Etsuro Fujita <etsuro.fujita@gmail.com> writes:
> On Sat, Aug 1, 2020 at 11:13 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Sure looks that way, doesn't it? I'm just now working to reproduce
>> on gaur's host and poke into it with a debugger. More news in an
>> hour or two (it's slow :-().
> Thanks for that!
I've concluded that this is probably a compiler bug. It doesn't fail
at optimization level -O0 or -O2, only -O1; and trying to step through
gen_partprune_steps_internal() suggests that the part_scheme local
variable is changing value, which it surely should not. Since gaur
is running an ancient gcc version, and -O1 is doubtless a pretty
under-tested optimization level, bugs there are not so surprising.
There wasn't any amazingly good reason to be using -O1 for gaur,
so I've switched the animal to use -O2. I expect it'll go back
to green in a few hours.
regards, tom lane