Automatic merge with MariaDB 5.1

14 files changed, 152 insertions, 55 deletions

diff --git a/client/mysqltest.cc b/client/mysqltest.cc
index 4cc7ef813c3..be4a3e06fbb 100644
--- a/client/mysqltest.cc
+++ b/client/mysqltest.cc
@@ -1255,15 +1255,6 @@ void die(const char *fmt, ...)
   DBUG_ENTER("die");
   DBUG_PRINT("enter", ("start_lineno: %d", start_lineno));
 
-  /*
-    Protect against dying twice
-    first time 'die' is called, try to write log files
-    second time, just exit
-  */
-  if (dying)
-    cleanup_and_exit(1);
-  dying= 1;
-
   /* Print the error message */
   fprintf(stderr, "mysqltest: ");
   if (cur_file && cur_file != file_stack)
@@ -1282,6 +1273,15 @@ void die(const char *fmt, ...)
   fprintf(stderr, "\n");
   fflush(stderr);
 
+  /*
+    Protect against dying twice
+    first time 'die' is called, try to write log files
+    second time, just exit
+  */
+  if (dying)
+    cleanup_and_exit(1);
+  dying= 1;
+
   log_file.show_tail(opt_tail_lines);
 
   /*
diff --git a/mysql-test/r/warnings_debug.result b/mysql-test/r/warnings_debug.result
new file mode 100644
index 00000000000..08908bf0437
--- /dev/null
+++ b/mysql-test/r/warnings_debug.result
@@ -0,0 +1,10 @@
+drop table if exists t1;
+create table t1 (a int primary key) engine=innodb;
+SET SESSION debug="+d,warn_during_ha_commit_trans";
+INSERT INTO t1 VALUES (1);
+Warnings:
+Warning	1196	Some non-transactional changed tables couldn't be rolled back
+SHOW WARNINGS;
+Level	Code	Message
+Warning	1196	Some non-transactional changed tables couldn't be rolled back
+drop table t1;
diff --git a/mysql-test/t/warnings_debug.test b/mysql-test/t/warnings_debug.test
new file mode 100644
index 00000000000..99d02330960
--- /dev/null
+++ b/mysql-test/t/warnings_debug.test
@@ -0,0 +1,19 @@
+--source include/have_innodb.inc
+--source include/have_debug.inc
+
+--disable_warnings
+drop table if exists t1;
+--enable_warnings
+
+create table t1 (a int primary key) engine=innodb;
+
+# Test that warnings produced during autocommit (after calling
+# set_ok_status()) are still reported to the client.
+SET SESSION debug="+d,warn_during_ha_commit_trans";
+INSERT INTO t1 VALUES (1);
+# The warning will be shown automatically by mysqltest; there was a bug where
+# this didn't happen because the warning was not counted when sending result
+# packet. Show the warnings manually also.
+SHOW WARNINGS;
+
+drop table t1;
diff --git a/sql/handler.cc b/sql/handler.cc
index 29aabda1343..48add3a1e53 100644
--- a/sql/handler.cc
+++ b/sql/handler.cc
@@ -1094,6 +1094,12 @@ int ha_commit_trans(THD *thd, bool all)
   my_xid xid= thd->transaction.xid_state.xid.get_my_xid();
   DBUG_ENTER("ha_commit_trans");
 
+  /* Just a random warning to test warnings pushed during autocommit. */
+  DBUG_EXECUTE_IF("warn_during_ha_commit_trans",
+    push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+                 ER_WARNING_NOT_COMPLETE_ROLLBACK,
+                 ER(ER_WARNING_NOT_COMPLETE_ROLLBACK)););
+
   /*
     We must not commit the normal transaction if a statement
     transaction is pending. Otherwise statement transaction
diff --git a/sql/protocol.cc b/sql/protocol.cc
index 786d6ef8544..8a9d712077d 100644
--- a/sql/protocol.cc
+++ b/sql/protocol.cc
@@ -203,7 +203,7 @@ net_send_ok(THD *thd,
   NET *net= &thd->net;
   uchar buff[MYSQL_ERRMSG_SIZE+10],*pos;
   bool error= FALSE;
-  DBUG_ENTER("my_ok");
+  DBUG_ENTER("net_send_ok");
 
