Tom Lane wrote:
> Looks right to me; but as a general tip, if you've made mistake X in
> place A, you might have made it in place B too. Have you checked the
> rest of dblink for forgotten SPI_finish calls?
Good tip -- will do. Though, I suspect the dblink_build_sql_* functions
have been much more lightly used than the rest of dblink, and that's the
only reason this bug has lurked for so long.
> (More generally, I wonder if AtEOXact_SPI oughtn't be fixed to emit
> a WARNING if the SPI stack isn't empty, except in the error case.
> Neglecting SPI_finish is analogous to a resource leak, and we have
> code in place to warn about other sorts of leaks.)
I'll take a look at this too.
> Sure. I believe we still intend a 7.3.5 shortly (maybe early December,
> since Thanksgiving week is a bad time to expect anything to happen).
>
OK -- then I'll plan to apply the final fix to REL7_3_STABLE as well.
Thanks,
Joe