Merge serg@work.mysql.com:/home/bk/mysql

into infomag.ape.relarn.ru:/usr/home/serg/src/mysql.safe/bk/mysql


Docs/manual.texi:
  Auto merged

Docs/manual.texi

@@ -13435,6 +13435,13 @@ mysql> CREATE TABLE test (
 For @code{BLOB} and @code{TEXT} columns, you must index a prefix of the
 column, you cannot index the entire thing.
 
+In @strong{MySQL} 3.23.23 or later, you can also create special
+@strong{fulltext} indexes. They are used for full-text search. Only
+@code{MyISAM} table type supports fulltext indexes. They can be created
+only from @code{VARCHAR}, @code{BLOB}, and @code{TEXT} columns.
+Indexing always happens over the entire column, partial indexing is not
+supported. See @ref{MySQL full-text search} for details of operation.
+
 @node Multiple-column indexes, Other-vendor column types, Indexes, Column types
 @subsection Multiple-column indexes
 
@@ -14150,6 +14157,17 @@ mysql> select STRCMP('text2', 'text');
 mysql> select STRCMP('text', 'text');
         -> 0
 @end example
+
+@findex MATCH ... AGAINST()
+@item MATCH (col1,col2,...) AGAINST (expr)
+@code{MATCH ... AGAINST()} is used for full-text search and returns
+relevance - similarity measure between the text in columns
+@code{(col1,col2,...)} and the query @code{expr}. Relevance is a
+positive floating point number. Zero relevance means no similarity.
+For @code{MATCH ... AGAINST()} to work, a @strong{fulltext index}
+must be created first. @xref{CREATE TABLE, , @code{CREATE TABLE}}.
+@code{MATCH ... AGAINST()} is available in @code{MySQL} 3.23.23 or later.
+For details and usage examples see @xref{MySQL full-text search}.
 @end table
 
 @findex Casts
@@ -16159,6 +16177,7 @@ create_definition:
   or    KEY [index_name] (index_col_name,...)
   or    INDEX [index_name] (index_col_name,...)
   or    UNIQUE [INDEX] [index_name] (index_col_name,...)
+  or    FULLTEXT [INDEX] [index_name] (index_col_name,...)
   or    [CONSTRAINT symbol] FOREIGN KEY index_name (index_col_name,...)
             [reference_definition]
   or    CHECK (expr)
@@ -16401,6 +16420,14 @@ When you use @code{ORDER BY} or @code{GROUP BY} with a @code{TEXT} or
 @code{BLOB} column, only the first @code{max_sort_length} bytes are used.
 @xref{BLOB,  , @code{BLOB}}.
 
+@item
+In @strong{MySQL} 3.23.23 or later, you can also create special
+@strong{fulltext} indexes. They are used for full-text search. Only
+@code{MyISAM} table type supports fulltext indexes. They can be created
+only from @code{VARCHAR}, @code{BLOB}, and @code{TEXT} columns.
+Indexing always happens over the entire column, partial indexing is not
+supported. See @ref{MySQL full-text search} for details of operation.
+
 @item
 The @code{FOREIGN KEY}, @code{CHECK} and @code{REFERENCES} clauses don't
 actually do anything.  The syntax for them is provided only for compatibility,
@@ -16570,6 +16597,7 @@ alter_specification:
   or    ADD INDEX [index_name] (index_col_name,...)
   or    ADD PRIMARY KEY (index_col_name,...)
   or    ADD UNIQUE [index_name] (index_col_name,...)
+  or    ADD FULLTEXT [index_name] (index_col_name,...)
   or    ALTER [COLUMN] col_name @{SET DEFAULT literal | DROP DEFAULT@}
   or    CHANGE [COLUMN] old_col_name create_definition
   or    MODIFY [COLUMN] create_definition
@@ -19741,7 +19769,7 @@ dropped only with explicit @code{REVOKE} commands or by manipulating the
 @section @code{CREATE INDEX} syntax
 
 @example
-CREATE [UNIQUE] INDEX index_name ON tbl_name (col_name[(length)],... )
+CREATE [UNIQUE|FULLTEXT] INDEX index_name ON tbl_name (col_name[(length)],... )
 @end example
 
 The @code{CREATE INDEX} statement doesn't do anything in @strong{MySQL} prior
@@ -19775,13 +19803,17 @@ which could save a lot of disk space and might also speed up @code{INSERT}
 operations!
 
