That avoids having to care about the lifetime of the
HeapTupleHeaderData passed to ExecStore[Buffer]HeapTuple(). That
doesn't make a huge difference for a plain HeapTupleTableSlot, but for
BufferHeapTupleTableSlot it can be a significant advantage, avoiding
the need to materialize slots where it's inconvenient to provide a
HeapTupleData with appropriate lifetime to point to the on-disk tuple.
It's quite possible that we'll want to add support functions for
constructing HeapTuples using that embedded HeapTupleData, but for now
callers do so themselves.
Author: Andres Freund
Discussion: https://postgr.es/m/
20180703070645.wchpu5muyto5n647@alap3.anarazel.de
index 8da0b84dd7e9c1909d97f02384cc144464ba8fd0..1e94beafac5b84601200b40fe39383b9b17c6a78 100644 (file)
#include "access/htup.h"
#include "access/tupdesc.h"
+#include "access/htup_details.h"
#include "storage/buf.h"
/*----------
HeapTuple tuple; /* physical tuple */
#define FIELDNO_HEAPTUPLETABLESLOT_OFF 2
uint32 off; /* saved state for slot_deform_heap_tuple */
+ HeapTupleData tupdata; /* optional workspace for storing tuple */
} HeapTupleTableSlot;
/* heap tuple residing in a buffer */