So I was testing the next step of plpgsql modification, namely actually
letting the parser hooks do something, and it promptly blew up in
trigger functions, like so:
+ ERROR: OLD used in query that is not in a rule
+ LINE 1: SELECT OLD
+ QUERY: SELECT OLD
+ CONTEXT: SQL statement in PL/PgSQL function "trigger_data" near line 35
The reason is that because plpgsql is no longer translating references
to its names into Params before letting the core parser see them, the
kluge in gram.y that changes "OLD" to "*OLD*" and "NEW" to "*NEW*"
kicks in, or actually decides to throw an error instead of kicking in.
I am wondering what is the point at all of having that kluge. It
certainly doesn't manage to make OLD/NEW not act like reserved words,
in fact rather more the opposite, as shown here. If we just made those
names be ordinary table alias names in rule queries, wouldn't things
work as well or better?
BTW, this brings up another point, which is that up to now it's often
been possible to use plpgsql variable names that conflict with
core-parser reserved words, so long as you didn't need to use the
reserved word with its special meaning. That will stop working when
this patch goes in. Doesn't bother me any, but if anyone thinks it's
a serious problem, speak now.
regards, tom lane