   if (! net->vio)	// hack for re-parsing queries
   {
diff --git a/sql/sql_class.h b/sql/sql_class.h
index 4682938fd6d..bb986d6377f 100644
--- a/sql/sql_class.h
+++ b/sql/sql_class.h
@@ -1224,6 +1224,13 @@ public:
     return m_total_warn_count;
   }
 
+  /* Used to count any warnings pushed after calling set_ok_status(). */
+  void increment_warning()
+  {
+    if (m_status != DA_EMPTY)
+      m_total_warn_count++;
+  }
+
   Diagnostics_area() { reset_diagnostics_area(); }
 
 private:
diff --git a/sql/sql_error.cc b/sql/sql_error.cc
index 9ea7facbe41..835e60cd6ba 100644
--- a/sql/sql_error.cc
+++ b/sql/sql_error.cc
@@ -159,6 +159,8 @@ MYSQL_ERROR *push_warning(THD *thd, MYSQL_ERROR::enum_warning_level level,
   }
   thd->warn_count[(uint) level]++;
   thd->total_warn_count++;
+  /* Make sure we also count warnings pushed after calling set_ok_status(). */
+  thd->main_da.increment_warning();
   DBUG_RETURN(err);
 }
 
diff --git a/storage/maria/ma_blockrec.c b/storage/maria/ma_blockrec.c
index 0eab7c62796..fd02e2ac0ec 100644
--- a/storage/maria/ma_blockrec.c
+++ b/storage/maria/ma_blockrec.c
@@ -1993,19 +1993,6 @@ static my_bool write_tail(MARIA_HA *info,
   /* Keep BLOCKUSED_USE_ORG_BITMAP */
   block->used|= BLOCKUSED_USED | BLOCKUSED_TAIL;
 
-  /* Increase data file size, if extended */
-  position= (my_off_t) block->page * block_size;
-  if (share->state.state.data_file_length <= position)
-  {
-    /*
-      We are modifying a state member before writing the UNDO; this is a WAL
-      violation. But for data_file_length this is ok, as long as we change
-      data_file_length after writing any log record (FILE_ID/REDO/UNDO) (see
-      collect_tables()).
-    */
-    _ma_set_share_data_file_length(share, position + block_size);
-  }
-
   if (block_is_read)
   {
     /* Current page link is last element in pinned_pages */
@@ -2021,17 +2008,33 @@ static my_bool write_tail(MARIA_HA *info,
     page_link->unlock= PAGECACHE_LOCK_READ_UNLOCK;
     res= 0;
   }
-  else if (!(res= pagecache_write(share->pagecache,
-                                  &info->dfile, block->page, 0,
-                                  row_pos.buff,share->page_type,
-                                  PAGECACHE_LOCK_READ,
-                                  PAGECACHE_PIN,
-                                  PAGECACHE_WRITE_DELAY, &page_link.link,
-                                  LSN_IMPOSSIBLE)))
+  else
   {
-    page_link.unlock= PAGECACHE_LOCK_READ_UNLOCK;
-    page_link.changed= 1;
-    push_dynamic(&info->pinned_pages, (void*) &page_link);
+    if (!(res= pagecache_write(share->pagecache,
+                               &info->dfile, block->page, 0,
+                               row_pos.buff,share->page_type,
+                               PAGECACHE_LOCK_READ,
+                               PAGECACHE_PIN,
+                               PAGECACHE_WRITE_DELAY, &page_link.link,
+                               LSN_IMPOSSIBLE)))
+    {
+      page_link.unlock= PAGECACHE_LOCK_READ_UNLOCK;
+      page_link.changed= 1;
+      push_dynamic(&info->pinned_pages, (void*) &page_link);
+    }
+
+    /* Increase data file size, if extended */
+    position= (my_off_t) block->page * block_size;
+    if (share->state.state.data_file_length <= position)
+    {
+      /*
+        We are modifying a state member before writing the UNDO; this is a WAL
+        violation. But for data_file_length this is ok, as long as we change
+        data_file_length after writing any log record (FILE_ID/REDO/UNDO) (see
+        collect_tables()).
+      */
+      _ma_set_share_data_file_length(share, position + block_size);
+    }
   }
   DBUG_RETURN(res);
 }
@@ -2068,7 +2071,7 @@ static my_bool write_full_pages(MARIA_HA *info,
   uint data_size= FULL_PAGE_SIZE(block_size);
   uchar *buff= info->keyread_buff;
   uint page_count, sub_blocks;
-  my_off_t position;
+  my_off_t position, max_position;
   DBUG_ENTER("write_full_pages");
   DBUG_PRINT("enter", ("length: %lu  page: %lu  page_count: %lu",
                        (ulong) length, (ulong) block->page,
@@ -2080,9 +2083,7 @@ static my_bool write_full_pages(MARIA_HA *info,
   page_count= block->page_count;
   sub_blocks= block->sub_blocks;
 
