Neil Conway <neilc@samurai.com> writes:
> Tom Lane wrote:
>> Alternatively we could make them local to any block that contains an
>> EXCEPTION clause, which would fix point 3 and also go a long way towards
>> addressing the unnecessary-overhead gripe. However that would mean that
>> an attempt to reference them from outside an exception handler would
>> probably fail outright, rather than deliver either NULLs or
>> 00000/"Successful completion".
> This behavior sounds fine to me.
I think the key distinction between this proposal and my other one
(that SQLSTATE/SQLERRM be procedure-local) is whether you want the error
status to be available to code that immediately follows the BEGIN block
containing the exception handler. That is, consider code like
BEGIN
-- do something perilous
EXCEPTION
WHEN OTHERS THEN -- nothing much
END;
IF SQLSTATE = '42000' THEN ...
At the moment I don't have a strong opinion about this. It seems
closely analogous to the question whether a loop iteration variable
should remain defined after the loop exits --- you can find cases
where that's handy, but you can also argue it shouldn't be used.
plpgsql itself is schizophrenic on the point (see integer versus
record FOR-loops), which means we don't have a solid precedent to go by.
regards, tom lane