bk commit into 4.1 tree (sergefp:1.2486) BUG#9622

[Sergey Petrunia]

Below is the list of changes that have just been committed into a local
4.1 repository of psergey. When psergey does a push these changes will
be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet
  1.2486 05/10/23 02:46:10 sergefp@stripped +3 -0
  BUG#9622: post-review-fixes: better comments

  mysys/my_handler.c
    1.20 05/10/23 02:46:04 sergefp@stripped +19 -7
    BUG#9622: post-review-fixes: better comments

  myisam/mi_check.c
    1.153 05/10/23 02:46:04 sergefp@stripped +32 -22
    BUG#9622: post-review-fixes: better comments

  include/myisam.h
    1.66 05/10/23 02:46:04 sergefp@stripped +1 -1
    BUG#9622: post-review-fixes: better comments

# This is a BitKeeper patch.  What follows are the unified diffs for the
# set of deltas contained in the patch.  The rest of the patch, the part
# that BitKeeper cares about, is below these diffs.
# User:	sergefp
# Host:	newbox.mylan
# Root:	/home/psergey/mysql-4.1-nulls-stats-r2

--- 1.65/include/myisam.h	2005-10-21 06:29:11 +04:00
+++ 1.66/include/myisam.h	2005-10-23 02:46:04 +04:00
@@ -323,7 +323,7 @@
   MI_STATS_METHOD_NULLS_NOT_EQUAL,
   /* Treat NULLs as equal when collecting statistics (like 4.0 did) */
   MI_STATS_METHOD_NULLS_EQUAL,
-  /* Ignore NULLs - count tuples without NULLs only */
+  /* Ignore NULLs - count only tuples without NULLs in the index components */
   MI_STATS_METHOD_IGNORE_NULLS
 } enum_mi_stats_method;
 

--- 1.152/myisam/mi_check.c	2005-10-21 06:29:11 +04:00
+++ 1.153/myisam/mi_check.c	2005-10-23 02:46:04 +04:00
@@ -603,10 +603,11 @@
     Process the next index tuple:
     1. Find out which prefix tuples of last_key don't contain NULLs, and
        update the array of notnull counters accordingly.
-    2. Find the first keypart number where the tuples are different(A), or
-       last_key has NULL value (B), and return it, so caller can count
-       number of unique tuples for each key prefix. We don't need (B) to be
-       counted, and that is compensated back in update_key_parts().
+    2. Find the first keypart number where the prev_key and last_key tuples
+       are different(A), or last_key has NULL value(B), and return it, so the 
+       caller can count number of unique tuples for each key prefix. We don't 
+       need (B) to be counted, and that is compensated back in 
+       update_key_parts().
 
   RETURN
     1 + number of first keypart where values differ or last_key tuple has NULL
@@ -619,11 +620,19 @@
   uint diffs[2];
   uint first_null_seg, kp;
 
-  /* Find first keypart where values are different or either of them is NULL */
+  /* 
+     Find the first keypart where values are different or either of them is
+     NULL. We get results in diffs array:
+     diffs[0]= 1 + number of first different keypart
+     diffs[1]=offset: (last_key + diffs[1]) points to first value in
+                      last_key that is NULL or different from corresponding
+                      value in prev_key.
+  */
   ha_key_cmp(keyseg, prev_key, last_key, USE_WHOLE_KEY, 
              SEARCH_FIND | SEARCH_NULL_ARE_NOT_EQUAL | SEARCH_RETURN_B_POS,
              diffs);
   HA_KEYSEG *seg= keyseg + diffs[0] - 1;
+
   /* Find first NULL in last_key */
   first_null_seg= ha_find_null(seg, last_key + diffs[1]) - keyseg;
   for (kp= 0; kp < first_null_seg; kp++)
@@ -4087,7 +4096,7 @@
 
 /*
   Update statistics for each part of an index
-  
+
   SYNOPSIS
     update_key_parts()
       keyinfo           IN  Index information (only key->keysegs used)
@@ -4095,25 +4104,26 @@
       unique            IN  Array of (#distinct tuples)
       notnull_tuples    IN  Array of (#tuples), or NULL
       records               Number of records in the table
-      
-  NOTES
+
+  DESCRIPTION
+    This function is called produce index statistics values from unique and 
+    notnull_tuples arrays after these arrays were produced with sequential
+    index scan (the scan is done in two places: chk_index() and
+    sort_key_write()).
+
     This function handles all 3 index statistics collection methods.
 
     Unique is an array:
-    unique[0]= (#different values of {keypart1}) - 1
-    unique[1]= (#different values of {keypart1,keypart2} tuple) - unique[0] - 1
-    ...
-    
-    For MI_STATS_METHOD_IGNORE_NULLS notnull_tuples is an array too:
-    notnull_tuples[0]= (# of {keypart1} tuples such that keypart1 is not NULL)
-    notnull_tuples[1]= (# of {keypart1,keypart2} tuples such that all 
-                        keypart{i} are not NULL)
-    ...
-    For all other statistics collection methods notnull_tuples=NULL.
-      
-    The 'unique' array is collected in one sequential scan through the entire
-    index. This is done in two places: in chk_index() and in sort_key_write().
-    notnull_tuples, if present, is collected during the same index scan.
+      unique[0]= (#different values of {keypart1}) - 1
+      unique[1]= (#different values of {keypart1,keypart2} tuple)-unique[0]-1
+      ...
+
+    For MI_STATS_METHOD_IGNORE_NULLS method, notnull_tuples is an array too:
+      notnull_tuples[0]= (#of {keypart1} tuples such that keypart1 is not NULL)
+      notnull_tuples[1]= (#of {keypart1,keypart2} tuples such that all 
+                          keypart{i} are not NULL)
+      ...
+    For all other statistics collection methods notnull_tuples==NULL.
 
     Output is an array:
     rec_per_key_part[k] = 

--- 1.19/mysys/my_handler.c	2005-10-21 06:29:11 +04:00
+++ 1.20/mysys/my_handler.c	2005-10-23 02:46:04 +04:00
@@ -86,18 +86,30 @@
 		position and this should also be compared
     diff_pos    OUT Number of first keypart where values differ, counting 
                 from one.
-                
-  NOTES
-    Number-keys can't be splited
   
   DESCRIPTION
-  
     If SEARCH_RETURN_B_POS flag is set, diff_pos must point to array of 2
-    values, first value has the meaning as described above, second value is:
-  
+    values, first value has the meaning as described in parameter
+    description above, the second value is:
+
     diff_pos[1]  OUT  (b + diff_pos[1]) points to first value in tuple b
                       that is different from corresponding value in tuple a.
-  
+
+  EXAMPLES 
+   Example1: if the function is called for tuples
+     ('aaa','bbb') and ('eee','fff'), then
+     diff_pos[0] = 1 (as 'aaa' != 'eee')
+     diff_pos[1] = 0 (offset from beggining of tuple b to 'eee' keypart).
+
+   Example2: if the index function is called for tuples
+     ('aaa','bbb') and ('aaa','fff'),
+     diff_pos[0] = 2 (as 'aaa' != 'eee')
+     diff_pos[1] = 3 (offset from beggining of tuple b to 'fff' keypart,
+                      here we assume that first key part is CHAR(3) NOT NULL)
+
+  NOTES
+    Number-keys can't be splited
+
   RETURN VALUES
     <0	If a < b
     0	If a == b