-  position= (my_off_t) (page + page_count) * block_size;
-  if (share->state.state.data_file_length < position)
-    _ma_set_share_data_file_length(share, position);
+  max_position= (my_off_t) (page + page_count) * block_size;
 
   /* Increase data file size, if extended */
 
@@ -2105,8 +2106,7 @@ static my_bool write_full_pages(MARIA_HA *info,
                           (ulong) block->page, (ulong) block->page_count));
 
       position= (page + page_count + 1) * block_size;
-      if (share->state.state.data_file_length < position)
-        _ma_set_share_data_file_length(share, position);
+      set_if_bigger(max_position, position);
     }
     lsn_store(buff, lsn);
     buff[PAGE_TYPE_OFFSET]= (uchar) BLOB_PAGE;
@@ -2134,6 +2134,8 @@ static my_bool write_full_pages(MARIA_HA *info,
     page++;
     DBUG_ASSERT(block->used & BLOCKUSED_USED);
   }
+  if (share->state.state.data_file_length < max_position)
+    _ma_set_share_data_file_length(share, max_position);
   DBUG_RETURN(0);
 }
 
@@ -3121,11 +3123,6 @@ static my_bool write_block_record(MARIA_HA *info,
   }
 #endif
 
-  /* Increase data file size, if extended */
-  position= (my_off_t) head_block->page * block_size;
-  if (share->state.state.data_file_length <= position)
-    _ma_set_share_data_file_length(share, position + block_size);
-
   if (head_block_is_read)
   {
     MARIA_PINNED_PAGE *page_link;
@@ -3154,6 +3151,11 @@ static my_bool write_block_record(MARIA_HA *info,
     page_link.unlock= PAGECACHE_LOCK_READ_UNLOCK;
     page_link.changed= 1;
     push_dynamic(&info->pinned_pages, (void*) &page_link);
+
+    /* Increase data file size, if extended */
+    position= (my_off_t) head_block->page * block_size;
+    if (share->state.state.data_file_length <= position)
+      _ma_set_share_data_file_length(share, position + block_size);
   }
 
