branches/innodb+: Merge revisions 2986:3152 from branches/zip:

  ------------------------------------------------------------------------
  r3036 | marko | 2008-11-12 12:34:30 +0200 (Wed, 12 Nov 2008) | 4 lines

  branches/zip: dtuple_validate(): When UNIV_DEBUG_VALGRIND is defined,
  rely solely on the UNIV_MEM_ASSERT_RW() check and disable the for loop
  that would only cause additional noise.
  ------------------------------------------------------------------------
  r3037 | marko | 2008-11-12 13:52:57 +0200 (Wed, 12 Nov 2008) | 6 lines

  branches/zip: row_vers_impl_x_locked_off_kernel(): Remove compilation
  warnings about prev_trx_id and vers_del being possibly uninitialized,
  by handling the case prev_version == NULL in a single if block.

  rb://45 approved by Inaam Rana.
  ------------------------------------------------------------------------
  r3131 | michael | 2008-11-17 14:56:56 +0200 (Mon, 17 Nov 2008) | 9 lines

  branches/zip:

  rb://53

  Improve innodb_supports_xa system variable handling and
  reduces the number of retrievals of the value from MySQL.

  Approved by: Marko, over IM
  ------------------------------------------------------------------------
  r3132 | michael | 2008-11-17 16:02:01 +0200 (Mon, 17 Nov 2008) | 5 lines

  branches/zip: rb://53

  Final version of rb://53, fixes the styling of a comment, makes
  the definition and the declaration of thd_supports_xa() identical commentwise.
  ------------------------------------------------------------------------
  r3141 | marko | 2008-11-19 16:39:55 +0200 (Wed, 19 Nov 2008) | 1 line

  branches/zip: buf_LRU_free_block(): Clarify the function comment.
  ------------------------------------------------------------------------
  r3144 | marko | 2008-11-20 11:39:49 +0200 (Thu, 20 Nov 2008) | 2 lines

  branches/zip: rec_get_nth_field_offs_old(): Add UNIV_UNLIKELY hints
  to assertion-like tests.
  ------------------------------------------------------------------------
  r3145 | marko | 2008-11-20 12:22:40 +0200 (Thu, 20 Nov 2008) | 20 lines

  branches/zip: Always check for "row too large" when executing SQL to create
  an index or table.  We have to skip this check when loading table definitions
  from the data dictionary, because we could otherwise refuse to load old
  tables (even uncompressed ones).  This addresses Issue #119.

  The first "row too large" check was implemented in MySQL 5.0.3
  to address MySQL Bug #5682.  In the InnoDB Plugin 1.0.2, a more
  accurate check was implemented in innodb_strict_mode.  We now
  make the check unconditional.

  dict_create_index_step(): Pass strict=TRUE to dict_index_add_to_cache().

  trx_is_strict(), thd_is_strict(): Remove.

  innodb-zip.test: Test in innodb_strict_mode=OFF.

  innodb_bug36169.test: Ensure that none of the tables can be created.

  rb://56 approved by Sunny Bains.
  ------------------------------------------------------------------------
  r3148 | marko | 2008-11-20 13:27:27 +0200 (Thu, 20 Nov 2008) | 3 lines

  branches/zip: rec_print_old(), rec_print_comp(): Dump each field in a
  separate line, so that the dumps can be read and compared more easily.
  ------------------------------------------------------------------------

buf/buf0lru.c

@@ -1287,9 +1287,15 @@ buf_LRU_make_block_old(
 
 /**********************************************************************
 Try to free a block.  If bpage is a descriptor of a compressed-only
-page, the descriptor object will be freed as well.  If this function
-returns BUF_LRU_FREED, it will not temporarily release
-buf_pool_mutex. */
+page, the descriptor object will be freed as well.
+
+NOTE: If this function returns BUF_LRU_FREED, it will not temporarily
+release buf_pool_mutex.  Furthermore, the page frame will no longer be
+accessible via bpage.
+
+The caller must hold buf_pool_mutex and buf_page_get_mutex(bpage) and
+release these two mutexes after the call.  No other
+buf_page_get_mutex() may be held when calling this function. */
 UNIV_INTERN
 enum buf_lru_free_block_status
 buf_LRU_free_block(

data/data0data.c

@@ -26,8 +26,10 @@ Created 5/30/1994 Heikki Tuuri
 for error checking */
 UNIV_INTERN byte	data_error;
 
+# ifndef UNIV_DEBUG_VALGRIND
 /* this is used to fool the compiler in dtuple_validate */
 UNIV_INTERN ulint	data_dummy;
+# endif /* !UNIV_DEBUG_VALGRIND */
 #endif /* UNIV_DEBUG */
 
