Fixed performance bug in lock tables

Docs/manual.texi:
  Updated changelog
mysys/thr_lock.c:
  Fixed bad performance bug when using SELECT, INSERT and UPDATE
strings/llstr.c:
  change llstr() to output signed strings

Docs/manual.texi

@@ -38723,6 +38723,12 @@ though, so Version 3.23 is not released as a stable version yet.
 @appendixsubsec Changes in release 3.23.28
 @itemize @bullet
 @item
+Fixed a major performance bug in the table locking code when you
+constantly had a LOT of @code{SELECT} running on a table on which you
+also did a lot of @code{UPDATE} and @code{INSERT}.  The symptom was that
+the @code{UPDATE} and @code{INSERT} queries was locked a long time
+while @code{SELECT} statements where executed without locks.
+@item
 One can now specify @code{interactive-timeout} in the option file that
 is read by @code{mysql_options()}. This makes it possible to force
 programs that runs for a long time (like @code{mysqlhotcopy}) to use
@@ -43268,6 +43274,8 @@ Secure connections (with SSL).
 @item
 Extend the optimizer to be able to optimize some
 @code{ORDER BY key_name DESC} queries.
+@item
+New key cache
 @end itemize
 
 @node TODO future, TODO sometime, TODO MySQL 4.0, TODO

mysys/thr_lock.c

@@ -653,7 +653,7 @@ void thr_unlock(THR_LOCK_DATA *data)
   data->type=TL_UNLOCK;				/* Mark unlocked */
   check_locks(lock,"after releasing lock",1);
 
-  if (!lock->write.data)			/* If no active read locks */
+  if (!lock->write.data)			/* If no active write locks */
   {
     data=lock->write_wait.data;
     if (!lock->read.data)			/* If no more locks in use */
@@ -756,7 +756,7 @@ void thr_unlock(THR_LOCK_DATA *data)
 			    (lock_type == TL_WRITE_CONCURRENT_INSERT ||
 			     lock_type == TL_WRITE_ALLOW_WRITE));
     }
-    else if (lock->read_wait.data)
+    else if (!data && lock->read_wait.data)
       free_all_read_locks(lock,0);
   }
 end:

strings/llstr.c

@@ -31,6 +31,6 @@
 
 char *llstr(longlong value,char *buff)
 {
-  longlong2str(value,buff,10);
+  longlong2str(value,buff,-10);
   return buff;
 }