 Note that you can only add an index on a column that can have @code{NULL}
-values or on a @code{BLOB}/@code{TEXT} column if you are useing
+values or on a @code{BLOB}/@code{TEXT} column if you are using
 @strong{MySQL} version 3.23.2 or newer and are using the @code{MyISAM}
 table type.
 
 For more information about how @strong{MySQL} uses indexes, see
 @ref{MySQL indexes, , @strong{MySQL} indexes}.
 
+Fulltext indexes can index only @code{VARCHAR}, @code{BLOB}, and
+@code{TEXT} columns, and only in @code{MyISAM} tables. Fulltext indexes
+are available from @strong{MySQL} 3.23.23. @ref{MySQL full-text search}.
+
 @findex DROP INDEX
 @node DROP INDEX, Comments, CREATE INDEX, Reference
 @section @code{DROP INDEX} syntax
@@ -34010,9 +34042,10 @@ working on the @strong{MySQL} code.
 
 @menu
 * MySQL threads::               MySQL threads
+* MySQL full-text search::      MySQL full-text search
 @end menu
 
-@node MySQL threads,  , MySQL internals, MySQL internals
+@node MySQL threads, MySQL full-text search, MySQL internals, MySQL internals
 @section MySQL threads
 
 The @strong{MySQL} server creates the the following threads:
@@ -34051,6 +34084,77 @@ started to read and apply updates from the master.
 @code{mysqladmin processlist} only shows the connection and @code{INSERT
 DELAYED} threads.
 
+@node MySQL full-text search, , MySQL threads, MySQL internals
+@section MySQL full-text search
+
+Since version 3.23.23, @strong{MySQL} has support for full-text indexing
+and searching. Full-text index in @strong{MySQL} is an
+index of type @code{FULLTEXT}. Fulltext indexes can be created from
+@code{VARCHAR}, @code{TEXT}, and @code{BLOB} columns at
+@code{CREATE TABLE} time or added later with @code{ALTER TABLE} or
+@code{CREATE INDEX}. Full-text search is performed with @code{MATCH} function.
+
+@example
+mysql> CREATE TABLE t (a VARCHAR(200), b TEXT, FULLTEXT (a,b));
+Query OK, 0 rows affected (0.00 sec)
+mysql> INSERT INTO t VALUES
+    ->   ('MySQL has now support', 'for full-text search'),
+    ->   ('Full-text indexes', 'are called collections'),
+    ->   ('Only MyISAM tables','support collections'),
+    ->   ('Function MATCH ... AGAINST()','is used to do a search'),
+    ->   ('Full-text search in MySQL', 'implements vector space model');
+Query OK, 5 rows affected (0.00 sec)
+Records: 5  Duplicates: 0  Warnings: 0
+mysql> SELECT * FROM t WHERE MATCH (a,b) AGAINST ('MySQL');
++---------------------------+-------------------------------+
+| a                         | b                             |
++---------------------------+-------------------------------+
+| MySQL has now support     | for full-text search          |
+| Full-text search in MySQL | implements vector-space-model |
++---------------------------+-------------------------------+
+2 rows in set (0.00 sec)
+mysql> SELECT *,MATCH a,b AGAINST ('collections support') as x FROM t;
++------------------------------+-------------------------------+--------+
+| a                            | b                             | x      |
++------------------------------+-------------------------------+--------+
+| MySQL has now support        | for full-text search          | 0.3834 |
+| Full-text indexes            | are called collections        | 0.3834 |
+| Only MyISAM tables           | support collections           | 0.7668 |
+| Function MATCH ... AGAINST() | is used to do a search        |      0 |
+| Full-text search in MySQL    | implements vector space model |      0 |
++------------------------------+-------------------------------+--------+
+5 rows in set (0.00 sec)
+@end example
+
+Function @code{MATCH} matches a natural language query @code{AGAINST} a
+text collection (which is simply the columns that are covered
+by fulltext index). For every row in a table it returns relevance -
+similarity measure between the text in that row (in the columns, that
+are part of the collection) and the query. When it used in a
+@code{WHERE} clause (see example above) the rows returned are
+automatically sorted with relevance decreasing. Relevance is a non-
+negative floating point number. Zero relevance means no similarity.
+Relevance is computed based on number of words in the row and number of
+unique words in that row, total number of words in the collection,
+number of documents (rows), that contain a particular word, etc.
+
+MySQL uses very simple parser to split text into words. "Word" is
+any sequence of letters, numbers, @code{'}, and @code{_}. Any "word"
+that is present in stopword list or just too short (3 characters or less)
+is ignored.
+
+Every correct word in the collection and in the query is weighted,
+according to their significance in the query or collection. This way, a
+word that is present in many documents will have lower weight (and may
+even have a zero weight), because it has lower semantic value in this
+particular collection. Otherwise, if the word is rare, it will receive a
+higher weight. Weights of the words are then combined to compute the
+relevance.
+
+Such a technique works best with big collections (in fact, it was
+carefully tuned up this way). For very small tables word distribution
+does not reflect adequately their semantical value, and this model
+may sometimes produce bizarre results.
 
