> Good point. We'd have to recognize btree splits (and possibly some
> other operations) as things that must be done anyway, even if their
> originating transaction is aborted.
>
> There already is a mechanism for doing that: xlog entries can
> be written without any transaction identifier (see XLOG_NO_TRAN).
> Seems to me that btree split XLOG records should be getting written
> that way now --- Vadim, don't you agree?
We would have to write two records per split instead of one as now.
Another way is new xlog AM method: we have XXX_redo, XXX_undo (unfunctional)
and XXX_desc (for debug output) now - add XXX_compact (or whatever)
able to modify record somehow for BAR. For heap, etc this method could
be {return} (or NULL) and for btree it could remove inserted tuple
from record (for aborted TX).
Vadim