29 @@ void	xt_ftracef(char *fmt, ...);
 //#define PBXT_HANDLER_TRACE
 #endif
 
+/*
+ * -----------------------------------------------------------------------
+ * CONNECTION TRACKING
+ */
+
+#define XT_TRACK_CONNECTIONS
+
+#ifdef XT_TRACK_CONNECTIONS
+#define XT_TRACK_MAX_CONNS		500
+
+typedef struct XTConnInfo {
+	xtThreadID			cu_t_id;
+	xtXactID			ci_curr_xact_id;
+	xtWord8				ci_xact_start;
+
+	xtXactID			ci_prev_xact_id;
+	xtWord8				ci_prev_xact_time;
+} XTConnInfoRec, *XTConnInfoPtr;
+
+extern XTConnInfoRec xt_track_conn_info[XT_TRACK_MAX_CONNS];
+
+void	xt_dump_conn_tracking(void);
+
+#endif
+
 #endif
diff --git a/storage/pbxt/src/util_xt.cc b/storage/pbxt/src/util_xt.cc
index 6e1db1f5f73..7e805e6e044 100644
--- a/storage/pbxt/src/util_xt.cc
+++ b/storage/pbxt/src/util_xt.cc
@@ -61,7 +61,7 @@ xtPublic xtWord8 xt_time_now(void)
 	return ms;
 }
 
-xtPublic void xt_free_nothing(struct XTThread XT_UNUSED(*thr), void XT_UNUSED(*x))
+xtPublic void xt_free_nothing(struct XTThread *XT_UNUSED(thread), void *XT_UNUSED(x))
 {
 }
 
diff --git a/storage/pbxt/src/xaction_xt.cc b/storage/pbxt/src/xaction_xt.cc
index be3cb650bcb..a27c9e65634 100644
--- a/storage/pbxt/src/xaction_xt.cc
+++ b/storage/pbxt/src/xaction_xt.cc
@@ -23,6 +23,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <time.h>
 #include <signal.h>
 
@@ -48,7 +52,7 @@
 #endif
 
 static void xn_sw_wait_for_xact(XTThreadPtr self, XTDatabaseHPtr db, u_int hsecs);
-static xtBool xn_get_xact_details(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)), int *flags, xtXactID *start, xtXactID *end, xtThreadID *thd_id);
+static xtBool xn_get_xact_details(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr XT_UNUSED(thread), int *flags, xtXactID *start, xtXactID *end, xtThreadID *thd_id);
 static xtBool xn_get_xact_pointer(XTDatabaseHPtr db, xtXactID xn_id, XTXactDataPtr *xact_ptr);
 
 /* ============================================================================================== */
@@ -203,7 +207,7 @@ typedef struct XNWaitFor {
 	xtXactID				wf_for_me_xn_id;		/* The transaction we are waiting for. */
 } XNWaitForRec, *XNWaitForPtr;
 
-static int xn_compare_wait_for(XTThreadPtr XT_UNUSED(self), register const void XT_UNUSED(*thunk), register const void *a, register const void *b)
+static int xn_compare_wait_for(XTThreadPtr XT_UNUSED(self), register const void *XT_UNUSED(thunk), register const void *a, register const void *b)
 {
 	xtXactID		*x = (xtXactID *) a;
 	XNWaitForPtr	y = (XNWaitForPtr) b;
@@ -215,7 +219,7 @@ static int xn_compare_wait_for(XTThreadPtr XT_UNUSED(self), register const void
 	return 1;
 }
 
-static void xn_free_wait_for(XTThreadPtr XT_UNUSED(self), void XT_UNUSED(*thunk), void XT_UNUSED(*item))
+static void xn_free_wait_for(XTThreadPtr XT_UNUSED(self), void *XT_UNUSED(thunk), void *XT_UNUSED(item))
 {
 }
 
@@ -446,7 +450,9 @@ xtPublic xtBool xt_xn_wait_for_xact(XTThreadPtr thread, XTXactWaitPtr xw, XTLock
 			xt_timed_wait_cond_ns(&my_wt->wt_cond, &my_wt->wt_lock, WAIT_FOR_XACT_TIME);
 		}
 
+		/* Unreachable
 		xt_unlock_mutex_ns(&my_wt->wt_lock);
+		*/
 	}
 
 	if (xw) {
@@ -753,12 +759,13 @@ xtPublic xtXactID xt_xn_get_curr_id(XTDatabaseHPtr db)
 	return curr_xn_id;
 }
 
-xtPublic XTXactDataPtr xt_xn_add_old_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)))
+xtPublic XTXactDataPtr xt_xn_add_old_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread)
 {
 	register XTXactDataPtr	xact;
 	register XTXactSegPtr 	seg;
 	register XTXactDataPtr	*hash;
 
+	(void) thread;
 	seg = &db->db_xn_idx[xn_id & XT_XN_SEGMENT_MASK];
 	XT_XACT_WRITE_LOCK(&seg->xs_tab_lock, thread);
 	hash = &seg->xs_table[(xn_id >> XT_XN_SEGMENT_SHIFTS) % XT_XN_HASH_TABLE_SIZE];
@@ -778,7 +785,7 @@ xtPublic XTXactDataPtr xt_xn_add_old_xact(XTDatabaseHPtr db, xtXactID xn_id, XTT
 		 */
 		db->db_sw_faster |= XT_SW_NO_MORE_XACT_SLOTS;
 		if (!(xact = (XTXactDataPtr) xt_malloc_ns(sizeof(XTXactDataRec)))) {
-			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 			return NULL;
 		}
 	}
@@ -797,7 +804,7 @@ xtPublic XTXactDataPtr xt_xn_add_old_xact(XTDatabaseHPtr db, xtXactID xn_id, XTT
 		seg->xs_last_xn_id = xn_id;
 
 	done_ok:
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 #ifdef HIGH_X
 	tot_alloced++;
 	if (tot_alloced > high_alloced)
@@ -806,12 +813,13 @@ xtPublic XTXactDataPtr xt_xn_add_old_xact(XTDatabaseHPtr db, xtXactID xn_id, XTT
 	return xact;
 }
 
-static XTXactDataPtr xn_add_new_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)))
+static XTXactDataPtr xn_add_new_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread)
 {
 	register XTXactDataPtr	xact;
 	register XTXactSegPtr 	seg;
 	register XTXactDataPtr	*hash;
 