 @page
 @node Environment variables, Users, MySQL internals, Top

include/Makefile.am

@@ -45,6 +45,7 @@ my_config.h:  ../config.h
 
 # This should be changed in the source and removed.
 my_global.h: global.h
+	-$(RM) my_global.h
 	$(CP) global.h my_global.h
 
 # These files should not be included in distributions since they are

include/ft_global.h

@@ -40,6 +40,8 @@ typedef struct st_ft_doclist {
   FT_DOC    doc[1];
 } FT_DOCLIST;
 
+extern const char *ft_precompiled_stopwords[];
+
 int ft_init_stopwords(const char **);
 
 FT_DOCLIST * ft_init_search(void *, uint, byte *, uint, my_bool);

myisam/ft_search.c

@@ -158,6 +158,7 @@ FT_DOCLIST * ft_init_search(void *info, uint keynr, byte *key,
   ALL_IN_ONE aio;
   FT_DOCLIST *dlist;
   FT_DOC     *dptr;
+  my_off_t    saved_lastpos;
 
 /* black magic ON */
   if ((int) (keynr = _mi_check_index((MI_INFO *)info,keynr)) < 0)
@@ -173,6 +174,8 @@ FT_DOCLIST * ft_init_search(void *info, uint keynr, byte *key,
   aio.keyinfo=aio.info->s->keyinfo+keynr;
   aio.key_root=aio.info->s->state.key_root[keynr];
 
+  saved_lastpos=aio.info->lastpos;
+
   if(!(wtree=ft_parse(NULL,key,key_len))) return NULL;
 
   init_tree(&aio.dtree,0,sizeof(FT_SUPERDOC),(qsort_cmp)&FT_SUPERDOC_cmp,0,
@@ -199,6 +202,7 @@ FT_DOCLIST * ft_init_search(void *info, uint keynr, byte *key,
   }
 
 err:
+  aio.info->lastpos=saved_lastpos;
   delete_tree(&aio.dtree);
   delete_tree(wtree);
   free(wtree);
@@ -217,7 +221,8 @@ int ft_read_next(FT_DOCLIST *handler, char *record)
 
   info->update&= (HA_STATE_CHANGED | HA_STATE_ROW_CHANGED);
 
-  if (!(*info->read_record)(info,handler->doc[handler->curdoc].dpos,record))
+  info->lastpos=handler->doc[handler->curdoc].dpos;
+  if (!(*info->read_record)(info,info->lastpos,record))
   {
     info->update|= HA_STATE_AKTIV;		/* Record is read */
     return 0;

myisam/fulltext.h

@@ -35,8 +35,6 @@
 
 extern const MI_KEYSEG ft_keysegs[FT_SEGS];
 
-extern const char *ft_precompiled_stopwords[];
-
 int  _mi_ft_cmp(MI_INFO *, uint, const byte *, const byte *);
 int  _mi_ft_add(MI_INFO *, uint, byte *, const byte *, my_off_t);
 int  _mi_ft_del(MI_INFO *, uint, byte *, const byte *, my_off_t);

myisam/mi_check.c

@@ -1083,7 +1083,7 @@ int mi_repair(MI_CHECK *param, register MI_INFO *info,
   if (!(sort_info->record=(byte*) my_malloc((uint) share->base.pack_reclength,
 					   MYF(0))))
   {
-    mi_check_print_error(param,"Not Enough memory for extra record");
+    mi_check_print_error(param,"Not enough memory for extra record");
     goto err;
   }
 
