Re: Assertion failure with a subtransaction and cursor

Heikki Linnakangas wrote:
> Tom Lane wrote:
>> So as far as I can tell at the moment, temp files really are the only
>> problem, and making them be managed by resource owners instead of a
>> subxact-based release policy should fix that.
>
> Ok, good.
>
>> I can go work on that, unless you wanted to?
>
> I started hacking on that when I posted, so I can finish it.

So, here's what I got this far, which is pretty straightforward.
Temporary files not opened in the interXact mode are registered with
CurrentResourceOwner, ensuring they're cleaned up at at the end of
(sub)transaction, or when the portal is closed if it's a temporary file
related to a cursor.

The logical next step would be to get rid of the interXact concept
altogether and always associate temporary files with the current
resource owner; the caller should switch to a sufficiently long-lived
one before calling. But I'm hesitant to change the signature of
OpenTemporaryFile, and tuplestore_begin_heap doesn't need the interXact
argument anymore either. Changing the signature of tuplestore_begin_heap
would certainly break user-defined set-returning C functions. So at
least in back-branches, perhaps we should leave those as dummy arguments
and elog(ERROR) if interXact is not passed in as false.

Google turned up one extra module that calls tuplestore_begin_heap with
interXact set to 'true', but frankly that one looks broken to me. It
should be using 'false':
http://cvs.pgfoundry.org/cgi-bin/cvsweb.cgi/python/be/src/call/pl.c?rev=1.20&content-type=text/x-cvsweb-markup

I left the cleanup of files/dirs opened with AllocateFile/Dir alone,
although it looks a bit inconsistent now that they don't use resource
owners as well.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com