From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | Andres Freund <andres(at)2ndquadrant(dot)com> |
Cc: | pgsql-hackers(at)postgresql(dot)org |
Subject: | Re: Why is it not sane to pass ExecStoreTuple(shouldFree=true) for tuples point into buffers |
Date: | 2014-04-07 19:58:31 |
Message-ID: | 27448.1396900711@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Andres Freund <andres(at)2ndquadrant(dot)com> writes:
> I've the need to persist a the result of an index_getnext() in a tuple
> slot. I don't want to unneccessarily duplicate the tuple data itself, so
> I'd like to use ExecStoreTuple(buffer = real_buffer) notion. But since
> the next index_getnext()/index_endscan() will overwrite/release the
> heaptuple I need to copy the HeapTupleData().
> It'd be rather useful to be able to do ExecStoreTuple(tuple, slot,
> some_buffer, true), i.e. hav ethe HeapTupleData struct freed *and* the
> buffer pinned. There's an Assert preventing that though.
There's an assumption that if you are asking to pin a buffer, the tuple
pointer you're passing is pointing into that buffer (and is therefore not
something that could be pfree'd). If it isn't pointing into a buffer,
the tuple slot is not the place to be keeping the buffer reference.
I'm disinclined to remove that Assert because failing to pin a buffer
when you *are* passing a pointer into it is a very bad, hard-to-find bug.
Admittedly the Assert is only a partial defense against that problem,
but it's better than nothing.
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Andres Freund | 2014-04-07 20:03:02 | Re: Why is it not sane to pass ExecStoreTuple(shouldFree=true) for tuples point into buffers |
Previous Message | Stephen Frost | 2014-04-07 19:35:09 | Re: "Transaction local" statistics are incorrect at speed |