@@ -1141,6 +1141,14 @@ int mi_repair(MI_CHECK *param, register MI_INFO *info,
   for (i=0 ; i < share->state.header.max_block_size ; i++)
     share->state.key_del[i]=  HA_OFFSET_ERROR;
 
+  /* I think mi_repair and mi_repair_by_sort should do the same
+     (according, e.g. to ha_myisam::repair), but as mi_repair doesn't
+     touch key_map it cannot be used to T_CREATE_MISSING_KEYS.
+     That is the next line for... (serg)
+  */
+
+  share->state.key_map= ((ulonglong)1L << share->base.keys)-1;
+
   info->state->key_file_length=share->base.keystart;
 
   lock_memory(param);			/* Everything is alloced */
@@ -1270,11 +1278,18 @@ static int writekeys(register MI_INFO *info,byte *buff,my_off_t filepos)
   for (i=0 ; i < info->s->base.keys ; i++)
   {
     if (((ulonglong) 1 << i) & info->s->state.key_map)
+    {
+      if (info->s->keyinfo[i].flag & HA_FULLTEXT )
+      {
+        if (_mi_ft_add(info,i,(char*) key,buff,filepos))  goto err;
+      }
+      else
       {
 	uint key_length=_mi_make_key(info,i,key,buff,filepos);
 	if (_mi_ck_write(info,i,key,key_length)) goto err;
       }
     }
+  }
   DBUG_RETURN(0);
 
  err:
@@ -1284,12 +1299,19 @@ static int writekeys(register MI_INFO *info,byte *buff,my_off_t filepos)
     while ( i-- > 0 )
     {
       if (((ulonglong) 1 << i) & info->s->state.key_map)
+      {
+	if (info->s->keyinfo[i].flag & HA_FULLTEXT)
+        {
+          if (_mi_ft_del(info,i,(char*) key,buff,filepos)) break;
+        }
+        else
 	{
 	  uint key_length=_mi_make_key(info,i,key,buff,filepos);
 	  if (_mi_ck_delete(info,i,key,key_length)) break;
 	}
       }
     }
+  }
   DBUG_PRINT("error",("errno: %d",my_errno));
   DBUG_RETURN(-1);
 } /* writekeys */
@@ -1919,8 +1941,22 @@ static int sort_key_read(SORT_INFO *sort_info, void *key)
 			 "Found too many records; Can`t continue");
     DBUG_RETURN(1);
   }
+  /* Hmm, repair_by_sort uses find_all_keys, and find_all_keys strictly
+     implies "one row - one key per keynr", while for ft_key one row/keynr
+     can produce as many keys as the number of unique words in the text
+     that's why I disabled repair_by_sort for ft-keys. (serg)
+   */
+  if (sort_info->keyinfo->flag & HA_FULLTEXT )
+  {
+    mi_check_print_error(sort_info->param,
+    			 "Can`t use repair_by_sort with FULLTEXT key");
+    DBUG_RETURN(1);
+  }
+  else
+  {
     VOID(_mi_make_key(info,sort_info->key,key,sort_info->record,
                       sort_info->filepos));
+  }
   DBUG_RETURN(sort_write_record(sort_info));
 } /* sort_key_read */
 
@@ -2984,7 +3020,14 @@ my_bool mi_test_if_sort_rep(MI_INFO *info, ha_rows rows)
     return FALSE;				/* Can't use sort */
   for (i=0 ; i < share->base.keys ; i++,key++)
   {
-    if (mi_too_big_key_for_sort(key,rows))
+/* It's to disable repair_by_sort for ft-keys.
+   Another solution would be to make ft-keys just too_big_key_for_sort,
+   but then they won't be disabled by dectivate_non_unique_index
+   and so they will be created at the first stage. As ft-key creation
+   is very time-consuming process, it's better to leave it to repair stage
+   but this repair shouldn't be repair_by_sort (serg)
+ */
+    if (mi_too_big_key_for_sort(key,rows) || (key->flag & HA_FULLTEXT))
       return FALSE;
   }
   return TRUE;

sql/ha_myisam.cc

@@ -372,6 +372,7 @@ int ha_myisam::repair(THD *thd, MI_CHECK &param)
   param.table_name = table->table_name;
   param.tmpfile_createflag = O_RDWR | O_TRUNC;
   param.using_global_keycache = 1;
+  param.thd=thd;
 
