Interface Changes

• The long broken and unused DBMConnection has been removed.
• Added a connection parameter to all class methods (patch 974755)
• Connection objects have a .module attribute, which points to the DB-API module. This is useful for getting access to the exception objects.

Features

• New UnicodeCol() that converts to and from Unicode in the database. See docs.

• Added indexing (from Jeremy Fitzhardinge). See the documentation for more.

• All connections are explicitly closed, not just garbage collected. Many database drivers don't close database connections properly when the connection object is garbage collected.
• New distinct option to selects, like MyClass.select(..., distinct=True)
• You can now do MyClass.selectBy(joinedTable=joinedTableInstance), where before you had to do MyClass.selectBy(joinedTableID=joinedTableInstance.id). (From Dave Cook)

Interface Changes

• Lazy updates. Add _lazyUpdate=True to your class, and updates will only be written when you call obj.syncUpdate() or obj.sync() (sync also refetches the data from the database, which syncUpdate does not do). When enabled, instances have a property dirty, which indicates if they have pending updates. Inserts are still done immediately.
• Separated database drivers (PostgresConnection, MySQLConnection, etc.) into separate packages. You can access the driver through URIs, like mysql://user:pass@host/dbname -- to set drivers after class creation you should use sqlobject.connectionForURI().
• The SQLObject package has been renamed to sqlobject. This makes it similar to several other packages, and emphasizes the distinction between the sqlobject package and the SQLObject class.
• Class instantiation now creates new rows (like .new() used to do), and the .get() method now retrieves objects that already have rows (like class instantiation used to do).
• We're now using a Subversion repository instead of CVS. It is located at http://svn.colorstudy.com/trunk/SQLObject
• If you pass forceDBName=True to the *Col constructors, then your column name doesn't have to be restricted to a-z, 0-9, and _.
• *Col constructors now support cascade: cascade=None (default) means no constraint; cascade=True means that if the foreign key is deleted, the object will be deleted; cascade=False means that the delete will fail; cascade="null" means that the column will be set to NULL. The constraints are only implemented in the DBMS, not in SQLObject (i.e., they will not work in databases like MySQL and SQLite).
• New _create(id, **kw) method that can be overridden to intercept and modify attempts to insert rows in the database.
• You can specify _idType in your class, like _idType = str. The default type is int; i.e., IDs are coerced to integers. This is a temporary interface; a more general specifier for primary keys will be added later.
• New classmethod createTableSQL() method for SQLObject classes, which returns the SQL that can be used to create the table. Analog to createTable().

Bugs

• SQLite booleans fixed.
• You can now use sqlite:/:memory: to store the database in memory.
• Some bugs resolved when caching is turned off (SF 956847)

Bugs

• Python 2.2 booleans fixed (SF: 903488)
• Longs (e.g., 1L) get converted properly (SF: 939965)

Interface Changes

• If you commit or rollback a transaction, you must call trans.begin() to restart the transaction. Any database access on the transaction inbetween commit/rollback and being will result in an AssertionError. (It's also acceptable to create a new transaction object instead of reusing the old one, but objects in that transaction will be invalid)

Bugs

• Using .select() would hold on to a connection, and also release it back to the connection pool. Very un-threadsafe and all-around bad.
• Fixed bug which did not release connections after database (query) error.
• When setting columns that use validators, the Pythonic (vs. database) representation wasn't being stored in the column. Now we roundtrip (through toPython and fromPython) the values when they get set.
• PostgreConnection is back to using sequences for ID generation, instead of oids. Long explanation -- oids can be unindexed in some versions of Postgres, or not even exist.
• When turning caching off and using transactions, got an attribute error on rollback.
• Rollback or commit didn't find objects that were expired from the cache but still in memory.
• Rollback or commit didn't free the connection object, so as you created more transactions it stole connections and didn't put them back in the pool.

Interface Changes

• Select results no longer have a __len__ method (i.e., you can't do len(Person.select(Person.q.firstName=='Bob'))). There is now a .count() method instead. __len__ gets called implicitly in several circumstances, like list(), which causes potentially expensive queries to COUNT(*).

Bugs

• Objects retrieved from a join now respect the transaction context of the original instance.
• .select().reversed() works.

Features

• Firebird support.
• Database-specific literal quoting (motivation: MySQL and Postgres use backslashes, Firebird and SQLite do not).
• Generic conversion/validation can be added to columns.
• BoolCol for portable boolean columns (BOOL on Postgres, INT on MySQL, etc.)
• Non-integer IDs. (Automatic table creation is not supported for non-integer IDs)
• Explicit IDs for new instances/rows (required for non-integer IDs).
• Instances can be synced with the database (in case there have been updates to the object since it was first fetched).
• Instances can be expired, so that they will be synced when they are next accessed.