   if (share->now_transactional && (tmp_data_used || blob_full_pages_exists))
@@ -5162,6 +5164,7 @@ my_bool _ma_scan_init_block_record(MARIA_HA *info)
   info->scan.number_of_rows= 0;
   info->scan.bitmap_pos= info->scan.bitmap_end;
   info->scan.bitmap_page= (pgcache_page_no_t) 0 - share->bitmap.pages_covered;
+  info->scan.max_page= share->state.state.data_file_length / share->block_size;
   /*
     We need to flush what's in memory (bitmap.map) to page cache otherwise, as
     we are going to read bitmaps from page cache in table scan (see
@@ -5363,6 +5366,11 @@ restart_bitmap_scan:
           page= (info->scan.bitmap_page + 1 +
                  (data - info->scan.bitmap_buff) / 6 * 16 + bit_pos - 1);
           info->scan.row_base_page= ma_recordpos(page, 0);
+          if (page >= info->scan.max_page)
+          {
+            DBUG_PRINT("info", ("Found end of file"));
+            DBUG_RETURN((my_errno= HA_ERR_END_OF_FILE));
+          }
           if (!(pagecache_read(share->pagecache,
                                &info->dfile,
                                page, 0, info->scan.page_buff,
diff --git a/storage/maria/maria_def.h b/storage/maria/maria_def.h
index bdc4d4c5ed2..b998ac4f3a0 100644
--- a/storage/maria/maria_def.h
+++ b/storage/maria/maria_def.h
@@ -476,7 +476,7 @@ typedef struct st_maria_block_scan
 {
   uchar *bitmap_buff, *bitmap_pos, *bitmap_end, *page_buff;
   uchar *dir, *dir_end;
-  pgcache_page_no_t bitmap_page;
+  pgcache_page_no_t bitmap_page, max_page;
   ulonglong bits;
   uint number_of_rows, bit_pos;
   MARIA_RECORD_POS row_base_page;
diff --git a/storage/xtradb/include/os0sync.h b/storage/xtradb/include/os0sync.h
index 0c22162b900..c230a03b6db 100644
--- a/storage/xtradb/include/os0sync.h
+++ b/storage/xtradb/include/os0sync.h
@@ -189,7 +189,7 @@ os_event_wait_low(
 
 /**********************************************************//**
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. In Unix the timeout is always infinite.
+a timeout is exceeded.
 @return	0 if success, OS_SYNC_TIME_EXCEEDED if timeout was exceeded */
 UNIV_INTERN
 ulint
diff --git a/storage/xtradb/include/srv0srv.h b/storage/xtradb/include/srv0srv.h
index dc455581350..8c64d5cee71 100644
--- a/storage/xtradb/include/srv0srv.h
+++ b/storage/xtradb/include/srv0srv.h
@@ -57,6 +57,9 @@ extern const char	srv_mysql50_table_name_prefix[9];
 thread starts running */
 extern os_event_t	srv_lock_timeout_thread_event;
 
+/* This event is set to tell the purge thread to shut down */
+extern os_event_t	srv_purge_thread_event;
+
 /* If the last data file is auto-extended, we add this many pages to it
 at a time */
 #define SRV_AUTO_EXTEND_INCREMENT	\
diff --git a/storage/xtradb/log/log0log.c b/storage/xtradb/log/log0log.c
index fade31037b5..b9f19aeff31 100644
--- a/storage/xtradb/log/log0log.c
+++ b/storage/xtradb/log/log0log.c
@@ -3102,6 +3102,7 @@ logs_empty_and_mark_files_at_shutdown(void)
 	algorithm only works if the server is idle at shutdown */
 
 	srv_shutdown_state = SRV_SHUTDOWN_CLEANUP;
+	os_event_set(srv_purge_thread_event);
 loop:
 	os_thread_sleep(100000);
 
diff --git a/storage/xtradb/os/os0sync.c b/storage/xtradb/os/os0sync.c
index 60467242e14..f9ab58c2ee4 100644
--- a/storage/xtradb/os/os0sync.c
+++ b/storage/xtradb/os/os0sync.c
@@ -31,6 +31,9 @@ Created 9/6/1995 Heikki Tuuri
 
 #ifdef __WIN__
 #include <windows.h>
+#else
+#include <sys/time.h>
+#include <time.h>
 #endif
 
