Hi,
On 2020-10-14 17:56:16 -0700, Andres Freund wrote:
> Oh dear. It's not as simple as that. The issue indeed are relocations,
> but we don't hit those errors. The issue rather is that the systemz
> specific relative redirection code thought that the only relative
> symbols are functions. So it creates a stub function to redirect
> them. Which turns out to not work well with variables like
> CurrentMemoryContext...
That might be a problem - but the main problem causing the crash at hand
is likely something else. The prototypes we create for
ExecAggTransReparent() were missing the 'zeroext' parameter for a the
'isnull' attribute, because the code for copying the attributes from
llvmjit_types.bc didn't go deep enough (i.e. I didn't quite grok the
pretty weird API). On s390x that lead to the newValue argument in
ExecAggTransReparent() having a 0 lower byte, but set higher bytes -
which then *sometimes* fooled the if (!newValueIsNull) check, which
assumed that the higher bits were unset.
I have a fix for this, but I've just stared at s390 assembly code for
~10h, never having done so before. So that'll have to wait for tomorrow.
It's quite possible that that fix would also help on other
architectures...
Greetings,
Andres Freund