+	(void) thread;
 	seg = &db->db_xn_idx[xn_id & XT_XN_SEGMENT_MASK];
 	XT_XACT_WRITE_LOCK(&seg->xs_tab_lock, thread);
 	hash = &seg->xs_table[(xn_id >> XT_XN_SEGMENT_SHIFTS) % XT_XN_HASH_TABLE_SIZE];
@@ -825,7 +833,7 @@ static XTXactDataPtr xn_add_new_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThread
 		 */
 		db->db_sw_faster |= XT_SW_NO_MORE_XACT_SLOTS;
 		if (!(xact = (XTXactDataPtr) xt_malloc_ns(sizeof(XTXactDataRec)))) {
-			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 			return NULL;
 		}
 	}
@@ -841,7 +849,7 @@ static XTXactDataPtr xn_add_new_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThread
 	xact->xd_flags = 0;
 
 	seg->xs_last_xn_id = xn_id;
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 #ifdef HIGH_X
 	tot_alloced++;
 	if (tot_alloced > high_alloced)
@@ -850,7 +858,7 @@ static XTXactDataPtr xn_add_new_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThread
 	return xact;
 }
 
-static xtBool xn_get_xact_details(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)), int *flags, xtXactID *start, xtWord4 *end, xtThreadID *thd_id)
+static xtBool xn_get_xact_details(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr XT_UNUSED(thread), int *flags, xtXactID *start, xtWord4 *end, xtThreadID *thd_id)
 {
 	register XTXactSegPtr 	seg;
 	register XTXactDataPtr	xact;
@@ -874,7 +882,7 @@ static xtBool xn_get_xact_details(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr
 		}
 		xact = xact->xd_next_xact;
 	}
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, FALSE);
 	return found;
 }
 
@@ -900,11 +908,11 @@ static xtBool xn_get_xact_pointer(XTDatabaseHPtr db, xtXactID xn_id, XTXactDataP
 		}
 		xact = xact->xd_next_xact;
 	}
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, FALSE);
 	return found;
 }
 
-static xtBool xn_get_xact_start(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)), xtLogID *log_id, xtLogOffset *log_offset)
+static xtBool xn_get_xact_start(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr XT_UNUSED(thread), xtLogID *log_id, xtLogOffset *log_offset)
 {
 	register XTXactSegPtr 	seg;
 	register XTXactDataPtr	xact;
@@ -922,12 +930,12 @@ static xtBool xn_get_xact_start(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr t
 		}
 		xact = xact->xd_next_xact;
 	}
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, FALSE);
 	return found;
 }
 
 /* NOTE: this function may only be used by the sweeper or the recovery process. */
-xtPublic XTXactDataPtr xt_xn_get_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)))
+xtPublic XTXactDataPtr xt_xn_get_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr XT_UNUSED(thread))
 {
 	register XTXactSegPtr 	seg;
 	register XTXactDataPtr	xact;
@@ -940,7 +948,7 @@ xtPublic XTXactDataPtr xt_xn_get_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThrea
 			break;
 		xact = xact->xd_next_xact;
 	}
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, FALSE);
 	return xact;
 }
 
@@ -948,11 +956,12 @@ xtPublic XTXactDataPtr xt_xn_get_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThrea
  * Delete a transaction, return TRUE if the transaction
  * was found.
  */
-xtPublic xtBool xt_xn_delete_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread __attribute__((unused)))
+xtPublic xtBool xt_xn_delete_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr thread)
 {
 	XTXactDataPtr	xact, pxact = NULL;
 	XTXactSegPtr 	seg;
 
+	(void) thread;
 	seg = &db->db_xn_idx[xn_id & XT_XN_SEGMENT_MASK];
 	XT_XACT_WRITE_LOCK(&seg->xs_tab_lock, thread);
 	xact = seg->xs_table[(xn_id >> XT_XN_SEGMENT_SHIFTS) % XT_XN_HASH_TABLE_SIZE];
@@ -963,13 +972,13 @@ xtPublic xtBool xt_xn_delete_xact(XTDatabaseHPtr db, xtXactID xn_id, XTThreadPtr
 			else
 				 seg->xs_table[(xn_id >> XT_XN_SEGMENT_SHIFTS) % XT_XN_HASH_TABLE_SIZE] = xact->xd_next_xact;
 			xn_free_xact(db, seg, xact);
-			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+			XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 			return TRUE;
 		}
 		pxact = xact;
 		xact = xact->xd_next_xact;
 	}
-	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread);
+	XT_XACT_UNLOCK(&seg->xs_tab_lock, thread, TRUE);
 	return FALSE;
 }
 
@@ -1253,6 +1262,10 @@ xtPublic xtBool xt_xn_begin(XTThreadPtr self)
 #ifdef TRACE_TRANSACTION
 	xt_ttracef(self, "BEGIN T%lu\n", (u_long) self->st_xact_data->xd_start_xn_id);
 #endif
+#ifdef XT_TRACK_CONNECTIONS
+	xt_track_conn_info[self->t_id].ci_curr_xact_id = self->st_xact_data->xd_start_xn_id;
+	xt_track_conn_info[self->t_id].ci_xact_start = xt_trace_clock();
+#endif
 	return OK;
 }
 
@@ -1375,6 +1388,13 @@ static xtBool xn_end_xact(XTThreadPtr thread, u_int status)
 
 		thread->st_xact_data = NULL;
 
+#ifdef XT_TRACK_CONNECTIONS
+		xt_track_conn_info[thread->t_id].ci_prev_xact_id = xt_track_conn_info[thread->t_id].ci_curr_xact_id;
+		xt_track_conn_info[thread->t_id].ci_prev_xact_time = xt_trace_clock() - xt_track_conn_info[thread->t_id].ci_xact_start;
+		xt_track_conn_info[thread->t_id].ci_curr_xact_id = 0;
+		xt_track_conn_info[thread->t_id].ci_xact_start = 0;
+#endif
+
 		xt_xn_wakeup_waiting_threads(thread);
 
 		/* {WAKE-SW} Waking the sweeper
@@ -1401,6 +1421,19 @@ static xtBool xn_end_xact(XTThreadPtr thread, u_int status)
 
