KNOWN BUGS:

- WHERE and NULL doesn't always work as expected:
  In a WHERE clause 'field = NULL', NULL is changed to 0 if field is
  numerical, else to "".  IS NULL and IS NOT NULL is changed in the same way.
  String fields which are NULL behaves like "" when comparing.
  Numerical fields which are NULL behaves like 0 when comparing.

  - This will be fixed as soon we have a 'full' WHERE clause.
    In the mean time, move the test to the HAVING clause.

- HAVING can only use fields in select expression list. Functions must be
  aliased.
- Blobs can't be used in groups or sorted. Because distinct does a implicit
  group it can't be used on blobs. A workaround for some cases is to use a sub
  string.  SELECT distinct LEFT(blob,400) from table.
- The communication buffer should be dynamicly enlarged on demand. Now a query
  will fail if the query or one row of the result doesn't fit in the buffer.
  The default buffer size is 16384 for the server and 8192 for the client. 
  The buffer size may be changed at startup.
- All calculation is done with 'double'. This means that one can't use the
  full range of a longlong.
- All fields types are fixed point fields. That means one must specify how
  many decimals a floating point fields shall have. All results will be
  returned with the correct number of decimals.
