When you try to use the NEW variable in a statement-level trigger, you get
ERROR: record "new" is not assigned yet
DETAIL: The tuple structure of a not-yet-assigned record is indeterminate.
which is not too friendly, given that people sometimes forget to specify
FOR EACH <sth> at all, get the default behaviour of FOR EACH STATEMENT
and scratch their heads. A quick search on the error detail reveals a
few such confused users already.
What's more, the documentation for PL/pgSQL triggers says that "This
variable is NULL in statement-level triggers" but it's not really a
NULL, for instance you can't execute "x is NULL" with it (you get the
I'm not that familiar with PL/pgSQL code, so I'm not sure how or if this
should be fixed. From a quick look it seems that the NEW and OLD
variables could be defined as row variables instead of record, since the
format of the tuple is known.
Would that make sense?
PS: If changing them to row types would induce too much code churn that
maybe we could use some trick to check if the error comes from using the
OLD or NEW variable and add a HINT to the error message?