On Jul 16, 2009, at 9:37 AM, Michael Widenius wrote:
> For example, MySQL may first do an ORDER by optimization and use a
> bitmap that only fetches the row id + what is in the ORDER BY part and
> then later it may fetch the whole rows.
I've been wondering for a while now if there wouldn't be a better way
to handle this then swapping out the bitmap guts for different actions.
The bitmap system seems a little fragile in its current state (I had
to fix a few bugs in it recently when we found corner cases where it
was not always signaling usage correctly).
It also causes bad assert cases if an engine makes use of the Field
objects in a way that is appropriate but not really thought out. There
are a number of engines out there that have to disable them to avoid
the asserts (which I think are aggressive and good).