 		/* Don't get too far ahead of the sweeper! */
 		if (writer) {
+#ifdef XT_WAIT_FOR_CLEANUP
+			xtXactID	wait_xn_id;
+			
+			/* This is the transaction that was committed 3 transactions ago: */
+			wait_xn_id = thread->st_prev_xact[thread->st_last_xact];
+			thread->st_prev_xact[thread->st_last_xact] = xn_id;
+			/* This works because XT_MAX_XACT_BEHIND == 2! */
+			ASSERT_NS((thread->st_last_xact + 1) % XT_MAX_XACT_BEHIND == thread->st_last_xact ^ 1);
+			thread->st_last_xact ^= 1;
+			while (xt_xn_is_before(db->db_xn_to_clean_id, wait_xn_id) && (db->db_sw_faster & XT_SW_TOO_FAR_BEHIND)) {
+				xt_critical_wait();
+			}
+#else
 			if ((db->db_sw_faster & XT_SW_TOO_FAR_BEHIND) != 0) {
 				xtWord8	then = xt_trace_clock() + (xtWord8) 20000;
 
@@ -1412,6 +1445,7 @@ static xtBool xn_end_xact(XTThreadPtr thread, u_int status)
 						break;
 				}
 			}
+#endif
 		}
 	}
 	return ok;
@@ -1854,7 +1888,7 @@ static xtBool xn_sw_cleanup_done(XTThreadPtr self, XTOpenTablePtr ot, xtRecordID
 	return FALSE;
 }
 
-static void xn_sw_clean_indices(XTThreadPtr self __attribute__((unused)), XTOpenTablePtr ot, xtRecordID rec_id, xtRowID row_id, xtWord1 *rec_data, xtWord1 *rec_buffer)
+static void xn_sw_clean_indices(XTThreadPtr XT_NDEBUG_UNUSED(self), XTOpenTablePtr ot, xtRecordID rec_id, xtRowID row_id, xtWord1 *rec_data, xtWord1 *rec_buffer)
 {
 	XTTableHPtr	tab = ot->ot_table;
 	u_int		cols_req;
@@ -2599,7 +2633,13 @@ xtPublic void xt_wait_for_sweeper(XTThreadPtr self, XTDatabaseHPtr db, int abort
 
 	if (db->db_sw_thread) {
 		then = time(NULL);
-		while (!xt_xn_is_before(xt_xn_get_curr_id(db), db->db_xn_to_clean_id)) { // was db->db_xn_to_clean_id <= xt_xn_get_curr_id(db)
+		/* Changed xt_xn_get_curr_id(db) to db->db_xn_curr_id,
+		 * This should work because we are not concerned about the difference
+		 * between xt_xn_get_curr_id(db) and db->db_xn_curr_id,
+		 * Which is just a matter of when transactions we can expect ot find
+		 * in memory (see {GAP-INC-ADD-XACT})
+		 */
+		while (!xt_xn_is_before(db->db_xn_curr_id, db->db_xn_to_clean_id)) { // was db->db_xn_to_clean_id <= xt_xn_get_curr_id(db)
 			xt_lock_mutex(self, &db->db_sw_lock);
 			pushr_(xt_unlock_mutex, &db->db_sw_lock);
 			xt_wakeup_sweeper(db);
diff --git a/storage/pbxt/src/xaction_xt.h b/storage/pbxt/src/xaction_xt.h
index 9a651fc2532..7bcc46dfcf3 100644
--- a/storage/pbxt/src/xaction_xt.h
+++ b/storage/pbxt/src/xaction_xt.h
@@ -36,28 +36,48 @@ struct XTOpenTable;
 
 #ifdef XT_USE_XACTION_DEBUG_SIZES
 
-#define XT_XN_DATA_ALLOC_COUNT	400
-#define XT_XN_SEGMENT_SHIFTS	1
-#define XT_XN_HASH_TABLE_SIZE	31
 #define XT_TN_NUMBER_INCREMENT	20
 #define XT_TN_MAX_TO_FREE		20
 #define XT_TN_MAX_TO_FREE_WASTE	3
 #define XT_TN_MAX_TO_FREE_CHECK	3
 #define XT_TN_MAX_TO_FREE_INC	3
 
+#define XT_XN_SEGMENT_SHIFTS	1
+
 #else
 
-#define XT_XN_DATA_ALLOC_COUNT	1250	// Number of pre-allocated transaction data structures per segment
-#define XT_XN_SEGMENT_SHIFTS	5		// (32)
-#define XT_XN_HASH_TABLE_SIZE	1279	// This is a prime number!
 #define XT_TN_NUMBER_INCREMENT	100		// The increment of the transaction number on restart
 #define XT_TN_MAX_TO_FREE		800		// The maximum size of the "to free" list
 #define XT_TN_MAX_TO_FREE_WASTE	400
 #define XT_TN_MAX_TO_FREE_CHECK	100		// Once we have exceeded the limit, we only try in intervals
 #define XT_TN_MAX_TO_FREE_INC	100
 
+//#define XT_XN_SEGMENT_SHIFTS	5		// (32)
+//#define XT_XN_SEGMENT_SHIFTS	6		// (64)
+//#define XT_XN_SEGMENT_SHIFTS	7		// (128)
+#define XT_XN_SEGMENT_SHIFTS	8		// (256)
+//#define XT_XN_SEGMENT_SHIFTS	9		// (512)
+
 #endif
 
+/* The hash table size (a prime number) */
+#if XT_XN_SEGMENT_SHIFTS == 1		// (1)
+#define XT_XN_HASH_TABLE_SIZE	1301
+#elif XT_XN_SEGMENT_SHIFTS == 5		// (32)
+#define XT_XN_HASH_TABLE_SIZE	1009
+#elif XT_XN_SEGMENT_SHIFTS == 6		// (64)
+#define XT_XN_HASH_TABLE_SIZE	503
+#elif XT_XN_SEGMENT_SHIFTS == 7		// (128)
+#define XT_XN_HASH_TABLE_SIZE	251
+#elif XT_XN_SEGMENT_SHIFTS == 8		// (256)
+#define XT_XN_HASH_TABLE_SIZE	127
+#elif XT_XN_SEGMENT_SHIFTS == 9		// (512)
+#define XT_XN_HASH_TABLE_SIZE	67
+#endif
+
+/* Number of pre-allocated transaction data structures per segment */
+#define XT_XN_DATA_ALLOC_COUNT	XT_XN_HASH_TABLE_SIZE
+
 #define XT_XN_NO_OF_SEGMENTS	(1 << XT_XN_SEGMENT_SHIFTS)
 #define XT_XN_SEGMENT_MASK		(XT_XN_NO_OF_SEGMENTS - 1)
 
