When you use pt-table-checksum with --where option, if index chosen at sub get_row_estimate is not a primary key, it will not be able to choose primary key as index

So, table structure is this:

CREATE TABLE t1 (
    a int(10) unsigned NOT NULL,
  b datetime DEFAULT NULL,
  PRIMARY KEY (a),
  KEY b(b)
) ENGINE=InnoDB DEFAULT CHARSET=latin1

Cardinality is this:
mysql> show indexes from test.a
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment |
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| t1 | 0 | PRIMARY | 1 | gni | A | 97330 | NULL | NULL | | BTREE | | |
| t1 | 1 | b | 1 | b | A | 97330 | NULL | NULL | YES | BTREE | | |
+----------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+

pt-table-checksum --databases=test --chunk-time=0.01 --chunk-size=4 h=127.0.0.1,P=21690,u=root,p=msandbox --empty-replicate-table 2>&1 --where "1=1" | tee output.txt

Assuming in get_row_estimate sub, b index was chosen,
so:

$mysql_index = lc($mysql_index);

so here $mysql_index = 'b';

On can_nibble sub:
   if ( !$where ) {
      $mysql_index = undef;
   }

Since where is equal to "1=1", $mysql_index will not be set to undef.

So, on function _find_best_index:

   //want_index is false because chunk-index was not set:
   if ( !$want_index && $args{mysql_index} ) {
      PTDEBUG && _d('MySQL wants to use index', $args{mysql_index});
      $want_index = $args{mysql_index};
   }
  //Now, want_index contains b index

   my $best_index;
   my @possible_indexes;

  Since $want_index has a value, it will not be able to reach code for autoselecting the best index where the PRIMARY KEY is a candidate.

   if ( $want_index ) {
      if ( $indexes->{$want_index}->{is_unique} ) {
         PTDEBUG && _d('Will use wanted index');
         $best_index = $want_index;
      }
      else {
         PTDEBUG && _d('Wanted index is a possible index');
         push @possible_indexes, $want_index;
      }
   }
   else {
      PTDEBUG && _d('Auto-selecting best index');
      foreach my $index ( $tp->sort_indexes($tbl_struct) ) {
         if ( $index eq 'PRIMARY' || $indexes->{$index}->{is_unique} ) {
            $best_index = $index;
            last;
         }
         else {
            push @possible_indexes, $index;
         }
      }
   }

The problem is if b index is chosen and b column contains nulls and the boundary of the chunk is null and null, this is how debug output would look like:
# NibbleIterator:6647 30375 First lower boundary: $VAR1 = [
# undef,
# undef
# ];
#
# NibbleIterator:6663 30375 Next lower boundary: $VAR1 = [
# undef,
# undef
# ];
#

Then you would get this error:

04-25T14:39:09 Error checksumming table test.t1: Use of uninitialized value in join or string at /usr/bin/pt-table-checksum line 6727.

I think there are two bugs here. The other bug is if the boundaries is both null, you should test for this and not get an perl exception. This is reported here: https://bugs.launchpad.net/percona-toolkit/+bug/1202026