Richard Guo <guofenglinux@gmail.com> writes: > On Tue, Aug 2, 2022 at 4:50 PM Martijn van Oosterhout <kleptog@gmail.com> > wrote: >> Now it's morning I've thought of a way to reproduce it more easily, see >> the attached script.
> Thanks for the report! I can reproduce it on HEAD.
FWIW, this reproduces the bug for me in v13 and v14, but not v15 or HEAD. While the method to fix the bug seems clear enough, it doesn't seem like we have a test case that's stable enough to be worth anything as a regression test. Hmmm...
Looking at what Richard has uncovered, it appears the issue is related to the table simply being big enough that it considers a parallel seqscan plan, and then fails. Something I can look into in the morning.
The part I haven't seen explained is why the generate_series() is important. My guess is that if you replace it with an expression it is no longer an SRF and it produces some completely different plan that prevents the problematic path being triggered.
