On Dec 1, 2008, at 7:19 PM, Brian Aker wrote:
Hi!
On Dec 1, 2008, at 4:38 PM, Roland Bouman wrote:
So, my proposal would be to allow engines to advertise any internal
row-identifiers (and probably, the methods to look up a row using
such
an identifier) that they may maintain to the API (should this be a
suitable key for replication). Of course, you can still demand that a
logical primary key be present in case an engine does not support
such
a feature.
Ok... we have what is called position() in the handler API but its
schematics are weak at best. Being able to say "give me a key" for
the row would work better for this anyway. The one thing that is a
bit odd though will be that I need to give a user level sort of
access to this. Right now there is nothing that we have that
represents a sort of ROWID. ROWID in MySQL was just primary key :)
Is that such a bad thing? I mean, there are cases where a PK might not
work well - a logging table for instance. In those cases, however,
having an auto_increment, while a bit silly, does work. It's not ideal
perhaps, but if the engine is storing this information anyway, I say
why not expose that to the user. There might be times when I want to
do something silly with PKs on tables that don't really need them
(again the logging table). Better to have that available to me than
have it hidden in API magic. That said, I say go with the simplest
path to accomplishing this goal properly. Isn't that one of the big
things Drizle is setting out to do, after all?
Not that I'm qualified to have an opinion on the matter or anything,
but there's my $0.02 anyway ;)
Tim S.
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help : https://help.launchpad.net/ListHelp