@@ -94,36 +114,34 @@ typedef struct XTXactData {
 
 } XTXactDataRec, *XTXactDataPtr;
 
-#define XT_XACT_USE_SPINLOCK
+#ifdef XT_NO_ATOMICS
+#define XT_XACT_USE_PTHREAD_RW
+#else
+//#define XT_XACT_USE_SKEWRWLOCK
+#define XT_XACT_USE_SPINXSLOCK
+#endif
 
-#ifdef XT_XACT_USE_FASTWRLOCK
-#define XT_XACT_LOCK_TYPE				XTFastRWLockRec
-#define XT_XACT_INIT_LOCK(s, i)			xt_fastrwlock_init(s, i)
-#define XT_XACT_FREE_LOCK(s, i)			xt_fastrwlock_free(s, i)	
-#define XT_XACT_READ_LOCK(i, s)			xt_fastrwlock_slock(i, s)
-#define XT_XACT_WRITE_LOCK(i, s)		xt_fastrwlock_xlock(i, s)
-#define XT_XACT_UNLOCK(i, s)			xt_fastrwlock_unlock(i, s)
-#elif defined(XT_XACT_USE_PTHREAD_RW)
+#if defined(XT_XACT_USE_PTHREAD_RW)
 #define XT_XACT_LOCK_TYPE				xt_rwlock_type
 #define XT_XACT_INIT_LOCK(s, i)			xt_init_rwlock(s, i)
 #define XT_XACT_FREE_LOCK(s, i)			xt_free_rwlock(i)	
 #define XT_XACT_READ_LOCK(i, s)			xt_slock_rwlock_ns(i)
 #define XT_XACT_WRITE_LOCK(i, s)		xt_xlock_rwlock_ns(i)
-#define XT_XACT_UNLOCK(i, s)			xt_unlock_rwlock_ns(i)
-#elif defined(XT_XACT_USE_RW_MUTEX)
-#define XT_XACT_LOCK_TYPE				XTRWMutexRec
-#define XT_XACT_INIT_LOCK(s, i)			xt_rwmutex_init(s, i)
-#define XT_XACT_FREE_LOCK(s, i)			xt_rwmutex_free(s, i)	
-#define XT_XACT_READ_LOCK(i, s)			xt_rwmutex_slock(i, (s)->t_id)
-#define XT_XACT_WRITE_LOCK(i, s)		xt_rwmutex_xlock(i, (s)->t_id)
-#define XT_XACT_UNLOCK(i, s)			xt_rwmutex_unlock(i, (s)->t_id)
+#define XT_XACT_UNLOCK(i, s, b)			xt_unlock_rwlock_ns(i)
+#elif defined(XT_XACT_USE_SPINXSLOCK)
+#define XT_XACT_LOCK_TYPE				XTSpinXSLockRec
+#define XT_XACT_INIT_LOCK(s, i)			xt_spinxslock_init_with_autoname(s, i)
+#define XT_XACT_FREE_LOCK(s, i)			xt_spinxslock_free(s, i)	
+#define XT_XACT_READ_LOCK(i, s)			xt_spinxslock_slock(i)
+#define XT_XACT_WRITE_LOCK(i, s)		xt_spinxslock_xlock(i, (s)->t_id)
+#define XT_XACT_UNLOCK(i, s, b)			xt_spinxslock_unlock(i, b)
 #else
-#define XT_XACT_LOCK_TYPE				XTSpinLockRec
-#define XT_XACT_INIT_LOCK(s, i)			xt_spinlock_init_with_autoname(s, i)
-#define XT_XACT_FREE_LOCK(s, i)			xt_spinlock_free(s, i)	
-#define XT_XACT_READ_LOCK(i, s)			xt_spinlock_lock(i)
-#define XT_XACT_WRITE_LOCK(i, s)		xt_spinlock_lock(i)
-#define XT_XACT_UNLOCK(i, s)			xt_spinlock_unlock(i)
+#define XT_XACT_LOCK_TYPE				XTSkewRWLockRec
+#define XT_XACT_INIT_LOCK(s, i)			xt_skewrwlock_init_with_autoname(s, i)
+#define XT_XACT_FREE_LOCK(s, i)			xt_skewrwlock_free(s, i)	
+#define XT_XACT_READ_LOCK(i, s)			xt_skewrwlock_slock(i)
+#define XT_XACT_WRITE_LOCK(i, s)		xt_skewrwlock_xlock(i, (s)->t_id)
+#define XT_XACT_UNLOCK(i, s, b)			xt_skewrwlock_unlock(i, b)
 #endif
 
 /* We store the transactions in a number of segments, each
diff --git a/storage/pbxt/src/xactlog_xt.cc b/storage/pbxt/src/xactlog_xt.cc
index 82c0d85b770..9a1c01a41a3 100644
--- a/storage/pbxt/src/xactlog_xt.cc
+++ b/storage/pbxt/src/xactlog_xt.cc
@@ -28,6 +28,10 @@
 
 #include "xt_config.h"
 
+#ifdef DRIZZLED
+#include <bitset>
+#endif
+
 #include <signal.h>
 
 #include "xactlog_xt.h"
@@ -600,7 +604,12 @@ void XTDatabaseLog::xlog_setup(XTThreadPtr self, XTDatabaseHPtr db, off_t inp_lo
 	
 		xt_init_mutex_with_autoname(self, &xl_write_lock);
 		xt_init_cond(self, &xl_write_cond);
+#ifdef XT_XLOG_WAIT_SPINS
 		xt_writing = 0;
+		xt_waiting = 0;
+#else
+		xt_writing = FALSE;
+#endif
 		xl_log_id = 0;
 		xl_log_file = 0;
 	
@@ -752,6 +761,7 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 	xtLogOffset	req_flush_log_offset;
 	size_t		part_size;
 	xtWord8		flush_time;
+	xtWord2		sum;
 
 	if (!size1) {
 		/* Just flush the buffer... */
@@ -790,13 +800,13 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 		 * enough space in the buffer, or a flush
 		 * is required.
 		 */
+		xtWord8	then;
 		 
