Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> Tracing it down, turns out that transformPartitionBoundValue gets from
> coerce_to_target_type a CoerceToDomain node. It then tries to apply
> expression_planner() to simplify the expression, but that one doesn't
> want anything to do with a domain coercion (for apparently good reasons,
> given other comments in that file).
Quite. Suppose you did
create domain overint as int;
create table pt (a overint) partition by range (a);
create table pt1 partition of pt for values from (0) to (100);
and the system took it, and then you did
alter domain overint add check (value > 100);
What happens now?
> However, if we take out the
> expression_planner() and replace it with a call to
> strip_implicit_coercions(), not only it magically starts working, but
> also the regression tests start failing with the attached diff, which
> seems a Good Thing to me.
Why would you find that to be a good thing? The prohibition against
mutable coercions seems like something we need here, for more or less
the same reason in the domain example.
regards, tom lane