   VOID(fn_format(fixed_name,file->filename,"",MI_NAME_IEXT,
 		     4+ (param.opt_follow_links ? 16 : 0)));

sql/item_func.cc

@@ -1837,26 +1837,8 @@ err:
 
 double Item_func_match::val()
 {
-  my_off_t docid=table->file->row_position();  // HAVE to do it here...
-
   if (first_call)
-  {
-    if (join_key=(table->file->get_index() == key &&
-                 (ft_handler=(FT_DOCLIST *)table->file->ft_handler)))
-      ;
-    else
-    {
-      /* join won't use this ft-key, but we must to init it anyway */
-      String *ft_tmp=0;
-      char tmp1[FT_QUERY_MAXLEN];
-      String tmp2(tmp1,sizeof(tmp1));
-
-      ft_tmp=key_item()->val_str(&tmp2);
-      ft_handler=(FT_DOCLIST *)
-         table->file->ft_init_ext(key, (byte*) ft_tmp->ptr(), ft_tmp->length());
-    }
-    first_call=0;
-  }
+    init_search();
 
   // Don't know how to return an error from val(), so NULL will be returned
   if ((null_value=(ft_handler==NULL)))
@@ -1873,6 +1855,7 @@ double Item_func_match::val()
 
     int a,b,c;
     FT_DOC  *docs=ft_handler->doc;
+    my_off_t docid=table->file->row_position();
 
     if ((null_value=(docid==HA_OFFSET_ERROR)))
       return 0.0;
@@ -1893,6 +1876,36 @@ double Item_func_match::val()
   }
 }
 
+void Item_func_match::init_search()
+{
+  if (!first_call)
+    return;
+  first_call=false;
+
+  if (master)
+  {
+    master->init_search();
+    ft_handler=master->ft_handler;
+    join_key=master->join_key;
+    return;
+  }
+
+  if (join_key)
+  {
+    ft_handler=((FT_DOCLIST *)table->file->ft_handler);
+    return;
+  }
+
+  /* join won't use this ft-key, but we must to init it anyway */
+  String *ft_tmp=0;
+  char tmp1[FT_QUERY_MAXLEN];
+  String tmp2(tmp1,sizeof(tmp1));
+
+  ft_tmp=key_item()->val_str(&tmp2);
+  ft_handler=(FT_DOCLIST *)
+     table->file->ft_init_ext(key, (byte*) ft_tmp->ptr(), ft_tmp->length());
+}
+
 bool Item_func_match::fix_fields(THD *thd,struct st_table_list *tlist)
 {
   List_iterator<Item> li(fields);
@@ -1982,6 +1995,24 @@ bool Item_func_match::fix_index()
   this->key=max_key;
   first_call=1;
   maybe_null=1;
+  join_key=0;
+
+  return 0;
+}
+
+bool Item_func_match::eq(const Item *item) const
+{
+  if (item->type() != FUNC_ITEM)
+    return 0;
+
+  if (func_name() != ((Item_func*)item)->func_name())
+    return 0;
+
+  Item_func_match *ifm=(Item_func_match*) item;
+
+  if (key == ifm->key && table == ifm->table &&
+      key_item()->eq(ifm->key_item()))
+    return 1;
 
   return 0;
 }

sql/item_func.h

@@ -839,19 +839,28 @@ public:
   TABLE *table;
   uint key;
   bool first_call, join_key;
+  Item_func_match *master;
   FT_DOCLIST *ft_handler;
 
