On Mon, Feb 12, 2018 at 9:11 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Andrew Gierth <andrew@tao11.riddles.org.uk> writes: > So what's happening here is that the function get_bug_id, being stable, > is being called speculatively at plan time for the query where it > appears in the WHERE clause. For whatever reason, the snapshot it's > being run in at that time is not the same one actually used for the > later execution of the query, and the plan-time snapshot doesn't see the > just-inserted row.
> It looks like what's going on here is that SPI does GetCachedPlan - > which is where planning will happen - _before_ establishing the new > snapshot in the non-read-only case (read_only is false here because the > calling function, test_bug(), is volatile).
Yeah, I came to the same conclusion. I think it's basically accidental that the given test case works before 9.2: the reason seems to be that in 9.1, the plancache doesn't pass through the parameter list containing the value of "my_text", so that the planner is unable to speculatively execute get_bug_id(). The order of operations in _SPI_execute_plan is just as wrong though.
I'm not sure I understand. When's the snapshot used for planning actually taken here?