 /*************************************************************************
@@ -232,11 +234,9 @@ dtuple_validate(
 	const dtuple_t*	tuple)	/* in: tuple */
 {
 	const dfield_t*	field;
-	const byte*	data;
 	ulint		n_fields;
 	ulint		len;
 	ulint		i;
-	ulint		j;
 
 	ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
 
@@ -252,8 +252,9 @@ dtuple_validate(
 
 		if (!dfield_is_null(field)) {
 
-			data = dfield_get_data(field);
-			UNIV_MEM_ASSERT_RW(data, len);
+			const byte*	data = dfield_get_data(field);
+#ifndef UNIV_DEBUG_VALGRIND
+			ulint		j;
 
 			for (j = 0; j < len; j++) {
 
@@ -262,6 +263,9 @@ dtuple_validate(
 						      code */
 				data++;
 			}
+#endif /* !UNIV_DEBUG_VALGRIND */
+
+			UNIV_MEM_ASSERT_RW(data, len);
 		}
 	}
 

dict/dict0crea.c

@@ -1079,7 +1079,7 @@ dict_create_index_step(
 		dulint	index_id = node->index->id;
 
 		err = dict_index_add_to_cache(node->table, node->index,
-					      FIL_NULL, trx_is_strict(trx));
+					      FIL_NULL, TRUE);
 
 		node->index = dict_index_get_if_in_cache_low(index_id);
 		ut_a(!node->index == (err != DB_SUCCESS));

handler/ha_innodb.cc

@@ -608,15 +608,17 @@ thd_is_select(
 }
 
 /**********************************************************************
-Returns true if the thread is executing in innodb_strict_mode. */
+Returns true if the thread supports XA,
+global value of innodb_supports_xa if thd is NULL. */
 extern "C" UNIV_INTERN
 ibool
-thd_is_strict(
-/*==========*/
-			/* out: true if thd is in strict mode */
-	void*	thd)	/* in: thread handle (THD*) */
+thd_supports_xa(
+/*============*/
+			/* out: true if thd has XA support */
+	void*	thd)	/* in: thread handle (THD*), or NULL to query
+			the global innodb_supports_xa */
 {
-	return(THDVAR((THD*) thd, strict_mode));
+	return(THDVAR((THD*) thd, support_xa));
 }
 
 /**********************************************************************
@@ -1238,9 +1240,6 @@ check_trx_exists(
 		trx->mysql_thd = thd;
 		trx->mysql_query_str = thd_query(thd);
 
-		/* Update the info whether we should skip XA steps that eat
-		CPU time */
-		trx->support_xa = THDVAR(thd, support_xa);
 	} else {
 		if (trx->magic_n != TRX_MAGIC_N) {
 			mem_analyze_corruption(trx);
@@ -2299,9 +2298,6 @@ innobase_commit(
 
 	trx = check_trx_exists(thd);
 
-	/* Update the info whether we should skip XA steps that eat CPU time */
-	trx->support_xa = THDVAR(thd, support_xa);
-
 	/* Since we will reserve the kernel mutex, we have to release
 	the search system latch first to obey the latching order. */
 
@@ -2428,9 +2424,6 @@ innobase_rollback(
 
 	trx = check_trx_exists(thd);
 
-	/* Update the info whether we should skip XA steps that eat CPU time */
-	trx->support_xa = THDVAR(thd, support_xa);
-
 	/* Release a possible FIFO ticket and search latch. Since we will
 	reserve the kernel mutex, we have to release the search system latch
 	first to obey the latching order. */
@@ -8843,7 +8836,10 @@ innobase_xa_prepare(
 		trx->active_trans = 2;
 	}
 
-	if (!THDVAR(thd, support_xa)) {
+	/* we use support_xa value as it was seen at transaction start
+	time, not the current session variable value. Any possible changes
+	to the session variable take effect only in the next transaction */
+	if (!trx->support_xa) {
 
 		return(0);
 	}

include/buf0lru.h

@@ -87,9 +87,15 @@ buf_LRU_insert_zip_clean(
 
 /**********************************************************************
 Try to free a block.  If bpage is a descriptor of a compressed-only
-page, the descriptor object will be freed as well.  If this function
-returns BUF_LRU_FREED, it will not temporarily release
-buf_pool_mutex. */
+page, the descriptor object will be freed as well.
+
+NOTE: If this function returns BUF_LRU_FREED, it will not temporarily
+release buf_pool_mutex.  Furthermore, the page frame will no longer be
+accessible via bpage.
+
+The caller must hold buf_pool_mutex and buf_page_get_mutex(bpage) and
+release these two mutexes after the call.  No other
+buf_page_get_mutex() may be held when calling this function. */
 UNIV_INTERN
 enum buf_lru_free_block_status
 buf_LRU_free_block(

include/ha_prototypes.h

@@ -207,13 +207,15 @@ innobase_get_charset(
 	void*	mysql_thd);	/* in: MySQL thread handle */
 
