On Oct 16, 2007, at 2:08 PM, Geoffrey Garen wrote:
After all active transactions are cleared, there is no context that
remembers that the database is corrupt, and the next statement to
be run would actually attempt to be executed.
I suppose user agents can volunteer to remember this and
automatically fail the next statement, but it's certainly not
specified.
Are you proposing that, once a database has been corrupted, all
transactions executed on it should fail, raising an
INVALID_STATE_ERR exception, for all time?
No, I was proposing that once a database has been corrupted, all
transactions executed on it should fail, raising an INVALID_STATE_ERR
exception, until the corruption is resolved. But my intentions have
changed (read below)
Once all active transactions are cleared, there's no need to
remember that the database was corrupt. The user agent should simply
recover from the corruption in an implementation-defined way --
either by deleting the database, performing an error-recovering
integrity check, asking the user to install cosmic ray shielding
around the house, or something else.
This is great and all, and I suspect its what most user agents would
do. "Nuke the database from orbit," and all. My concern is about any
already-open-database handles.
INVALID_STATE_ERR is already overloaded and I think the corruption
case is a particularly problematic one - I guess what I'm after is a
special condition for the corruption case built in to the spec.
Say corruption is detected and we present this error to the script
somehow - an exception for example. Then we nuke the database and
recreate it from scratch. Unless the error we presented to the script
was explicit about the condition, the script might not know that we
just deleted the entire database and it is now empty. They might
execute some query that will succeed on the new, fresh database and
they didn't even realize we reset the database behind their back.
They might think that some cached data they have in memory is still
persistent in the database, instead of knowing they could recover a
little by writing it back out.
It would be nice to have a way to indicate to the script "There was a
catastrophic event and we reset your database, assume you're starting
over from scratch."
~Brady