Interface Changes

• SQLBuilder.sqlRepr renamed to SQLBuilder.sqlrepr, signature changed to sqlrepr(value, databaseName) to quote value, where databaseName is one of "mysql", "postgres", "sqlite", "firebird".
• sqlRepr magic method renamed to __sqlrepr__, and takes new databaseName argument.
• When using explicit booleans, use Col.TRUE and Col.FALSE for backward compatibility with Python 2.2. This is not required for BoolCol, however (which converts all true values to TRUE and false values to FALSE)
• SQLObject has a sqlrepr method, so you can construct queries with something like "WHERE last_name = %s" % Person.sqlrepr('Bob')

Bugs

• Released all locks with finally:, so that bugs won't cause frozen locks.
• Tons of transaction fixes. Transactions pretty much work.
• A class can have multiple foreign keys pointing to the same table (e.g., spouse = ForeignKey("Person"); supervisor = ForeignKey("Person"))

Features

• You can specify columns in a new, preferred manner:

  class SomeObject(SQLObject):
      someColumn = Col()
  

  Equivalent to:

  class SomeObject(SQLObject):
      _columns = [Col('someColumn')]
  

  Ditto joins.

• Cache objects have a clear method, which empties all objects. However, weak references to objects are maintained, so the integrity of the cache can be ensured.

• SQLObject subclasses can be further subclassed, adding or removing column definitions (as well as changing settings like connection, style, etc). Each class still refers to a single concrete table in the database -- the class hierarchy is not represented in the database.

• Each SQLObject subclass can have an associated style, as given in the _style attribute. This object is used to map between Python and database names (e.g., the column name for a Python attribute). Some samples are available in the Style module.

• Postgres support for _fromDatabase (reading a table definition from the database, and creating a class from that).

• Postgres id columns more permissive, you don't have to create a specially named sequence (or implicitly create that sequence through SERIAL). lastoid is used instead.

• MySQL uses localhost as the default host, and the empty string as the default password.

• Added functions for use with queries: ISNULL, ISNOTNULL. == and != can be used with None, and is translated into ISNULL, ISNOTNULL.

• Classes can be part of a specific registry. Since classes are referred to by name in several places, the names have to be unique. This can be problematic, so you can add a class variable _registry, the value of which should be a string. Classes references are assumed to be inside that registry, and class names need only be unique among classes in that registry.

• SomeClass.select() selects all, instead of using SomeClass.select('all'). You can also use None instead of 'all'.

• Trying to fetch non-existent objects raises SQLObjectNotFound, which is a subclass of the builtin exception LookupError. This may not be raised if _cacheValues is False and you use the ID to fetch an object (but alternateID fetches will raise the exception in either case).

• Can order by descending order, with the reversed option to the select method, or by prefixing the column with a "-".

• Ordering with joins works better -- you can order with multiple columns, as well as descending ordering.

Bugs

• DateTimeCol uses TIMESTAMP for Postgres. Note that the Python type name is used for column names, not necessarily the SQL standard name.
• Foreign key column names are slightly more permissive. They still need to end in id, but it's case insensitive.
• _defaultOrder should be the python attribute's name, not the database name.
• SomeClass.q.colName uses proper Python attributes for colName, and proper database names when executed in the database.
• SQLite select results back to being proper iterator.
• SomeClass.q.colName now does proper translation to database names, using dbName, etc., instead of being entirely algorithm-driven.
• Raise TypeError if you pass an unknown argument to the new method.
• You can override the _get_* or _set_* version of a property without overriding the other.
• Python 2.3 compatible.
• Trying to use Col('id') or id = Col() will raise an exception, instead of just acting funky.
• ForeignKey columns return None if the associated column is NULL in the database (used to just act weird).
• Instantiating an object with an id of None will give an error, instead of just acting weird.

Internal

• Col class separated into Col and SOCol (and same for all other *Col classes). Col defines a column, SOCol is that definition bound to a particular SQLObject class.
• Instance variable _SO_columns holds the SOCol instances.

Features

• Table creation (SQL schema generation) via new class method createTable. And of course a dropTable method to go with.
• Add and remove columns at runtime, optionally modifying the schema in the database (via ALTER). (Does not work in SQLite)
• New column classes (see Col module), indicates type
• Classes can be created by parsing an already existant table (MySQL only).
• Objects are not cached indefinitely. Cached objects are expired into a weak dictionary (it allows objects to be garbage collected if nowhere else in the program is using the object, but until it is collected it's still available to the cache). Some cache control, pass nocache=True to your connection object to eliminate as much caching as possible. See Cache module for a bit more.
• New DBMConnection, implements a database-like backend without any database to speak of, including queries (so long as you use SQLBuilder and don't generate your where clauses manually). Actual SQL generation is done entirely by the database connection, allowing portability across very different backends.
• Postgres table IDs should be created with type SERIAL (which implicitly creates a sequence).
• New _defaultOrder class variable gives a default for the orderBy parameter to select queries.

Bugs

• LIMIT/OFFSET (select result slicing) works in Postgres and SQLite.
• tableExists method from DBConnection works in same.
• mxDateTime not required (never should have been, always just an option).

Bugs

• Fixed caching of new objects

Features

• SQLite support
• Select statements are lazily generated, retrieve full rows for speed, and are slicable (select docs).
• alternateID option for Col objects -- select individual objects via UNIQUE columns, e.g., a username (Col docs).