Re: preserving forensic information when we freeze

On Fri, Dec 20, 2013 at 7:51 AM, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:
>> I wondered that, too, but it's not well-defined for all tuples. What
>> happens if you pass in constructed tuple rather than an on-disk tuple?
>
> Those should be discernible I think, t_self/t_tableOid won't be set for
> generated tuples.

I went looking for existing precedent for code that does things like
this and found record_out, which does this:

HeapTupleHeader rec = PG_GETARG_HEAPTUPLEHEADER(0);
...
/* Extract type info from the tuple itself */
tupType = HeapTupleHeaderGetTypeId(rec);
tupTypmod = HeapTupleHeaderGetTypMod(rec);
tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
ncolumns = tupdesc->natts;

/* Build a temporary HeapTuple control structure */
tuple.t_len = HeapTupleHeaderGetDatumLength(rec);
ItemPointerSetInvalid(&(tuple.t_self));
tuple.t_tableOid = InvalidOid;
tuple.t_data = rec;

This appears to be a typical pattern, although interestingly I noticed
that row_to_json() doesn't bother setting t_tableOid or t_self, which
I think it's supposed to do. The problem I see here is that this code
seems to imply that a function passed a record doesn't actually have
enough information to know what sort of a thing it's getting. The use
of HeapTupleHeaderGetTypeId and HeapTupleHeaderGetTypMod implies that
it's safe to assume that t_choice will contain DatumTupleFields rather
than HeapTupleFields, which doesn't seem to bode well for your
approach.

Am I missing a trick?

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company