Thread

Re: Instability of phycodorus in pg_upgrade tests with JIT

Alexander Lakhin <exclusion@gmail.com> — 2025-10-25T16:00:00Z

Hello Tom and Andres,

25.10.2025 00:31, Tom Lane wrote:
> Sure, we'd need to change our docs about the oldest supported LLVM
> version if we go that way.
>
>>> I wonder if we should just write these off as "probably an LLVM bug".

As I wrote upthread, I could not reproduce the issue with the same old
LLVM versions.

>> I'm not sure that's really convincing, given that REL_16_STABLE seems to not
>> have an issue?
> The other side of that coin is that no other LLVM-using animal is
> showing similar instability.  Sure, it's plausible that we changed
> something in v15 or so that stopped the problem, but is it worth the
> effort to try to find out what?  And if we did find it, would we
> care to risk back-porting it?

My collection [2] contains also reports from other animals: petalura,
desmoxytes, dragonet.


> (If you want to research this, I'm not standing in the way.
> But I think there are better uses for your time.)

I wanted to research this, but failed, to my disappointment.


[1] https://www.postgresql.org/message-id/563ee5af-8ee2-484f-b50a-1c8fbdd16171%40gmail.com
[2] 
https://wiki.postgresql.org/wiki/Known_Buildfarm_Test_Failures#check-pg_upgrade_fails_on_LLVM-enabled_animals_due_to_double_free_or_corruption

Best regards,
Alexander