On Thu, Jan 19, 2017 at 6:09 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
I wrote: > Hmm ... that line was last touched by ab1f0c822, so I'm betting that > I broke it somehow, but I'm not sure how. > It looks like S_3 might have been parsed from a totally empty source > string? But if that's the trigger, I'd think it'd be entirely trivial > to reproduce.
Oh, duh: the reason it's not trivial to reproduce is you have to try to bind an empty prepared statement *in an already-aborted transaction*.