Currently, when creating the db_schema.dba from a MySQL table (but same thing for all other dbs), that contains an auto increment field with a type different than integer, so for example: tinyint, smallint, mediumint, bigint
the type is lost, and just integer will be used on next load of this db_schema.dba

Here is an export of such a table (just an extract of a db_schema.dba):

  [...]
  'djrefs_types' =>
  array (
    'name' => 'djrefs_types',
    'fields' =>
    array (
      'id' =>
      array (
        'type' => 'auto_increment',
        'default' => false,
      ),
      [...]

It comes from the fact, the type, that usually contains the type of field (logic is perfect), is also used to say if the field is auto_increment (and then we loose the real type)

However, it's 2 different things: a field can have a type (tinyint, smallint, etc..), and 'also' can be auto increment (or not)

So why not have an export that just activate a new attribute in the array that says: 'auto_increment' => true ?, and leave the type untouched

the extract of the export would be then:

  [...]
  'djrefs_types' =>
  array (
    'name' => 'djrefs_types',
    'fields' =>
    array (
      'id' =>
      array (
        'length' => 3,
        'type' => 'tinyint',
        'auto_increment' => true,
        'default' => false,
      ),
      [...]

The patch to perform this is quite easy I think, and can be done for all dbs. Because the part that needs to be patched exist (roughtly) in the same form for all dbs:

            if ( $autoinc )
            {
                unset( $field['length'] );
                $field['not_null'] = 0;
                $field['default'] = false;
                $field['type'] = 'auto_increment';
            }

On Oracle:

            if ( $isAutoIncCol )
            {
                $field['type'] = 'auto_increment';
                $field['default'] = false;
            }