 		/*
 		 * The objective of the following code is to
 		 * pick one writer, out of all threads.
-		 * The others rest will wait for the writer.
+		 * The rest will wait for the writer.
 		 */
-		xtBool i_am_writer;
 
 		if (write_reason == WR_FLUSH) {
 			/* Before we flush, check if we should wait for running
@@ -805,8 +815,7 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 			if (xl_db->db_xn_writer_count - xl_db->db_xn_writer_wait_count - xl_db->db_xn_long_running_count > 0 && xl_last_flush_time) {
 				/* Wait for about as long as the last flush took,
 				 * the idea is to saturate the disk with flushing...: */
-				xtWord8	then = xt_trace_clock() + (xtWord8) xl_last_flush_time;
-
+				then = xt_trace_clock() + (xtWord8) xl_last_flush_time;
 				for (;;) {
 					xt_critical_wait();
 					/* If a thread leaves this loop because times up, or
@@ -831,6 +840,55 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 			}
 		}
 
+#ifdef XT_XLOG_WAIT_SPINS
+		/* Spin for 1/1000s: */
+		then = xt_trace_clock() + (xtWord8) 1000;
+		for (;;) {
+			if (!xt_atomic_tas4(&xt_writing, 1))
+				break;
+
+			/* If I am not the writer, then I just waited for the
+			 * writer. So it may be that my requirements have now
+			 * been met!
+			 */
+			if (write_reason == WR_FLUSH) {
+				/* If the reason was to flush, then
+				 * check the last flush sequence, maybe it is passed
+				 * our required sequence.
+				 */
+				if (xt_comp_log_pos(req_flush_log_id, req_flush_log_offset, xl_flush_log_id, xl_flush_log_offset) <= 0) {
+					/* The required flush position of the log is before
+					 * or equal to the actual flush position. This means the condition
+					 * for this thread have been satified (via group commit).
+					 * Nothing more to do!
+					 */
+					ASSERT_NS(xt_comp_log_pos(xl_write_log_id, xl_write_log_offset, xl_append_log_id, xl_append_log_offset) <= 0);
+					return OK;
+				}
+			}
+			else {
+				/* It may be that there is now space in the append buffer: */
+				if (xl_append_buf_pos + size1 + size2 <= xl_size_of_buffers)
+					goto copy_to_log_buffer;
+			}
+
+			if (xt_trace_clock() >= then) {
+				xt_lock_mutex_ns(&xl_write_lock);
+				xt_waiting++;
+				if (!xt_timed_wait_cond_ns(&xl_write_cond, &xl_write_lock, 500)) {
+					xt_waiting--;
+					xt_unlock_mutex_ns(&xl_write_lock);
+					return FALSE;
+				}
+				xt_waiting--;
+				xt_unlock_mutex_ns(&xl_write_lock);
+			}
+			else
+				xt_critical_wait();
+		}
+#else
+		xtBool i_am_writer;
+
 		i_am_writer = FALSE;
 		xt_lock_mutex_ns(&xl_write_lock);
 		if (xt_writing) {
@@ -873,6 +931,7 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 				
 			goto write_log_to_file;
 		}
+#endif
 
 		/* I am the writer, check the conditions, again: */
 		if (write_reason == WR_FLUSH) {
@@ -881,8 +940,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 				/* The writers required flush position is before or equal
 				 * to the actual position, so the writer is done...
 				 */
+#ifdef XT_XLOG_WAIT_SPINS
+				xt_writing = 0;
+				if (xt_waiting)
+					xt_cond_wakeall(&xl_write_cond);
+#else
 				xt_writing = FALSE;
 				xt_cond_wakeall(&xl_write_cond);
+#endif
 				ASSERT_NS(xt_comp_log_pos(xl_write_log_id, xl_write_log_offset, xl_append_log_id, xl_append_log_offset) <= 0);
 				return OK;
 			}
@@ -923,8 +988,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 						xt_unlock_mutex_ns(&xl_db->db_wr_lock);
 					}
 				}
+#ifdef XT_XLOG_WAIT_SPINS
+				xt_writing = 0;
+				if (xt_waiting)
+					xt_cond_wakeall(&xl_write_cond);
+#else
 				xt_writing = FALSE;
 				xt_cond_wakeall(&xl_write_cond);
+#endif
 				ASSERT_NS(xt_comp_log_pos(xl_write_log_id, xl_write_log_offset, xl_append_log_id, xl_append_log_offset) <= 0);
 				return ok;
 			}
@@ -934,8 +1005,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 			 * to copy our data into the buffer:
 			 */
 			if (xl_append_buf_pos + size1 + size2 <= xl_size_of_buffers) {
+#ifdef XT_XLOG_WAIT_SPINS
+				xt_writing = 0;
+				if (xt_waiting)
+					xt_cond_wakeall(&xl_write_cond);
+#else
 				xt_writing = FALSE;
 				xt_cond_wakeall(&xl_write_cond);
+#endif
 				goto copy_to_log_buffer;
 			}
 		}
@@ -1085,8 +1162,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 				 * position, continue writing: */
 				goto rewrite;
 
+#ifdef XT_XLOG_WAIT_SPINS
+			xt_writing = 0;
+			if (xt_waiting)
+				xt_cond_wakeall(&xl_write_cond);
+#else
 			xt_writing = FALSE;
 			xt_cond_wakeall(&xl_write_cond);
+#endif
 			ASSERT_NS(xt_comp_log_pos(xl_write_log_id, xl_write_log_offset, xl_append_log_id, xl_append_log_offset) <= 0);
 			return OK;
 		}
@@ -1100,8 +1183,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 		if (xl_append_buf_pos + size1 + size2 > xl_size_of_buffers)
 			goto rewrite;
 
+#ifdef XT_XLOG_WAIT_SPINS
+		xt_writing = 0;
+		if (xt_waiting)
+			xt_cond_wakeall(&xl_write_cond);
+#else
 		xt_writing = FALSE;
 		xt_cond_wakeall(&xl_write_cond);
+#endif
 	}
 
 	copy_to_log_buffer:
@@ -1146,8 +1235,6 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 		case XT_LOG_ENT_DELETE_BG:
 		case XT_LOG_ENT_DELETE_FL:
 		case XT_LOG_ENT_DELETE_FL_BG:
-			xtWord2 sum;
-			
 			sum = XT_GET_DISK_2(record->xu.xu_checksum_2) ^ XT_CHECKSUM_2(xl_append_log_id);
 			XT_SET_DISK_2(record->xu.xu_checksum_2, sum);
 
