Support for the MaxDB database.
Overview
The maxdb dialect is experimental and has only been tested on 7.6.03.007
and 7.6.00.037. Of these, only 7.6.03.007 will work with SQLAlchemy’s ORM.
The earlier version has severe LEFT JOIN limitations and will return
incorrect results from even very simple ORM queries.
Only the native Python DB-API is currently supported. ODBC driver support
is a future enhancement.
Implementation Notes
With the 7.6.00.37 driver and Python 2.5, it seems that all DB-API
generated exceptions are broken and can cause Python to crash.
For ‘somecol.in_([])’ to work, the IN operator’s generation must be changed
to cast ‘NULL’ to a numeric, i.e. NUM(NULL). The DB-API doesn’t accept a
bind parameter there, so that particular generation must inline the NULL value,
which depends on [ticket:807].
The DB-API is very picky about where bind params may be used in queries.
Bind params for some functions (e.g. MOD) need type information supplied.
The dialect does not yet do this automatically.
Max will occasionally throw up ‘bad sql, compile again’ exceptions for
perfectly valid SQL. The dialect does not currently handle these, more
research is needed.
MaxDB 7.5 and Sap DB <= 7.4 reportedly do not support schemas. A very
slightly different version of this dialect would be required to support
those versions, and can easily be added if there is demand. Some other
required components such as an Max-aware ‘old oracle style’ join compiler
(thetas with (+) outer indicators) are already done and available for
integration- email the devel list if you’re interested in working on
this.
Versions tested: 7.6.03.07 and 7.6.00.37, native Python DB-API
- MaxDB has severe limitations on OUTER JOINs, which are essential to ORM
eager loading. And rather than raise an error if a SELECT can’t be serviced,
the database simply returns incorrect results.
- Version 7.6.03.07 seems to JOIN properly, however the docs do not show the
OUTER restrictions being lifted (as of this writing), and no changelog is
available to confirm either. If you are using a different server version and
your tasks require the ORM or any semi-advanced SQL through the SQL layer,
running the SQLAlchemy test suite against your database is HIGHLY
recommended before you begin.
- Version 7.6.00.37 is LHS/RHS sensitive in FROM lhs LEFT OUTER JOIN rhs ON
lhs.col=rhs.col vs rhs.col=lhs.col!
- Version 7.6.00.37 is confused by SELECT DISTINCT col as alias FROM t ORDER
BY col - these aliased, DISTINCT, ordered queries need to be re-written to
order by the alias name.
- Version 7.6.x supports creating a SAVEPOINT but not its RELEASE.
- MaxDB supports autoincrement-style columns (DEFAULT SERIAL) and independent
sequences. When including a DEFAULT SERIAL column in an insert, 0 needs to
be inserted rather than NULL to generate a value.
- MaxDB supports ANSI and “old Oracle style” theta joins with (+) outer join
indicators.
- The SQLAlchemy dialect is schema-aware and probably won’t function correctly
on server versions (pre-7.6?). Support for schema-less server versions could
be added if there’s call.
- ORDER BY is not supported in subqueries. LIMIT is not supported in
subqueries. In 7.6.00.37, TOP does work in subqueries, but without limit not
so useful. OFFSET does not work in 7.6 despite being in the docs. Row number
tricks in WHERE via ROWNO may be possible but it only seems to allow
less-than comparison!
- Version 7.6.03.07 can’t LIMIT if a derived table is in FROM: SELECT * FROM
(SELECT * FROM a) LIMIT 2
- MaxDB does not support sql’s CAST and can only usefullly cast two types.
There isn’t much implicit type conversion, so be precise when creating
PassiveDefaults in DDL generation: ‘3’ and 3 aren’t the same.
sapdb.dbapi
- As of 2007-10-22 the Python 2.4 and 2.5 compatible versions of the DB-API
are no longer available. A forum posting at SAP states that the Python
driver will be available again “in the future”. The last release from MySQL
AB works if you can find it.
- sequence.NEXTVAL skips every other value!
- No rowcount for executemany()
- If an INSERT into a table with a DEFAULT SERIAL column inserts the results
of a function INSERT INTO t VALUES (LENGTH(‘foo’)), the cursor won’t have
the serial id. It needs to be manually yanked from tablename.CURRVAL.
- Super-duper picky about where bind params can be placed. Not smart about
converting Python types for some functions, such as MOD(5, ?).
- LONG (text, binary) values in result sets are read-once. The dialect uses a
caching RowProxy when these types are present.
- Connection objects seem like they want to be either `close()`d or garbage
collected, but not both. There’s a warning issued but it seems harmless.