 /**********************************************************************
-Returns true if the thread is executing in innodb_strict_mode. */
+Returns true if the thread supports XA,
+global value of innodb_supports_xa if thd is NULL. */
 
 ibool
-thd_is_strict(
-/*==========*/
-			/* out: true if thd is in strict mode */
-	void*	thd);	/* in: thread handle (THD*) */
+thd_supports_xa(
+/*============*/
+			/* out: true if thd supports XA */
+	void*	thd);	/* in: thread handle (THD*), or NULL to query
+			the global innodb_supports_xa */
 
 /**********************************************************************
 Returns the lock wait timeout for the current connection. */

include/trx0trx.h

@@ -407,15 +407,6 @@ trx_is_interrupted(
 #define trx_is_interrupted(trx) FALSE
 #endif /* !UNIV_HOTBACKUP */
 
-/**************************************************************************
-Determines if the currently running transaction is in innodb_strict_mode. */
-UNIV_INTERN
-ibool
-trx_is_strict(
-/*==========*/
-			/* out: TRUE if strict */
-	trx_t*	trx);	/* in: transaction */
-
 /***********************************************************************
 Calculates the "weight" of a transaction. The weight of one transaction
 is estimated as the number of altered rows + the number of locked rows.

mysql-test/innodb-zip.result

@@ -122,7 +122,7 @@ table_schema	table_name	row_format
 test	t1	Compressed
 test	t2	Compact
 drop table t1,t2;
-SET SESSION innodb_strict_mode = on;
+SET SESSION innodb_strict_mode = off;
 CREATE TABLE t1(
 c TEXT NOT NULL, d TEXT NOT NULL,
 PRIMARY KEY (c(767),d(767)))

mysql-test/innodb-zip.test

@@ -84,7 +84,8 @@ SELECT table_schema, table_name, row_format
 FROM information_schema.tables WHERE engine='innodb';
 drop table t1,t2;
 
-SET SESSION innodb_strict_mode = on;
+# The following should fail even in non-strict mode.
+SET SESSION innodb_strict_mode = off;
 --error ER_TOO_BIG_ROWSIZE
 CREATE TABLE t1(
 	c TEXT NOT NULL, d TEXT NOT NULL,

mysql-test/innodb_bug36169.test

@@ -22,6 +22,7 @@ SET GLOBAL innodb_file_per_table=ON;
 # Generating 10 tables
 # Creating a table with 94 columns and 24 indexes
 DROP TABLE IF EXISTS `table0`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table0`
 (`col0` BOOL,
 `col1` BOOL,
@@ -145,6 +146,7 @@ KEY `idx23` (`col7`(163),`col31`,`col71`,`col14`)
 
 # Creating a table with 10 columns and 32 indexes
 DROP TABLE IF EXISTS `table1`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table1`
 (`col0` CHAR (113),
 `col1` FLOAT,
@@ -192,6 +194,7 @@ KEY `idx31` (`col4`(1),`col0`)
 
 # Creating a table with 141 columns and 18 indexes
 DROP TABLE IF EXISTS `table2`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table2`
 (`col0` BOOL,
 `col1` MEDIUMINT,
@@ -356,6 +359,7 @@ KEY `idx17` (`col24`(250),`col7`,`col92`,`col45`)
 
 # Creating a table with 199 columns and 1 indexes
 DROP TABLE IF EXISTS `table3`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table3`
 (`col0` SMALLINT,
 `col1` SET ('test1','test2','test3'),
@@ -561,6 +565,7 @@ KEY `idx0` (`col39`,`col23`)
 
 # Creating a table with 133 columns and 16 indexes
 DROP TABLE IF EXISTS `table4`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table4`
 (`col0` VARCHAR (60),
 `col1` NUMERIC,
@@ -715,6 +720,7 @@ KEY `idx15` (`col4`(246),`col130`,`col115`,`col3`(141))
 
