Transaction-handling issues
Ed Pasma <[email protected]>
| Newsgroups | gmane.comp.python.db.pysqlite.user |
|---|---|
| Message-ID | <[email protected]> |
Dear all, One thing of the PySQLite interface that I find confusing is the transaction handling. It is convenient to have automatic transaction- begin but the way it combines with explicit transactions is confusing. Following are my issues, and a change proposal. When automatic transactions are enabled, as per the default, PySQLite automatically issues a begin statement before it first executes an update or other dml-statement. In this mode, begin should never be added programmaticaly as an SQL-statement. Doing so will raise an error as soon as PySQLite does the automatic begin. The error mesage in this case is: "SQL Logic error or missing database". The type of begin statement that is automaticly inserted can be any of the supported SQLite types (DEFERRED, IMMEDIATE, EXCLUSIVE). By default it is "" which is equivalent to DEFERRED. What confuses here is IMMEDIATE. As it is only inserted at the first dml, it actually becomes equivalent to DEFERRED. Commit should, with the default operation mode, only be programmed as a call to the corresponding method. Commit as an SQL-statement should not be used. Doing so is punished by an error: "cannot commit - no transaction is active". While the changes really have been committed. With automatic transaction-begin disabled (isolation_level=None), the other way around only commit as an SQL-statement may be used. Using the commit method in this mode, does not cause an error. But neither does it really commit changes. Now I like to know if the following proposal makes sense. To make PySQLite handle commit as an SQL-statement equally to the method. The only distinction would be that the method nay be called also if there are no changes to commit. Secondly, but this may be more dificult, to allow a transaction to be explicitly started using BEGIN also when in the default operating mode. The interface should then be aware of the new transaction state. I assume that this proposal is not against the dbapi v.2 specification and that it would be "downwards compatible". Best regards, Edzard Pasma