From e2119738eea97f40dd88bd794754be4eea4fe0d1 Mon Sep 17 00:00:00 2001
From: Sergey Vojtovich <svoj@mysql.com>
Date: Fri, 28 Nov 2008 18:17:13 +0400
Subject: BUG#37245 - Full text search problem

Certain boolean mode queries with truncation operator did
not return matching records and calculate relevancy
incorrectly.

myisam/ft_boolean_search.c:
  Sort ftb->list in ascending order. This helps to fix binary
  search in ft_boolean_find_relevance() without rewriting it
  much.

  Fixed binary search in ft_boolean_find_relevance(), so it finds
  right-most element in an array.

  Fixed that ft_boolean_find_relevance() didn't return match for
  words with truncation operator in case query has other non-
  matching words.
mysql-test/r/fulltext.result:
  A test case for BUG#37245.
mysql-test/t/fulltext.test:
  A test case for BUG#37245.
---
 myisam/ft_boolean_search.c | 38 ++++++++++++++++++++++++++++++++------
 1 file changed, 32 insertions(+), 6 deletions(-)

(limited to 'myisam')

diff --git a/myisam/ft_boolean_search.c b/myisam/ft_boolean_search.c
index 57de75ee4be..255c51fd33a 100644
--- a/myisam/ft_boolean_search.c
+++ b/myisam/ft_boolean_search.c
@@ -122,11 +122,11 @@ static int FTB_WORD_cmp(my_off_t *v, FTB_WORD *a, FTB_WORD *b)
 
 static int FTB_WORD_cmp_list(CHARSET_INFO *cs, FTB_WORD **a, FTB_WORD **b)
 {
-  /* ORDER BY word DESC, ndepth DESC */
-  int i= mi_compare_text(cs, (uchar*) (*b)->word+1,(*b)->len-1,
-                             (uchar*) (*a)->word+1,(*a)->len-1,0,0);
+  /* ORDER BY word, ndepth */
+  int i= mi_compare_text(cs, (uchar*) (*a)->word + 1, (*a)->len - 1,
+                             (uchar*) (*b)->word + 1, (*b)->len - 1, 0, 0);
   if (!i)
-    i=CMP_NUM((*b)->ndepth,(*a)->ndepth);
+    i= CMP_NUM((*a)->ndepth, (*b)->ndepth);
   return i;
 }
 
@@ -674,23 +674,49 @@ float ft_boolean_find_relevance(FT_INFO *ftb, byte *record, uint length)
                               (byte *) end, &word, TRUE))
     {
       int a, b, c;
+      /*
+        Find right-most element in the array of query words matching this
+        word from a document.
+      */
       for (a=0, b=ftb->queue.elements, c=(a+b)/2; b-a>1; c=(a+b)/2)
       {
         ftbw=ftb->list[c];
         if (mi_compare_text(ftb->charset, (uchar*) word.pos, word.len,
                             (uchar*) ftbw->word+1, ftbw->len-1,
-                            (my_bool) (ftbw->flags&FTB_FLAG_TRUNC),0) >0)
+                            (my_bool) (ftbw->flags & FTB_FLAG_TRUNC), 0) < 0)
           b=c;
         else
           a=c;
       }
+      /*
+        If there were no words with truncation operator, we iterate to the
+        beginning of an array until array element is equal to the word from
+        a document. This is done mainly because the same word may be
+        mentioned twice (or more) in the query.
+
+        In case query has words with truncation operator we must iterate
+        to the beginning of the array. There may be non-matching query words
+        between matching word with truncation operator and the right-most
+        matching element. E.g., if we're looking for 'aaa15' in an array of
+        'aaa1* aaa14 aaa15 aaa16'.
+
+        Worse of that there still may be match even if the binary search
+        above didn't find matching element. E.g., if we're looking for
+        'aaa15' in an array of 'aaa1* aaa14 aaa16'. The binary search will
+        stop at 'aaa16'.
+      */
       for (; c>=0; c--)
       {
         ftbw=ftb->list[c];
         if (mi_compare_text(ftb->charset, (uchar*) word.pos, word.len,
                             (uchar*) ftbw->word+1,ftbw->len-1,
                             (my_bool) (ftbw->flags&FTB_FLAG_TRUNC),0))
-          break;
+        {
+          if (ftb->with_scan & FTB_FLAG_TRUNC)
+            continue;
+          else
+            break;
+        }
         if (ftbw->docid[1] == docid)
           continue;
         ftbw->docid[1]=docid;
-- 
cgit v1.2.1