 # Creating a table with 176 columns and 13 indexes
 DROP TABLE IF EXISTS `table5`;
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table5`
 (`col0` MEDIUMTEXT,
 `col1` VARCHAR (90),
@@ -910,6 +916,7 @@ KEY `idx12` (`col24`)
 # Creating a table with 179 columns and 46 indexes
 DROP TABLE IF EXISTS `table6`;
 -- error ER_TOO_BIG_ROWSIZE
+--error ER_TOO_BIG_ROWSIZE
 CREATE TABLE IF NOT EXISTS `table6`
 (`col0` ENUM ('test1','test2','test3'),
 `col1` MEDIUMBLOB,

rem/rem0rec.c

@@ -681,13 +681,13 @@ rec_get_nth_field_offs_old(
 	ut_ad(rec && len);
 	ut_ad(n < rec_get_n_fields_old(rec));
 
-	if (n > REC_MAX_N_FIELDS) {
+	if (UNIV_UNLIKELY(n > REC_MAX_N_FIELDS)) {
 		fprintf(stderr, "Error: trying to access field %lu in rec\n",
 			(ulong) n);
 		ut_error;
 	}
 
-	if (rec == NULL) {
+	if (UNIV_UNLIKELY(rec == NULL)) {
 		fputs("Error: rec is NULL pointer\n", stderr);
 		ut_error;
 	}
@@ -1599,11 +1599,11 @@ rec_print_old(
 			fprintf(file, " SQL NULL, size %lu ",
 				rec_get_nth_field_size(rec, i));
 		}
+
 		putc(';', file);
+		putc('\n', file);
 	}
 
-	putc('\n', file);
-
 	rec_validate_old(rec);
 }
 
@@ -1642,9 +1642,8 @@ rec_print_comp(
 			fputs(" SQL NULL", file);
 		}
 		putc(';', file);
+		putc('\n', file);
 	}
-
-	putc('\n', file);
 }
 
 /*******************************************************************

row/row0vers.c

@@ -48,16 +48,13 @@ row_vers_impl_x_locked_off_kernel(
 	rec_t*		clust_rec;
 	ulint*		clust_offsets;
 	rec_t*		version;
-	rec_t*		prev_version;
 	dulint		trx_id;
-	dulint		prev_trx_id;
 	mem_heap_t*	heap;
 	mem_heap_t*	heap2;
 	dtuple_t*	row;
 	dtuple_t*	entry	= NULL; /* assignment to eliminate compiler
 					warning */
 	trx_t*		trx;
-	ulint		vers_del;
 	ulint		rec_del;
 	ulint		err;
 	mtr_t		mtr;
@@ -141,6 +138,11 @@ row_vers_impl_x_locked_off_kernel(
 	version = clust_rec;
 
 	for (;;) {
+		rec_t*		prev_version;
+		ulint		vers_del;
+		row_ext_t*	ext;
+		dulint		prev_trx_id;
+
 		mutex_exit(&kernel_mutex);
 
 		/* While we retrieve an earlier version of clust_rec, we
@@ -157,66 +159,63 @@ row_vers_impl_x_locked_off_kernel(
 						  heap, &prev_version);
 		mem_heap_free(heap2); /* free version and clust_offsets */
 
-		if (prev_version) {
-			row_ext_t*	ext;
+		if (prev_version == NULL) {
+			mutex_enter(&kernel_mutex);
+
+			if (!trx_is_active(trx_id)) {
+				/* Transaction no longer active: no
+				implicit x-lock */
 
-			clust_offsets = rec_get_offsets(
-				prev_version, clust_index, NULL,
-				ULINT_UNDEFINED, &heap);
-
-			vers_del = rec_get_deleted_flag(prev_version,
-							comp);
-			prev_trx_id = row_get_rec_trx_id(prev_version,
-							 clust_index,
-						 	 clust_offsets);
-
-			/* If the trx_id and prev_trx_id are
-			different and if the prev_version is marked
-			deleted then the prev_trx_id must have
-			already committed for the trx_id to be able to
-			modify the row. Therefore, prev_trx_id cannot
-			hold any implicit lock. */
-			if (0 != ut_dulint_cmp(trx_id, prev_trx_id)
-			    && vers_del) {
-	
-				mutex_enter(&kernel_mutex);
 				break;
 			}
 
-			/* The stack of versions is locked by mtr.
-			Thus, it is safe to fetch the prefixes for
-			externally stored columns. */
-			row = row_build(ROW_COPY_POINTERS, clust_index,
-					prev_version, clust_offsets,
-					NULL, &ext, heap);
-			entry = row_build_index_entry(row, ext, index, heap);
-			/* entry may be NULL if a record was inserted
-			in place of a deleted record, and the BLOB
-			pointers of the new record were not
-			initialized yet.  But in that case,
-			prev_version should be NULL. */
-			ut_a(entry);
-		}
+			/* If the transaction is still active,
+			clust_rec must be a fresh insert, because no
+			previous version was found. */
+			ut_ad(err == DB_SUCCESS);
 
