Martin Butter <martin(dot)butter(at)splendiddata(dot)com> writes:
> Extension test_spi_in_preload_session just executes 'select 1' in its
> _PG_init() function.
> Until Postgres version 14beta1 this works fine, but in version 14beta2
> the process dies.
I don't believe we have ever considered that (i.e., execution of general
SQL from a _PG_init function) to be a supported case. If it seemed to
work before, that was completely accidental. There is no transaction
context in place when process_session_preload_libraries runs.
regards, tom lane