@@ -1158,6 +1245,10 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 				xl_db->db_xn_total_writer_count++;
 			}
 			break;
+		case XT_LOG_ENT_REC_REMOVED_BI:
+			sum = XT_GET_DISK_2(record->xu.xu_checksum_2) ^ XT_CHECKSUM_2(xl_append_log_id);
+			XT_SET_DISK_2(record->xu.xu_checksum_2, sum);
+			break;
 		case XT_LOG_ENT_ROW_NEW:
 		case XT_LOG_ENT_ROW_NEW_FL:
 			record->xl.xl_checksum_1 ^= XT_CHECKSUM_1(xl_append_log_id);
@@ -1209,8 +1300,14 @@ xtBool XTDatabaseLog::xlog_append(XTThreadPtr thread, size_t size1, xtWord1 *dat
 	return OK;
 
 	write_failed:
+#ifdef XT_XLOG_WAIT_SPINS
+	xt_writing = 0;
+	if (xt_waiting)
+		xt_cond_wakeall(&xl_write_cond);
+#else
 	xt_writing = FALSE;
 	xt_cond_wakeall(&xl_write_cond);
+#endif
 	return FAILED;
 }
 
@@ -1595,7 +1692,7 @@ void XTDatabaseLog::xlog_seq_close(XTXactSeqReadPtr seq)
 	seq->xseq_log_eof = 0;
 }
 
-xtBool XTDatabaseLog::xlog_seq_start(XTXactSeqReadPtr seq, xtLogID log_id, xtLogOffset log_offset, xtBool missing_ok __attribute__((unused)))
+xtBool XTDatabaseLog::xlog_seq_start(XTXactSeqReadPtr seq, xtLogID log_id, xtLogOffset log_offset, xtBool XT_UNUSED(missing_ok))
 {
 	if (seq->xseq_rec_log_id != log_id) {
 		seq->xseq_rec_log_id = log_id;
@@ -2094,7 +2191,9 @@ xtBool XTDatabaseLog::xlog_seq_next(XTXactSeqReadPtr seq, XTXactLogBufferDPtr *r
 			goto return_empty;
 		}
 		default:
-			ASSERT_NS(FALSE);
+			/* It is possible to land here after a crash, if the
+			 * log was not completely written.
+			 */
 			seq->xseq_record_len = 0;
 			goto return_empty;
 	}
@@ -2304,7 +2403,13 @@ static void xlog_wr_wait_for_log_flush(XTThreadPtr self, XTDatabaseHPtr db)
 				 * the wait, and the sweeper has nothing to do, and the checkpointer.
 				 */
 				if (db->db_xn_curr_id == last_xn_id &&
-					xt_xn_is_before(xt_xn_get_curr_id(db), db->db_xn_to_clean_id) && // db->db_xn_curr_id < db->db_xn_to_clean_id
+					/* Changed xt_xn_get_curr_id(db) to db->db_xn_curr_id,
+					 * This should work because we are not concerned about the difference
+					 * between xt_xn_get_curr_id(db) and db->db_xn_curr_id,
+					 * Which is just a matter of when transactions we can expect ot find
+					 * in memory (see {GAP-INC-ADD-XACT})
+					 */
+					xt_xn_is_before(db->db_xn_curr_id, db->db_xn_to_clean_id) && // db->db_xn_curr_id < db->db_xn_to_clean_id
 					!db->db_restart.xres_is_checkpoint_pending(db->db_xlog.xl_write_log_id, db->db_xlog.xl_write_log_offset)) {
 					/* There seems to be no activity at the moment.
 					 * this might be a good time to write the log data.
diff --git a/storage/pbxt/src/xactlog_xt.h b/storage/pbxt/src/xactlog_xt.h
index 391b646b53f..2db11898ec2 100644
--- a/storage/pbxt/src/xactlog_xt.h
+++ b/storage/pbxt/src/xactlog_xt.h
@@ -373,6 +373,13 @@ typedef struct XTXactLogFile {
 /*
  * The transaction log. Each database has one.
  */
+ 
+/* Does not seem to make much difference... */
+#ifndef XT_NO_ATOMICS
+/* This function uses atomic ops: */
+//#define XT_XLOG_WAIT_SPINS
+#endif
+
 typedef struct XTDatabaseLog {
 	struct XTDatabase		*xl_db;
 
@@ -390,7 +397,12 @@ typedef struct XTDatabaseLog {
 	/* The writer log buffer: */
 	xt_mutex_type			xl_write_lock;
 	xt_cond_type			xl_write_cond;
+#ifdef XT_XLOG_WAIT_SPINS
+	xtWord4					xt_writing;						/* 1 if a thread is writing. */
+	xtWord4					xt_waiting;						/* Count of the threads waiting on the xl_write_cond. */
+#else
 	xtBool					xt_writing;						/* TRUE if a thread is writing. */
+#endif
 	xtLogID					xl_log_id;						/* The number of the write log. */
 	XTOpenFilePtr			xl_log_file;					/* The open write log. */
 
diff --git a/storage/pbxt/src/xt_config.h b/storage/pbxt/src/xt_config.h
index 6571ebdaebe..52c5d061c3d 100644
--- a/storage/pbxt/src/xt_config.h
+++ b/storage/pbxt/src/xt_config.h
@@ -96,4 +96,22 @@ const int max_connections = 500;
 #define XT_SOLARIS
 #endif
 
+/*
+ * Definition of which atomic operations to use:
+ */
+#ifdef XT_WIN
+/* MS Studio style embedded assembler for x86 */
+#define XT_ATOMIC_WIN32_X86
+#elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
+/* Use GNU style embedded assembler for x86 */
+#define XT_ATOMIC_GNUC_X86
+#elif defined(XT_SOLARIS)
+/* Use Sun atomic operations library
+ * http://docs.sun.com/app/docs/doc/816-5168/atomic-ops-3c?a=view
+ */
+#define XT_ATOMIC_SOLARIS_LIB
+#else
+#define XT_NO_ATOMICS
+#endif
+
 #endif
diff --git a/storage/pbxt/src/xt_defs.h b/storage/pbxt/src/xt_defs.h
index 16981ddc672..98867f746c9 100644
--- a/storage/pbxt/src/xt_defs.h
+++ b/storage/pbxt/src/xt_defs.h
@@ -187,7 +187,16 @@ typedef struct XTPathStr {
 	char				ps_path[XT_VAR_LENGTH];
 } *XTPathStrPtr;
 
