Bug #33379 valgrind error in parts/partition_bit_myisam

Description:
test case partition_bit_myisam.test in suite parts fails with valgrind errors

How to repeat:
compile with valgrind
./mtr --valgrind --suite=parts partition_bit_myisam

Suggested fix:
--- 1.424/sql/field.cc	2007-12-05 23:31:15 +01:00
+++ edited/sql/field.cc	2007-12-19 19:59:40 +01:00
@@ -8774,7 +8774,7 @@
   : Field(ptr_arg, len_arg, null_ptr_arg, null_bit_arg,
           unireg_check_arg, field_name_arg),
     bit_ptr(bit_ptr_arg), bit_ofs(bit_ofs_arg), bit_len(len_arg & 7),
-    bytes_in_rec(len_arg / 8)
+    bytes_in_rec((len_arg + 7) / 8)
 {
   flags|= UNSIGNED_FLAG;
   /*

I.e. field is created with wrong number of bytes in record, affects field->pack_lenght_in_rec() function that is used in partitioning.

suggested fix is wrong!

better to fix by changing pack_length_in_rec() function instead.

As Kristian Nielsen proposed:
The problem lies in Field::hash (which Field_bit is using).
I will try to implement the Field_bit::hash function using my_charset_bin.coll->hash_sort() with a correct buffer.

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/41633

ChangeSet@1.2516, 2008-02-04 12:28:42+01:00, mattiasj@client-10-129-10-147.upp.off.mysql.com +4 -0
  Bug#33379: valgrind error in parts/partition_bit_myisam
  
  Problem was that Field_bit used Field::hash() function that did not
  know about using null-byte for storing bits.
  Resulting in wrong length, which was caught by valgrind.
  
  Soluiton: created a Field_bit::hash() that uses Field_bit::val_int()
  and my_charset_bin-collation function hash_sort.

As Kristian commented, the previous patch was endianess dependent, I will remake it endian independent, using mi_intNstore into a local buffer.

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/41855

ChangeSet@1.2516, 2008-02-07 10:12:56+01:00, mattiasj@witty.ndb.mysql.com +4 -0
  Bug#33379: valgrind error in parts/partition_bit_myisam
  
  Problem was that Field_bit used Field::hash() function that did not
  know about using null-byte for storing bits.
  Resulting in wrong length, which was caught by valgrind.
  
  Solution: created a Field_bit::hash() that uses Field_bit::val_int()
  and my_charset_bin-collation function hash_sort.
  Also use the store function for platform independs

Discussed with Mikael, and we decided to change the switch/case to use the mi_int8store directly. new patch will be committed soon

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/41860

ChangeSet@1.2516, 2008-02-07 12:28:38+01:00, mattiasj@witty.local +4 -0
  Bug#33379: valgrind error in parts/partition_bit_myisam
  
  Problem was that Field_bit used Field::hash() function that did not
  know about using null-byte for storing bits.
  Resulting in wrong length, which was caught by valgrind.
  
  Solution: created a Field_bit::hash() that uses Field_bit::val_int()
  and my_charset_bin-collation function hash_sort.
  Also use the store function for platform independs

Pushed into mysql-6.0-engines and mysql-5.1-engines

Pushed into 5.1.24-rc

Pushed into 6.0.5-alpha

There don't appear to be any user-visible changes here; closed bug without taking further action.

I think the user-visible change here is that before this fix, PARTITION BY KEY on
bitfields did not work correctly. I think rows could end up in the wrong
partition, causing wrong results to be returned from queries or other similar
problems.

We do not have a testcase for this though.

Still, might be good to document that this bug fixes potential problems with
PARTITION BY KEY on bitfield columns.

Setting this back to Documenting status per developer comment above.

Documented bugfix as follows in the 5.1.24 and 6.0.5 changelogs:

        A table defined using PARTITION BY KEY and having a BIT column referenced 
        in the partitioning key did not behave correctly; some rows could be 
        inserted into the wrong partition, causing wrong results to be returned 
        from queries.

Also noted fix in the 5.1.23-ndb-6.3.11 changelog.