On Wed, Jun 05, 2024 at 01:12:41PM +0900, Kyotaro Horiguchi wrote: > At Mon, 27 May 2024 11:31:24 -0300, Ranier Vilela <ranier...@gmail.com> wrote > in >> The function *plpgsql_inline_handler* can use uninitialized >> variable retval, if PG_TRY fails. >> Fix like function*plpgsql_call_handler* wich declare retval as >> volatile and initialize to (Datum 0).
You forgot to read elog.h, explaining under which circumstances variables related to TRY/CATCH block should be marked as volatile. There is a big "Note:" paragraph. It is not the first time that this is mentioned on this list: but sending a report without looking at the reason why a change is justified makes everybody waste time. That's not productive. > If PG_TRY fails, retval is not actually accessed, so no real issue > exists. Commit 7292fd8f1c changed plpgsql_call_handler() to the > current form, but as stated in its commit message, it did not fix a > real issue and was solely to silence compiler. This complain was from lapwing, that uses a version of gcc which produces a lot of noise with incorrect issues. It is one of the only 32b buildfarm members, so it still has a lot of value. > I believe we do not need to modify plpgsql_inline_handler() unless > compiler actually issues a false warning for it. If we were to do something, that would be to remove the volatile from plpgsql_call_handler() at the end once we don't have in the buildfarm compilers that complain about it, because there is no reason to use a volatile in this case. :) -- Michael
signature.asc
Description: PGP signature