-#define XT_UNUSED(x)	x __attribute__((__unused__))
+//#define XT_UNUSED(x)		x __attribute__((__unused__))
+#define XT_UNUSED(x)
+
+/* Only used when DEBUG is on: */
+#ifdef DEBUG
+#define XT_NDEBUG_UNUSED(x)	x
+#else
+//#define XT_NDEBUG_UNUSED(x)	x __attribute__((__unused__))
+#define XT_NDEBUG_UNUSED(x)
+#endif
 
 /* ----------------------------------------------------------------------
  * MAIN CONSTANTS
@@ -267,8 +276,10 @@ typedef struct XTPathStr {
  * the row list is scanned.
  *
  * For more details see [(9)].
+ * 223, 1019, 3613
  */
-#define XT_ROW_RWLOCKS					223
+#define XT_ROW_RWLOCKS					1019
+//#define XT_ROW_RWLOCKS					223
 
 /*
  * These are the number of row lock "slots" per table.
@@ -292,31 +303,20 @@ typedef struct XTPathStr {
  */
 #define XT_OPEN_TABLE_FREE_TIME			30
 
-#ifdef XT_USE_GLOBAL_DEBUG_SIZES
-/*
- * DEBUG SIZES!
- * Reduce the thresholds to make things happen faster.
+/* Define this in order to use memory mapped files
+ * (record and row pointer files only).
+ *
+ * This makes no difference in sysbench R/W performance
+ * test.
  */
+//#define XT_USE_ROW_REC_MMAP_FILES
 
-//#undef XT_ROW_RWLOCKS
-//#define XT_ROW_RWLOCKS				2
-
-//#undef XT_TAB_MIN_VAR_REC_LENGTH
-//#define XT_TAB_MIN_VAR_REC_LENGTH		20
-
-//#undef XT_ROW_LOCK_COUNT
-//#define XT_ROW_LOCK_COUNT				(XT_ROW_RWLOCKS * 2)
-
-//#undef XT_INDEX_PAGE_SHIFTS
-//#define XT_INDEX_PAGE_SHIFTS			12
-
-//#undef XT_INDEX_WRITE_BUFFER_SIZE
-//#define XT_INDEX_WRITE_BUFFER_SIZE	(40 * 1024)
-
-#endif
-
-/* Define this in order to use memory mapped files: */
-#define XT_USE_ROW_REC_MMAP_FILES
+/* Define this if sequential scan should load data into the 
+ * record cache.
+ *
+ * This is the way InnoDB behaves.
+ */
+#define XT_SEQ_SCAN_LOADS_CACHE
 
 /* Define this in order to use direct I/O on index files: */
 /* NOTE: DO NOT ENABLE!
@@ -326,32 +326,34 @@ typedef struct XTPathStr {
  */
 //#define XT_USE_DIRECT_IO_ON_INDEX
 
-#ifdef XT_USE_ROW_REC_MMAP_FILES
-
-#define XT_SEQ_SCAN_FROM_MEMORY
-#define XT_ROW_REC_FILE_PTR				XTMapFilePtr
-#define XT_PWRITE_RR_FILE				xt_pwrite_fmap
-#define XT_PREAD_RR_FILE				xt_pread_fmap
-#define XT_FLUSH_RR_FILE				xt_flush_fmap
-#define XT_CLOSE_RR_FILE_NS				xt_close_fmap_ns
-
-#else
-
-#define XT_ROW_REC_FILE_PTR				XTOpenFilePtr
-#define XT_PWRITE_RR_FILE				xt_pwrite_file
-#define XT_PREAD_RR_FILE				xt_pread_file
-#define XT_FLUSH_RR_FILE				xt_flush_file
-#define XT_CLOSE_RR_FILE_NS				xt_close_file_ns
+/*
+ * Define this variable if PBXT should do lazy deleting in indexes
+ * Note, even if the variable is not defined, PBXT will handle
+ * lazy deleted items in an index.
+ *
+ * NOTE: This can cause significant degrade of index scan speed.
+ * 25% on sysbench readonly index scan tests.
+ */
+//#define XT_USE_LAZY_DELETE
 
-#endif
+/*
+ * Define this variable if a connection should wait for the
+ * sweeper to clean up previous transactions executed by the
+ * connection, before continuing.
+ *
+ * The number of transactions that the sweeper is aload to
+ * lag can be dynamic, but there is a limit (XT_MAX_XACT_BEHIND)
+ */
+#define XT_WAIT_FOR_CLEANUP
 
-#ifdef XT_SEQ_SCAN_FROM_MEMORY
-#define XT_LOCK_MEMORY_PTR(x, f, a, s, v, c)	do { x = xt_lock_fmap_ptr(f, a, s, v, c); } while (0)
-#define XT_UNLOCK_MEMORY_PTR(f, v)				xt_unlock_fmap_ptr(f, v);
-#else
-#define XT_LOCK_MEMORY_PTR(x, f, a, v, c)
-#define XT_UNLOCK_MEMORY_PTR(f, v)
-#endif
+/*
+ * This seems to be the optimal value, at least according to
+ * sysbench/sysbench run --test=oltp --num-threads=128 --max-requests=50000 --mysql-user=root 
+ * --oltp-table-size=100000 --oltp-table-name=sb_pbxt --mysql-engine-trx=yes
+ *
+ * Using 8, 16 and 128 threads.
+ */
+#define XT_MAX_XACT_BEHIND				2
 
 /* {NO-ACTION-BUG}
  * Define this to implement NO ACTION correctly
@@ -405,6 +407,60 @@ typedef struct XTPathStr {
 #define XT_ADD_PTR(p, l)				((void *) ((char *) (p) + (l)))
 
