Hmm, thanks for the report.
I can repeat the aboved issue on master, even on pg10 and pg 11.
I analyzed this issue, and I found that ATExecAddColumn(), we forgot to call CommandCounterIncrement() in if (colDef->inhcount > 0) {...} branch.
So the third(a->d) updates the first(a->b->c->d) tuple.
Attached patch is my quickly fixed solution.
--
The following bug has been logged on the website:
Bug reference: 18297
Logged by: Alexander Lakhin
Email address: exclusion@gmail.com
PostgreSQL version: 16.1
Operating system: Ubuntu 22.04
Description:
The following query:
CREATE TABLE a ();
CREATE TABLE b () INHERITS (a);
CREATE TABLE c () INHERITS (b);
CREATE TABLE d () INHERITS (a,b,c);
ALTER TABLE a ADD COLUMN i int;
fails with:
NOTICE: merging definition of column "i" for child "d"
ERROR: tuple already updated by self
While with a simpler hierarchy:
CREATE TABLE a ();
CREATE TABLE b () INHERITS (a);
CREATE TABLE c () INHERITS (a,b);
the column added successfully:
ALTER TABLE a ADD COLUMN i int;
NOTICE: merging definition of column "i" for child "c"
ALTER TABLE
In the failed case the error occurred when table d was processed the third
time. First (following chain a -> b -> c-> d) the table got the column i
added, second (a -> b -> d) it got merged column definition, third
(a -> d) an attempt to merge the column definition once more failed.
This error can be seen at least on REL_12_STABLE .. master.
