BUG #18327: Column naming inconsistency for boolean literals in ELSE clauses of CASE expressions.
The following bug has been logged on the website:
Bug reference: 18327
Logged by: Michael Bondarenko
Email address: work.michael.2956@gmail.com
PostgreSQL version: 14.10
Operating system: MacOS Sonoma 14.1.1 (23B81)
Description:
I'm running my queries on the TPC-H schema, which can be found at:
https://github.com/dimitri/tpch-citus/blob/master/schema/tpch-schema.sql
I found the following inconsistency:
```
tpch=# select case when true then true else true end;
case
------
t
(1 row)
tpch=# select case when true then true else (select true) end;
bool
------
t
(1 row)
```
As you can see, the auto-generated aliases do not correspond. The expected
behaviour was for the first query to also name the column 'bool'.
I base my expectations on:
1) This query naming the result 'bool':
```
tpch=# select true;
bool
------
t
(1 row)
```
2) These queries propagating the column name from the else clause of the
case expression:
2.1) Propagation of the aggregate function name
```
tpch=# select case when true then true else bool_or(true) end;
bool_or
---------
t
(1 row)
```
2.2) Propagation of the [compound] column identifier:
```
tpch=# select case when true then 1 else t1.n_nationkey end from nation as
t1 limit 1;
n_nationkey
-------------
1
(1 row)
```
2.3) Propagation of the simple column identifier:
```
tpch=# select case when true then 1 else n_nationkey end from nation limit
1;
n_nationkey
-------------
1
(1 row)
```
2.4) And even nested case expressions:
```
tpch=# select case when true then 1 else (
case when true then 1 else t1.n_nationkey end
) end from nation as t1 limit 1;
n_nationkey
-------------
1
(1 row)
```
On Sun, Feb 4, 2024, 11:19 PG Bug reporting form <noreply@postgresql.org>
wrote:
The following bug has been logged on the website:
Bug reference: 18327
Logged by: Michael Bondarenko
Email address: work.michael.2956@gmail.com
PostgreSQL version: 14.10
Operating system: MacOS Sonoma 14.1.1 (23B81)
Description:I'm running my queries on the TPC-H schema, which can be found at:
https://github.com/dimitri/tpch-citus/blob/master/schema/tpch-schema.sqlI found the following inconsistency:
This one doesn't have a name to use. Using case is better than picking one
of the branches to make up a name from.
```
tpch=# select case when true then true else true end;
case
------
t
(1 row)
All of these do. And in any case we never publish how we choose our
aliases in this circumstance so it cannot be a bug.
The subquery forces a name to be chosen for the output relation. Then
since only one branch has a name to provide it is used.
tpch=# select case when true then true else (select true) end;
bool
------
t
(1 row)
```As you can see, the auto-generated aliases do not correspond. The expected
behaviour was for the first query to also name the column 'bool'.I base my expectations on:
1) This query naming the result 'bool':
```
tpch=# select true;
bool
------
t
(1 row)
```2) These queries propagating the column name from the else clause of the
case expression:
2.1) Propagation of the aggregate function name
```
tpch=# select case when true then true else bool_or(true) end;
bool_or
---------
t
(1 row)
```
This is indeed all consistent in that literals don't have names while other
expression tend to.
David J.
"David G. Johnston" <david.g.johnston@gmail.com> writes:
This is indeed all consistent in that literals don't have names while other
expression tend to.
Actually, pre-v15 the literal constants "true" and "false" do have
names for this purpose. For historical reasons they were parsed
into the equivalent of 't'::bool and 'f'::bool, and then the name
selection rule for a typecast took effect.
Starting in v15 they're treated like other sorts of literals,
meaning they have no assigned name:
regression=# select true;
?column?
----------
t
(1 row)
Many of these other examples change as a consequence.
regards, tom lane