 /* ----------------------------------------------------------------------
+ * DEFINES DEPENDENT ON  CONSTANTS
+ */
+
+#ifdef XT_USE_ROW_REC_MMAP_FILES
+
+#define XT_ROW_REC_FILE_PTR						XTMapFilePtr
+#define XT_PWRITE_RR_FILE						xt_pwrite_fmap
+#define XT_PREAD_RR_FILE						xt_pread_fmap
+#define XT_FLUSH_RR_FILE						xt_flush_fmap
+#define XT_CLOSE_RR_FILE_NS						xt_close_fmap_ns
+
+#define XT_LOCK_MEMORY_PTR(x, f, a, s, v, c)	do { x = xt_lock_fmap_ptr(f, a, s, v, c); } while (0)
+#define XT_UNLOCK_MEMORY_PTR(f, d, e, v)		do { xt_unlock_fmap_ptr(f, v); d = NULL; } while (0)
+
+#else
+
+#define XT_ROW_REC_FILE_PTR						XTOpenFilePtr
+#define XT_PWRITE_RR_FILE						xt_pwrite_file
+#define XT_PREAD_RR_FILE						xt_pread_file
+#define XT_FLUSH_RR_FILE						xt_flush_file
+#define XT_CLOSE_RR_FILE_NS						xt_close_file_ns
+
+#define XT_LOCK_MEMORY_PTR(x, f, a, s, v, c)	do { if (!xt_lock_file_ptr(f, &x, a, s, v, c)) x = NULL; } while (0)
+#define XT_UNLOCK_MEMORY_PTR(f, d, e, v)		do { if (e) { xt_unlock_file_ptr(f, d, v); d = NULL; } } while (0)
+
+#endif
+
+/* ----------------------------------------------------------------------
+ * DEBUG SIZES!
+ * Reduce the thresholds to make things happen faster.
+ */
+
+#ifdef XT_USE_GLOBAL_DEBUG_SIZES
+
+//#undef XT_ROW_RWLOCKS
+//#define XT_ROW_RWLOCKS				2
+
+//#undef XT_TAB_MIN_VAR_REC_LENGTH
+//#define XT_TAB_MIN_VAR_REC_LENGTH		20
+
+//#undef XT_ROW_LOCK_COUNT
+//#define XT_ROW_LOCK_COUNT				(XT_ROW_RWLOCKS * 2)
+
+//#undef XT_INDEX_PAGE_SHIFTS
+//#define XT_INDEX_PAGE_SHIFTS			8	// 256
+//#undef XT_BLOCK_SIZE_FOR_DIRECT_IO
+//#define XT_BLOCK_SIZE_FOR_DIRECT_IO	256
+
+//#undef XT_INDEX_WRITE_BUFFER_SIZE
+//#define XT_INDEX_WRITE_BUFFER_SIZE	(40 * 1024)
+
+#endif
+
+/* ----------------------------------------------------------------------
  * BYTE ORDER
  */
 
@@ -645,6 +701,14 @@ typedef struct xtIndexNodeID {
 #define XT_XACT_ID_SIZE			4
 #define XT_CHECKSUM4_XACT(x)	(x)
 
+#ifdef XT_WIN
+#define __FUNC__				__FUNCTION__
+#elif defined(XT_SOLARIS)
+#define __FUNC__				"__func__"
+#else
+#define __FUNC__				__PRETTY_FUNCTION__
+#endif
+
 /* ----------------------------------------------------------------------
  * GLOBAL VARIABLES
  */
@@ -669,6 +733,7 @@ extern xtBool				pbxt_crash_debug;
 #define MYSQL_THD							Session *
 #define THR_THD								THR_Session
 #define STRUCT_TABLE						class Table
+#define TABLE_SHARE							TableShare
 
 #define MYSQL_TYPE_STRING					DRIZZLE_TYPE_VARCHAR
 #define MYSQL_TYPE_VARCHAR					DRIZZLE_TYPE_VARCHAR
@@ -687,6 +752,7 @@ extern xtBool				pbxt_crash_debug;
 
 #define mx_tmp_use_all_columns(x, y)		(x)->use_all_columns(y)
 #define mx_tmp_restore_column_map(x, y)		(x)->restore_column_map(y)
+#define MX_BIT_FAST_TEST_AND_SET(x, y)		bitmap_test_and_set(x, y)
 
 #define MX_TABLE_TYPES_T					handler::Table_flags
 #define MX_UINT8_T							uint8_t
@@ -696,6 +762,7 @@ extern xtBool				pbxt_crash_debug;
 #define MX_CHARSET_INFO						struct charset_info_st
 #define MX_CONST_CHARSET_INFO				const struct charset_info_st			
 #define MX_CONST							const
+
 #define my_bool								bool
 #define int16								int16_t
 #define int32								int32_t
@@ -712,6 +779,9 @@ extern xtBool				pbxt_crash_debug;
 
 #define HA_CAN_SQL_HANDLER					0
 #define HA_CAN_INSERT_DELAYED				0
+#define HA_BINLOG_ROW_CAPABLE				0
+#define HA_BINLOG_STMT_CAPABLE				0
+#define HA_CACHE_TBL_TRANSACT				0
 
 #define max									cmax
 #define min									cmin
@@ -734,6 +804,7 @@ extern xtBool				pbxt_crash_debug;
 #define thd_tablespace_op					session_tablespace_op
 #define thd_alloc							session_alloc
 #define thd_make_lex_string					session_make_lex_string
+#define column_bitmaps_signal()
 
 #define my_pthread_setspecific_ptr(T, V)	pthread_setspecific(T, (void*) (V))
 
@@ -750,6 +821,9 @@ extern xtBool				pbxt_crash_debug;
                                    (((uint32_t) (((const unsigned char*) (A))[1])) << 16) +\
                                    (((uint32_t) (((const unsigned char*) (A))[0])) << 24)))
 
+class PBXTStorageEngine;
+typedef PBXTStorageEngine handlerton;
+
 #else // DRIZZLED
 /* The MySQL case: */
 #if MYSQL_VERSION_ID >= 60008
@@ -760,6 +834,7 @@ extern xtBool				pbxt_crash_debug;
 
 #define mx_tmp_use_all_columns				dbug_tmp_use_all_columns
 #define mx_tmp_restore_column_map(x, y)		dbug_tmp_restore_column_map((x)->read_set, y)
+#define MX_BIT_FAST_TEST_AND_SET(x, y)		bitmap_fast_test_and_set(x, y)
 
 #define MX_TABLE_TYPES_T					ulonglong
 #define MX_UINT8_T							uint8
@@ -772,6 +847,11 @@ extern xtBool				pbxt_crash_debug;
 
 #endif // DRIZZLED
 
+#define MX_BITMAP							MY_BITMAP
+#define MX_BIT_SIZE()						n_bits
+#define MX_BIT_IS_SUBSET(x, y)				bitmap_is_subset(x, y)
+#define MX_BIT_SET(x, y)					bitmap_set_bit(x, y)
+
 #ifndef XT_SCAN_CORE_DEFINED
 #define XT_SCAN_CORE_DEFINED
 xtBool	xt_mm_scan_core(void);