On Thu, 27 Mar 2025 at 18:33, Álvaro Herrera <
alvherre@alvh.no-ip.org> wrote:
>
> On 2025-Mar-27, Andrew Dunstan wrote:
>
> > I don't think we can backpatch this. It's a behaviour change.
>
> I agree, we can't.
>
> Also, if we're going to enforce this rule, then pg_upgrade --check needs
> to alert users that they have a database name that's no longer valid.
> That needs to be part of this patch as well. We should also ensure that
> these are the only problem characters, which IMO means it should add a
> test for pg_dumpall that creates a database whose name has all possible
> characters and ensures that it is dumpable.
>
> --
> Álvaro Herrera Breisgau, Deutschland —
https://www.EnterpriseDB.com/> "La virtud es el justo medio entre dos defectos" (Aristóteles)
Thanks Álvaro.
Yes, I also agree with you. "pg_upgrade --check" should alert users regarding database names because pg_upgrade is failing if we have \n\r in dbname(old cluster).
pg_upgrade --check: This is passing but pg_upgrade is failing.
............
Checking for new cluster tablespace directories ok
*Clusters are compatible*
pg_upgrade:
.........................
Creating dump of global objects ok
Creating dump of database schemas
shell command argument contains a newline or carriage return: "dbname='invalid
db'"
As a part of this patch, we can teach pg_upgrade to alert users regarding database names with invalid characters or we can keep old behavior as upgrade is already failing without this patch also.
I will try to make a separate patch to teach "pg_upgrade --check" to alert users regarding database/user/role with \n\r in the old cluster.
Here, I am attaching an updated patch for review.
This patch has changes for: CREATE DATABASE, CREATE ROLE, CREATE USER and RENAME DATABASE/USER/ROLE and have some tests also. (EXCEPT RENAME test case)