   Item_func_match(List<Item> &a, Item *b): Item_real_func(b),
-  fields(a), table(0), ft_handler(0)
-  {}
-  ~Item_func_match() { ft_close_search(ft_handler);
-                       if(join_key) table->file->ft_handler=0; }
+  fields(a), table(0), ft_handler(0), master(0)  {}
+  ~Item_func_match()
+  {
+    if (!master)
+    {
+      ft_close_search(ft_handler);
+      if(join_key)
+        table->file->ft_handler=0;
+    }
+  }
   const char *func_name() const { return "match"; }
   enum Functype functype() const { return FT_FUNC; }
   void update_used_tables() {}
   bool fix_fields(THD *thd,struct st_table_list *tlist);
-  bool fix_index();
-
+  bool eq(const Item *) const;
   double val();
   longlong val_int() { return val()!=0.0; }
+
+  bool fix_index();
+  void init_search();
 };

sql/lex.h

@@ -81,7 +81,6 @@ static SYMBOL symbols[] = {
   { "CHANGED",		SYM(CHANGED),0,0},
   { "CHECK",		SYM(CHECK_SYM),0,0},
   { "CHECKSUM",		SYM(CHECKSUM_SYM),0,0},
-  { "COLLECTION",       SYM(COLLECTION),0,0},
   { "COLUMN",		SYM(COLUMN_SYM),0,0},
   { "COLUMNS",		SYM(COLUMNS),0,0},
   { "COMMENT",		SYM(COMMENT_SYM),0,0},
@@ -142,6 +141,7 @@ static SYMBOL symbols[] = {
   { "FROM",		SYM(FROM),0,0},
   { "FOR",		SYM(FOR_SYM),0,0},
   { "FULL",		SYM(FULL),0,0},
+  { "FULLTEXT",		SYM(FULLTEXT_SYM),0,0},
   { "FUNCTION",		SYM(UDF_SYM),0,0},
   { "GRANT",		SYM(GRANT),0,0},
   { "GRANTS",	        SYM(GRANTS),0,0},

sql/mysqld.cc

@@ -1455,7 +1455,7 @@ int main(int argc, char **argv)
     sql_print_error("Can't init databases");
     exit(1);
   }
-  ft_init_stopwords(NULL);                                          /* SerG */
+  ft_init_stopwords(ft_precompiled_stopwords);       /* SerG */
 
 #ifdef __WIN__
 #define MYSQL_ERR_FILE "mysql.err"

sql/sql_base.cc

@@ -2172,18 +2172,22 @@ bool remove_table_from_cache(THD *thd, const char *db,const char *table_name)
   DBUG_RETURN(result);
 }
 
-/*
-   Will be used for ft-query optimization someday.
-   SerG.
- */
 int setup_ftfuncs(THD *thd,TABLE_LIST *tables, List<Item_func_match> &ftfuncs)
 {
-  List_iterator<Item_func_match> li(ftfuncs);
-  Item_func_match *ftf;
+  List_iterator<Item_func_match> li(ftfuncs), li2(ftfuncs);
+  Item_func_match *ftf, *ftf2;
 
   while ((ftf=li++))
+  {
     if (ftf->fix_index())
       return 1;
+    li2.rewind();
+    while ((ftf2=li2++) != ftf)
+    {
+      if (ftf->eq(ftf2) && !ftf2->master)
+        ftf2->master=ftf;
+    }
+  }
 
   return 0;
 }

sql/sql_select.cc

@@ -1284,11 +1284,11 @@ add_ft_keys(DYNAMIC_ARRAY *keyuse_array,
   KEYUSE keyuse;
 
   keyuse.table= cond_func->table;
-  keyuse.val =  cond_func->key_item();
+  keyuse.val =  cond_func;
   keyuse.key =  cond_func->key;
 #define FT_KEYPART   (MAX_REF_PARTS+10)
   keyuse.keypart=FT_KEYPART;
-  keyuse.used_tables=keyuse.val->used_tables();
+  keyuse.used_tables=cond_func->key_item()->used_tables();
   VOID(insert_dynamic(keyuse_array,(gptr) &keyuse));
 }
 
@@ -1670,7 +1670,7 @@ find_best(JOIN *join,table_map rest_tables,uint idx,double record_count,
 	    else
 	      tmp=best_time;			// Do nothing
 	  }
-          } /* not ftkey */
+          } /* not ft_key */
 	  if (tmp < best_time - records/(double) TIME_FOR_COMPARE)
 	  {
 	    best_time=tmp + records/(double) TIME_FOR_COMPARE;
@@ -1882,9 +1882,12 @@ get_best_combination(JOIN *join)
       keyinfo=table->key_info+key;
       if (ftkey)
       {
-        ft_tmp=keyuse->val->val_str(&tmp2);
+        Item_func_match *ifm=(Item_func_match *)keyuse->val;
+
+        ft_tmp=ifm->key_item()->val_str(&tmp2);
         length=ft_tmp->length();
         keyparts=1;
+        ifm->join_key=1;
       }
       else
       {
@@ -1924,7 +1927,7 @@ get_best_combination(JOIN *join)
       byte *key_buff=j->ref.key_buff;
       if (ftkey)
       {
-        j->ref.items[0]=keyuse->val;
+        j->ref.items[0]=((Item_func*)(keyuse->val))->key_item();
         if (!keyuse->used_tables &&
             !(join->select_options & SELECT_DESCRIBE))
         {

