Simon Riggs <simon.riggs@enterprisedb.com> writes:
> New version of patch, now just a one-line patch!
Of course, there's all the documentation and comments that you falsified.
Also, what of the SubTransSetParent call in ProcessTwoPhaseBuffer?
(The one in ProcArrayApplyXidAssignment is actually okay, though the
comment making excuses for it no longer is.)
Also, if we're going to go over to a one-level structure in pg_subtrans,
we really ought to simplify the code in subtrans.c accordingly, and
get rid of the extra lookup currently done for the top parent's parent.
I still wonder whether we'll regret losing information about the
subtransaction tree structure, as discussed in the other thread [1].
That seems like the main barrier to proceeding with this.
regards, tom lane
[1] https://www.postgresql.org/message-id/flat/CANbhV-HYfP0ebZRERkpt84ZCDsNX-UYJGYsjfS88jtbYzY%2BKcQ%40mail.gmail.com