Peter Eisentraut <peter_e@gmx.net> writes:
> The following happens in latest CVS and a fresh database:
> create table test (a int);
> insert into test values (1);
> alter table test add column b text check (b <> '');
This bug's been there awhile I fear. The failure occurs when
AlterTableAddColumn needs to add a check constraint AND the
new column causes AlterTableCreateToastTable to do its thing.
The reason there is a bug is that AlterTableCreateToastTable
gratuitously does a heap_mark4update, thereby selecting the un-updated
version of the pg_class tuple as its basis for modification (and
ignoring the HeapTupleSelfUpdated return code that warned that there
was a problem).
I've said before that I do not like heap_mark4update in catalog
manipulations, and here's a perfect example of why it's a bad idea.
regards, tom lane