sql/sql_yacc.yy

@@ -137,7 +137,6 @@ bool my_yyoverflow(short **a, YYSTYPE **b,int *yystacksize);
 %token  CHANGED_FILES
 %token	CHECKSUM_SYM
 %token	CHECK_SYM
-%token  COLLECTION
 %token	COLUMNS
 %token	COLUMN_SYM
 %token	CONSTRAINT
@@ -162,6 +161,7 @@ bool my_yyoverflow(short **a, YYSTYPE **b,int *yystacksize);
 %token	FOREIGN
 %token	FROM
 %token	FULL
+%token  FULLTEXT_SYM
 %token	GRANT
 %token	GRANTS
 %token	GREATEST_SYM
@@ -457,7 +457,7 @@ bool my_yyoverflow(short **a, YYSTYPE **b,int *yystacksize);
 	expr_list udf_expr_list when_list ident_list
 
 %type <key_type>
-	key_type opt_unique
+	key_type opt_unique_or_fulltext
 
 %type <string_list>
 	key_usage_list
@@ -628,7 +628,7 @@ create:
 	}
 	create2
 
-	| CREATE opt_unique INDEX ident ON table_ident
+	| CREATE opt_unique_or_fulltext INDEX ident ON table_ident
 	  {
 	    Lex->sql_command= SQLCOM_CREATE_INDEX;
 	    if (!add_table_to_list($6,NULL))
@@ -643,21 +643,6 @@ create:
 	    Lex->key_list.push_back(new Key($2,$4.str,Lex->col_list));
 	    Lex->col_list.empty();
 	  }
-        | CREATE COLLECTION ident ON table_ident
-          {
-            Lex->sql_command= SQLCOM_CREATE_INDEX;
-            if (!add_table_to_list($5,NULL))
-              YYABORT;
-            Lex->create_list.empty();
-            Lex->key_list.empty();
-            Lex->col_list.empty();
-            Lex->change=NullS;
-          }
-          '(' key_list ')'
-          {
-            Lex->key_list.push_back(new Key(Key::FULLTEXT,$3.str,Lex->col_list));
-            Lex->col_list.empty();
-          }
 	| CREATE DATABASE opt_if_not_exists ident
 	  {
 	    Lex->sql_command=SQLCOM_CREATE_DB;
@@ -964,7 +949,8 @@ delete_option:
 key_type:
 	opt_constraint PRIMARY_SYM KEY_SYM  { $$= Key::PRIMARY; }
 	| key_or_index			    { $$= Key::MULTIPLE; }
-	| COLLECTION			    { $$= Key::FULLTEXT; }
+	| FULLTEXT_SYM			    { $$= Key::FULLTEXT; }
+	| FULLTEXT_SYM key_or_index	    { $$= Key::FULLTEXT; }
 	| opt_constraint UNIQUE_SYM	    { $$= Key::UNIQUE; }
 	| opt_constraint UNIQUE_SYM key_or_index { $$= Key::UNIQUE; }
 
@@ -976,9 +962,10 @@ keys_or_index:
 	KEYS {}
 	| INDEX {}
 
-opt_unique:
+opt_unique_or_fulltext:
 	/* empty */	{ $$= Key::MULTIPLE; }
 	| UNIQUE_SYM	{ $$= Key::UNIQUE; }
+	| FULLTEXT_SYM	{ $$= Key::FULLTEXT; }
 
 key_list:
 	key_list ',' key_part order_dir { Lex->col_list.push_back($3); }
@@ -2443,7 +2430,6 @@ keyword:
 	| WORK_SYM		{}
 	| YEAR_SYM		{}
         | SLAVE                 {}
-        | COLLECTION            {}
 
 /* Option functions */