currently we have limitations: no views, only works with
innodb/aria/myisam/xtradb etc
I was looking through some of the other storage engines to get a better
idea of how things work, with the goal of fixing the memory problem. I
think I am close, we need to just use the proper memory share
arrangement and possibly a hash of tables we already opened, or
something. Anwyay.
One thing I (re)realised, looking at ha_connect, is that conceptually
oqgraph could work with anything at all that has table semantics, given
the right interface.
So in theory, we could restructure oqgraph to work like ha_connect and I
think federatedx, which resolve to using the mysql client API to talk to
the database.
If I was able to modify oqgraph to access the backing store like that
instead of directly hammering the raw mysql internals, it means we could
work with anything. Which means you could even chain it to ha_connect
and access data in other databases like postgres or god forbid SQL Server!
The big open question, is whether there would be any significant
performance hit using the mysql client library to access data on the
same host?
just a thought bubble