"David G. Johnston" <david.g.johnston@gmail.com> writes:
> Over in news [1] Josh Drake and Eric Ridge discovered the undocumented
> feature "IS [NOT] OF"; introduced seemingly as an "oh-by-the-way" in 2002
> via commit eb121ba2cfe [2].
> Is there any reason not to document this back to 9.5,
As far as I can tell from reading the SQL spec, this has nothing much in
common with the SQL feature of that name except for the syntax. The SQL
feature seems to be a *run time* test on subtype inclusion, not something
that can be answered in parse analysis. Even if I'm getting that wrong,
it's clear that the spec intends IS OF to return true for subtype
relationships, not only exact type equality which is the only thing
transformAExprOf considers.
So my vote would be to rip it out, not document it. Somebody can try
again in future, perhaps. But if we document it we're just locking
ourselves into a SQL incompatibility.
regards, tom lane