-		mutex_enter(&kernel_mutex);
+			/* It was a freshly inserted version: there is an
+			implicit x-lock on rec */
 
-		if (!trx_is_active(trx_id)) {
-			/* Transaction no longer active: no implicit x-lock */
+			trx = trx_get_on_id(trx_id);
 
 			break;
 		}
 
-		/* If the transaction is still active, the previous version
-		of clust_rec must be accessible if not a fresh insert; we
-		may assert the following: */
+		clust_offsets = rec_get_offsets(prev_version, clust_index,
+						NULL, ULINT_UNDEFINED, &heap);
 
-		ut_ad(err == DB_SUCCESS);
+		vers_del = rec_get_deleted_flag(prev_version, comp);
+		prev_trx_id = row_get_rec_trx_id(prev_version, clust_index,
+						 clust_offsets);
 
-		if (prev_version == NULL) {
-			/* It was a freshly inserted version: there is an
-			implicit x-lock on rec */
+		/* If the trx_id and prev_trx_id are different and if
+		the prev_version is marked deleted then the
+		prev_trx_id must have already committed for the trx_id
+		to be able to modify the row. Therefore, prev_trx_id
+		cannot hold any implicit lock. */
+		if (vers_del && 0 != ut_dulint_cmp(trx_id, prev_trx_id)) {
 
-			trx = trx_get_on_id(trx_id);
+			mutex_enter(&kernel_mutex);
+			break;
+		}
+
+		/* The stack of versions is locked by mtr.  Thus, it
+		is safe to fetch the prefixes for externally stored
+		columns. */
+		row = row_build(ROW_COPY_POINTERS, clust_index, prev_version,
+				clust_offsets, NULL, &ext, heap);
+		entry = row_build_index_entry(row, ext, index, heap);
+		/* entry may be NULL if a record was inserted in place
+		of a deleted record, and the BLOB pointers of the new
+		record were not initialized yet.  But in that case,
+		prev_version should be NULL. */
+		ut_a(entry);
+
+		mutex_enter(&kernel_mutex);
+
+		if (!trx_is_active(trx_id)) {
+			/* Transaction no longer active: no implicit x-lock */
 
 			break;
 		}
@@ -226,6 +225,11 @@ row_vers_impl_x_locked_off_kernel(
 		if prev_version would require rec to be in a different
 		state. */
 
+		/* The previous version of clust_rec must be
+		accessible, because the transaction is still active
+		and clust_rec was not a fresh insert. */
+		ut_ad(err == DB_SUCCESS);
+
 		/* We check if entry and rec are identified in the alphabetical
 		ordering */
 		if (0 == cmp_dtuple_rec(entry, rec, offsets)) {

trx/trx0trx.c

@@ -34,22 +34,6 @@ UNIV_INTERN sess_t*		trx_dummy_sess = NULL;
 the kernel mutex */
 UNIV_INTERN ulint	trx_n_mysql_transactions = 0;
 
-/**************************************************************************
-Determines if the currently running transaction is in innodb_strict_mode. */
-UNIV_INTERN
-ibool
-trx_is_strict(
-/*==========*/
-			/* out: TRUE if strict */
-	trx_t*	trx)	/* in: transaction */
-{
-#ifndef UNIV_HOTBACKUP
-	return(trx && trx->mysql_thd && thd_is_strict(trx->mysql_thd));
-#else /* UNIV_HOTBACKUP */
-	return(FALSE);
-#endif /* UNIV_HOTBACKUP */
-}
-
 /*****************************************************************
 Set detailed error message for the transaction. */
 UNIV_INTERN
@@ -694,6 +678,14 @@ trx_start(
 {
 	ibool	ret;
 
+	/* Update the info whether we should skip XA steps that eat CPU time
+	For the duration of the transaction trx->support_xa is not reread
+	from thd so any changes in the value take effect in the next
+	transaction. This is to avoid a scenario where some undo
+	generated by a transaction, has XA stuff, and other undo,
+	generated by the same transaction, doesn't. */
+	trx->support_xa = thd_supports_xa(trx->mysql_thd);
+
 	mutex_enter(&kernel_mutex);
 
 	ret = trx_start_low(trx, rseg_id);