On Tue, Nov 17, 2009 at 08:45:34AM -0000, Vladimir Kolesnikov wrote:
> I think in case of FKs the whole chain should be discarded (we are just
> caching after all). But I'm the whole situation happens because DDL is
> not durable neither on server nor on engine side, so formally this looks
> more like a feature request. I can add a condition to bypass this test
> line for PBXT (we have already one anyway for DELETE IGNORE), if we
> don't want to waste time now for it.
DDL that is crash safe can be done in Drizzle pretty easily. If engine
data dictionary is crash safe, and can accommodate a blob for the
table proto (or stores enough information to recreate it), then it's
pretty trivial to do.
(and you'd be beating innodb to the punch :)
--
Stewart Smith
On Tue, Nov 17, 2009 at 08:45:34AM -0000, Vladimir Kolesnikov wrote:
> I think in case of FKs the whole chain should be discarded (we are just
> caching after all). But I'm the whole situation happens because DDL is
> not durable neither on server nor on engine side, so formally this looks
> more like a feature request. I can add a condition to bypass this test
> line for PBXT (we have already one anyway for DELETE IGNORE), if we
> don't want to waste time now for it.
DDL that is crash safe can be done in Drizzle pretty easily. If engine
data dictionary is crash safe, and can accommodate a blob for the
table proto (or stores enough information to recreate it), then it's
pretty trivial to do.
(and you'd be beating innodb to the punch :)
--
Stewart Smith