Re: Delete cascade with three levels bug ?

Robert Haas <robertmhaas@gmail.com> writes:
> On Tue, Oct 27, 2009 at 10:41 AM, Euler Taveira de Oliveira
>> BTW, is it worth preventing such a crash putting an elog message in trigger.c?

> It doesn't seem right to allow a catalog change that results in an
> assertion failure.  Seems like we should either prevent the catalog
> change, or have an elog() there rather than Assert().

We can't prevent a superuser from making stupid catalog changes.  I
agree though that it's not very nice to Assert or dump core afterwards.
Changing the assert to an elog seems like an appropriate answer,
assuming that this isn't masking some more significant issue.
        regards, tom lane