 #include "ut0mem.h"
@@ -407,14 +410,14 @@ os_event_wait_low(
 
 /**********************************************************//**
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. In Unix the timeout is always infinite.
+a timeout is exceeded.
 @return	0 if success, OS_SYNC_TIME_EXCEEDED if timeout was exceeded */
 UNIV_INTERN
 ulint
 os_event_wait_time(
 /*===============*/
 	os_event_t	event,	/*!< in: event to wait */
-	ulint		time)	/*!< in: timeout in microseconds, or
+	ulint		wtime)	/*!< in: timeout in microseconds, or
 				OS_SYNC_INFINITE_TIME */
 {
 #ifdef __WIN__
@@ -422,8 +425,8 @@ os_event_wait_time(
 
 	ut_a(event);
 
-	if (time != OS_SYNC_INFINITE_TIME) {
-		err = WaitForSingleObject(event->handle, (DWORD) time / 1000);
+	if (wtime != OS_SYNC_INFINITE_TIME) {
+		err = WaitForSingleObject(event->handle, (DWORD) wtime / 1000);
 	} else {
 		err = WaitForSingleObject(event->handle, INFINITE);
 	}
@@ -439,13 +442,47 @@ os_event_wait_time(
 		return(1000000); /* dummy value to eliminate compiler warn. */
 	}
 #else
-	UT_NOT_USED(time);
+	int	err;
+	int	ret = 0;
+	ulint	tmp;
+	ib_int64_t	old_count;
+	struct timeval tv_start;
+	struct timespec timeout;
+
+	if (wtime == OS_SYNC_INFINITE_TIME) {
+		os_event_wait(event);
+		return 0;
+	}
+
+	/* Compute the absolute point in time at which to time out. */
+	gettimeofday(&tv_start, NULL);
+	tmp = tv_start.tv_usec + wtime;
+	timeout.tv_sec = tv_start.tv_sec + (tmp / 1000000);
+	timeout.tv_nsec = (tmp % 1000000) * 1000;
+
+	os_fast_mutex_lock(&(event->os_mutex));
+	old_count = event->signal_count;
+
+	for (;;) {
+		if (event->is_set == TRUE || event->signal_count != old_count)
+			break;
+
+		err = pthread_cond_timedwait(&(event->cond_var),
+					     &(event->os_mutex), &timeout);
+		if (err == ETIMEDOUT) {
+			ret = OS_SYNC_TIME_EXCEEDED;
+			break;
+		}
+	}
 
-	/* In Posix this is just an ordinary, infinite wait */
+	os_fast_mutex_unlock(&(event->os_mutex));
 
-	os_event_wait(event);
+	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
+
+		os_thread_exit(NULL);
+	}
 
-	return(0);
+	return ret;
 #endif
 }
 
diff --git a/storage/xtradb/srv/srv0srv.c b/storage/xtradb/srv/srv0srv.c
index c1d0f255c64..43799aab196 100644
--- a/storage/xtradb/srv/srv0srv.c
+++ b/storage/xtradb/srv/srv0srv.c
@@ -704,6 +704,8 @@ UNIV_INTERN srv_slot_t*	srv_mysql_table = NULL;
 
 UNIV_INTERN os_event_t	srv_lock_timeout_thread_event;
 
+UNIV_INTERN os_event_t	srv_purge_thread_event;
+
 UNIV_INTERN srv_sys_t*	srv_sys	= NULL;
 
 /* padding to prevent other memory update hotspots from residing on
@@ -1009,6 +1011,7 @@ srv_init(void)
 	}
 
 	srv_lock_timeout_thread_event = os_event_create(NULL);
+	srv_purge_thread_event = os_event_create(NULL);
 
 	for (i = 0; i < SRV_MASTER + 1; i++) {
 		srv_n_threads_active[i] = 0;
@@ -3337,9 +3340,10 @@ loop:
 		mutex_exit(&kernel_mutex);
 
 		sleep_ms = 10;
+		os_event_reset(srv_purge_thread_event);
 	}
 
-	os_thread_sleep( sleep_ms * 1000 );
+	os_event_wait_time(srv_purge_thread_event, sleep_ms * 1000);
 
 	history_len = trx_sys->rseg_history_len;
 	if (history_len > 1000)