Pavel Stehule <pavel.stehule@gmail.com> writes:
> There is only one issue, that should be solved first. I introduced non
> standard diagnostics field "column_names", because there is not
> possible get "column_name" value for check constraints now. A correct
> implementation of COLUMN_NAME field needs a explicit relation between
> pg_constraint and pg_attribute - maybe implemented as new column to
> pg_constraint. Do you agree?
No, I don't. You're adding complication to solve a problem that doesn't
need to be solved. The standard says to return the name of the
constraint for a constraint-violation failure. It does not say anything
about naming the associated column(s). COLUMN_NAME is only supposed to
be defined for certain kinds of errors, and this isn't one of them.
regards, tom lane