Add DATABASE_PRE_SYNCDB to settings

[Johan Bergström <bugs@…>]

DATABASE_PRE_SYNCDB is a way of letting users execute SQL before tables are created (for instance through syncdb).
The main reason for this is users which don't have control over their db - in practise, no rights to set default modes (which has been brought up ​earlier). This would allow users to set the "correct" charset and storage engine for his/her tables.
Another useful feature is that django now has the possibility to create a database.
Here's a "real world" example:

DATABASE_PRE_SYNCDB = (
    "CREATE DATABASE IF NOT EXISTS %s" % DATABASE_NAME,
    "SET storage_engine=INNODB",
)

Alternative approaches to this has been suggested, such as:

• Alter tables after the table has been created (refs: wiki:AlterModelOnSyncDB)
• Export SQL, modify and import manually

I personally think this implementation is cleaner and allows greater flexibility.

[Johan Bergström <bugs@…>]

database_pre_syncdb patch

[George Vilches <gav@…>]

If we're going allow modifications before all the sync activities, why not just create a pre_syncdb signal. That way it won't be yet another setting to clutter the 99% case, but people could easily add the dispatcher to a management.py in their own app.

Or, with the new management refactor, maybe a user should just add their own command entirely, and have their command call the syncdb command?

[James Bennett]

Replying to George Vilches <gav@thataddress.com>:

If we're going allow modifications before all the sync activities, why not just create a pre_syncdb signal. That way it won't be yet another setting to clutter the 99% case, but people could easily add the dispatcher to a management.py in their own app.

Or, with the new management refactor, maybe a user should just add their own command entirely, and have their command call the syncdb command?

I like both of these ideas. Feel like writing up an implementation of one of them?

[George Vilches]

The patch above implements the pre_dbsync signal idea against r6799.

[Adam Nelson]

Patch needs update to latest svn revision.

[Silver_Ghost]

Updated patch. Tests included.

[Silver_Ghost]

I put tests in regressiontests/syncdb according to ​advice from Russell.

[Silver_Ghost]

Test for check that pre_syncdb is emitted at the right time added.

[Silver_Ghost]

Updated patch with tests.

[Silver_Ghost]

Added send of pre_syncdb signal in flush command and documentation. Documentation probably is not good enough but I can't write better because of my poor English.

[Chris Beaven]

It seems like this ticket has been hijacked for a different purpose. I'm wontfixing the idea of a DATABASE_PRE_SYNCDB setting, but not a pre_syncdb signal.

We have ticket #11398 to track that, I'll move Silver_Ghost's patch there.